NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #55 most downloaded on PyPI
multidict implementation
Latest stable
7.2.1
Last release 29 days ago
24 Oct 2026
Ships unpredictably
gaps range from 8 days to 8 months
Nearly every release is documented
notes for 59 of the last 60 stable releases
8 versions withdrawn
withdrawn after publishing
11 years old
157 releases · first in 2016
Drag across the ruler to compare, or tap a release to open it.
Fixed a regression in 7.2.0 where a subclass of MultiDictProxy or CIMultiDictProxy that converted its argument in its own __init__() raised TypeError
Fixed a regression in 7.2.0 where a subclass of
MultiDictProxy or CIMultiDictProxy
that converted its argument in its own __init__() raised
TypeError before that __init__() ran; such a subclass got an
empty multidict until its __init__() set the proxied one
-- by @asvetlov.
(2026-10-24)
Fixed a regression in 7.2.0 where a subclass of ~multidict.MultiDictProxy or ~multidict.CIMultiDictProxy that converted its argument in its own __init__() raised TypeError before that __init__() ran; such a subclass got an empty multidict until its __init__() set the proxied one -- by asvetlov.
Related issues and pull requests on GitHub: 1824, 1825.
(2026-11-09)
Modernized the GitHub issue and pull request templates to guide contributors through bug reports and feature requests more consistently – by @Polandia94 .
Related issues and pull requests on GitHub: #1519 .
Synchronized the .yamllint configuration with the tox-dev template by adding the quoted-strings rule and removing redundant quotes in GitHub Actions workflows.
Related issues and pull requests on GitHub: #1830 .
(2026-10-24)
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Removals and backward incompatible breaking changes
7.2.0 drops Python 3.10, fixes a crash in a proxy built without
__init__() and a MemoryError from a huge __length_hint__(),
and is otherwise a performance release for the C extension: lookups,
adds, updates and deletes each needed a few percent fewer instructions.
Stopped a huge __length_hint__() from raising MemoryError in
extend(), update()
and merge(); the C extension capped the room
it reserved for the hint at the largest table the constructor allocates
up front -- by @asvetlov.
Related issues and pull requests on GitHub:
#1767.
Fixed a segmentation fault when using a MultiDictProxy
or CIMultiDictProxy built by __new__() without
__init__(); __new__() started requiring the proxied multidict, the
same argument as the constructor -- by @asvetlov.
Related issues and pull requests on GitHub:
#1782.
Sped up reading a multidict in the C extension. md[key],
get() and getone()
became up to 7% faster by returning the lookup's result by value instead
of through a pointer to the caller's local, which had made
-fstack-protector-strong add a stack canary to every call. On the GIL
build a get() miss became 5% faster and
key in md inlined its hash table lookup. On the free-threaded build
getall() stopped taking the multidict's lock
for a multidict without repeated keys, the way
get() already avoided it. len(md.keys())
and iterating over the keys of a multidict that holds a key more than
once became 4% to 13% faster -- by @asvetlov.
Related issues and pull requests on GitHub:
#1775, #1776, #1781, #1800, #1817.
Sped up building and updating a multidict in the C extension. On the GIL
build, constructing a multidict from a sequence of pairs needed 8% to 10%
fewer instructions and update() 3.5% to 4.5%
fewer. On the free-threaded build, building a
MultiDict became up to 14% faster, since a table that
is both full and unable to hold the new key in its compact form was grown
and moved in one rebuild, and add(),
md[key] = value for a new key and
setdefault() needed 2% to 3% fewer
instructions, about 1% fewer on the GIL build. Updating a multidict with
existing keys needed 1.5% to 2.9% fewer instructions, while adding a new
key needed about 0.5% more on the GIL build -- by @asvetlov.
Related issues and pull requests on GitHub:
#1763, #1774, #1780, #1785, #1802.
Sped up replacing and removing items in the C extension.
add(), getall(),
md[key] = value and del md[key] stopped computing the identity
of a key that, or whose istr canonical form, already
serves as one, as md[key] did, and checked a
CIMultiDict key holding uppercase letters only once.
On the GIL build del md[key] needed 4% to 5% fewer instructions,
md[key] = value about 2% fewer and pop()
1% fewer. Replacing a key that had duplicates needed 0.8% to 1.7% fewer
instructions, and clear() became 6% faster on
the free-threaded build -- by @asvetlov.
Related issues and pull requests on GitHub:
#1769, #1770, #1779, #1796, #1809.
Dropped support for Python 3.10 as it has reached end of life
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1777.
Refreshed the instruction-count tables in the benchmarking reference
against the current tree -- by @asvetlov.
Added the CPython 3.15 trove classifier and started testing the
GIL and free-threaded 3.15 builds in CI -- by @asvetlov.
Related issues and pull requests on GitHub:
#1766.
Split the release publishing and signing across four parallel jobs, so
that each one finishes well within the lifetime of its short-lived PyPI
upload token -- by @asvetlov.
Changed the clear() benchmarks to clear a thousand prebuilt
multidicts of 0 to 99 items each, copied outside the measured region,
since the copy inside the timed loop and the single small table it
measured made the result swing by up to 19% between runs of the same
commit -- by @asvetlov.
Related issues and pull requests on GitHub:
#1784.
Pinned Hypothesis in the MyPy pre-commit hooks and dropped a
type: ignore comment from the Hypothesis tests that became unused
once Hypothesis annotated run_state_machine_as_test(), which failed
the MyPy lint job -- by @asvetlov.
Related issues and pull requests on GitHub:
#1820.
Fixed a test of a huge __length_hint__() failing on 32-bit builds,
where its 2**40 hint did not fit a Py_ssize_t
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1821.
Split the C extension's helpers per table kind, so that the lookup
loops of pop(), popall(), item replacement and update(), the
table rebuild and the deferred delete of an entry call the helper for
their own kind instead of testing it again, and moved the choice between
overwriting and dropping each entry of a repeated key into the per-kind
loops of update(). Folded helpers with a single caller into that
caller: choosing the kind of a new table, reading a compact entry's
identity and hash, filling a new entry, replacing a key's values,
deleting a key and looking up the first value of a key that needs a
computed identity. Merged the two per-kind checks for a deleted entry
into one, and removed the separate watched copies of delete, pop and
replace, folding the add that follows a missed replace into the replace
itself -- by @asvetlov.
Related issues and pull requests on GitHub:
#1771, #1772, #1773, #1786, #1788,
#1789, #1790, #1792, #1804, #1805,
#1810, #1811.
Removed duplicated code from the C extension: the (key, default)
methods shared their argument parsing and missing-key result, and one
macro each replaced the copies of the lock-free compact entry match on
the free-threaded build and of the per-kind dispatch of a single key
lookup. Turned the watcher record, flush and pending check helpers and
_growth_rate() into macros, so that a multidict without watchers
skipped evaluating the event arguments, renamed the lowercase
function-like macros to upper case, and dropped the leading underscore
from the watcher functions that the macros call from other headers
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1793, #1794, #1795, #1806, #1807,
#1808, #1812.
Adjusted how the C extension spends GCC's inlining budget: marked most
checks that lead to a raised exception as unlikely, so the compiler
moves the error paths out of the hot code, kept the event name lookup of
a failing watcher callback report and the cleanup of a dropped watcher
event log out of line, and dropped the inline hint from the helpers
that compute a table's size and that rebuild, build or free a table's
keys -- by @asvetlov.
Related issues and pull requests on GitHub:
#1783, #1791, #1801, #1815.
Moved the C extension's repr() of multidicts, proxies and views and
its equality comparison out of the hash table header, each into a header
of its own -- by @asvetlov.
Removals and backward incompatible breaking changes
7.1.0 is a correctness and housekeeping release. It changes what
keys() yields, fixes a long list of crashes and
inconsistencies that Python code run in the middle of an operation could
trigger, stores the common all-str multidicts in half the space, and
deprecates multidict.upstr.
Behaviour changes. keys() and iteration over a
multidict yield each key once, as spelled in its first item, so
len(d.keys()) counts distinct keys; items(),
values() and len(d) still see every item. A
str subclass's own lower() is no longer called to compute the
case-insensitive identity of a CIMultiDict key.
Robustness. Finalizers, __eq__(), __repr__(), lower() and
argument iterators that read or mutate the multidict in the middle of
__init__(), update(), merge(), del d[key],
popall(), popitem(),
to_dict() or repr() no longer crash the C
extension, lose or duplicate pairs, or make the two backends disagree. Two
data races against lock-free readers on the free-threaded build are gone
too.
Memory. A MultiDict whose keys are all exact
str, and a CIMultiDict whose keys are all exact
istr, are stored in a compact table without a separate
identity and hash per entry, which cuts their size by about 45%.
Fixed building the C extension on platforms where char is unsigned,
such as ARM, PowerPC, s390x and RISC-V, and with GCC 9, which flags
CPython's own headers from 3.12 on; both failed on a sign conversion
warning since -Wconversion took effect. GCC 9 builds the extension
without -Wconversion -- by @asvetlov.
Fixed getversion() returning only the low 32 bits of
the version on Windows and on 32-bit platforms, where two different
versions could compare equal, and fixed a race in the pure-Python
implementation on GIL builds that could give two multidicts mutated at
the same time from different threads the same version
-- by @asvetlov.
Fixed in on an items view truncating the length of a non-tuple,
non-list operand to 32 bits in the C extension, which let an object
reporting 2**32 + 2 items be indexed as a pair and raise
TypeError instead of returning False -- by @asvetlov.
Related issues and pull requests on GitHub:
#1611.
Fixed adding to a full table of the C extension, which compacted it in
place whenever it held deleted entries and so regained only as many
slots as were deleted: a delete-then-add loop on a full table rebuilt
it on every add. The table was resized by the number of live entries
instead, as the pure-Python implementation did, which also let it
shrink after deletions. Fixed extend(),
update() and merge()
reserving room by comparing table sizes, which ignored the room taken by
deleted entries; the reservation checked the free room instead, in both
implementations -- by @asvetlov.
Fixed calling __init__() again on a live multidict losing and leaking
whatever a finalizer of an old value added to it in the C extension, and
miscounting len() in the pure-Python implementation, by installing
the new contents and size before releasing the old pairs
-- by @asvetlov.
Fixed two data races on free-threaded builds against lock-free readers:
copy() and __init__() from a multidict of the same kind read the
source's reader count while its readers updated it, and __init__()
on a live multidict rewrote the case-insensitivity flag its lookups
read -- by @asvetlov.
Related issues and pull requests on GitHub:
#1621.
Fixed MultiDict.__init__() called on a CIMultiDict making it
case-sensitive in the C extension, and CIMultiDict.__init__()
rejecting a MultiDict there, by taking the mode from the
instance rather than from the class whose __init__() was called, as
the pure-Python implementation already did -- by @asvetlov.
Related issues and pull requests on GitHub:
#1622.
Fixed the C extension dropping the existing items when a multidict was
re-initialized from itself, or from a proxy of itself, together with keyword
arguments, as in d.__init__(d, key=value); the pure-Python
implementation was not affected -- by @asvetlov.
Related issues and pull requests on GitHub:
#1666.
Fixed a use-after-free in popitem() when
building the result ran Python code that mutated the multidict: a
str subclass key's __str__() on
CIMultiDict, or on Python 3.10 and 3.11 a garbage
collection's finalizers, by removing the pair first, and fixed the
pure-Python popitem() miscounting len() in the same case
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1625.
Fixed del md[key] and popall() also
removing, and popall() also returning, the pairs a finalizer of a
removed key or value added while the call was still running, by
releasing the removed pairs only once every match is gone, and fixed the
pure-Python implementation raising IndexError or miscounting
len() in the same case, for md[key] = value too
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1627.
Fixed a crash when Python code run between the items of
update() (the argument's own iterator, a key's
lower() or a finalizer) read the multidict, where the pure-Python
implementation returned None for the pairs the call was about to
remove, and fixed update() and merge()
overwriting or losing pairs when that code changed the multidict. An
update from another multidict raised RuntimeError when a key's
lower() mutated the source -- by @asvetlov.
Fixed a use-after-free in to_dict() on Python
3.10 and 3.11, where allocating a value list can run a garbage collection
whose finalizers mutate the multidict, by refusing that mutation with
RuntimeError as it is refused from a key's __hash__()
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1630.
Fixed repr() of a multidict reading freed memory on the GIL build
when a key's or value's __repr__ called an extend(), update()
or merge() that grew the table and then failed, by bumping the version
on every resize -- by @asvetlov.
Related issues and pull requests on GitHub:
#1658.
Fixed CIMultiDict returning a key added as a
str subclass with the spelling of the subclass's own
__str__() instead of the key's, and stopped the C extension from
replacing such a key with its istr while reading it, which
could run the key's finalizer inside the read. Documented that
CIMultiDict converts keys to istr
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1640.
Fixed the istr keys that
extend(), update(),
merge() and the MultiDict
constructor took from a CIMultiDict: they carried the
key's own spelling as their case-insensitive identity instead of its
lower-cased form, so a CIMultiDict keyed by one of them
found it under no spelling of the key. Changed the pure-Python
implementation to hand over those keys as istr too, as
the C extension did -- by @asvetlov.
Related issues and pull requests on GitHub:
#1647.
Fixed a lock-free lookup on the free-threaded build reading the hash
through a key of a compact table that a concurrent delete had already
freed, by taking a reference to such a key first unless it is the very
object being looked up -- by @asvetlov.
Related issues and pull requests on GitHub:
#1653.
Changed CIMultiDictProxy.__init__() called without an argument to
report the same TypeError message as MultiDictProxy and the
pure-Python implementation -- by @asvetlov.
Related issues and pull requests on GitHub:
#1677.
Made MultiDict in the C extension store a mapping whose
keys are all exact str in half the space, keeping no separate
identity or hash per entry, and did the same for a
CIMultiDict whose keys are all exact
istr, taking each identity from the key's canonical
form, which cut its size by about 45%. Copying such a mapping became
about 55% cheaper in instructions and clear()
about 27%; any other key moves the mapping to the full layout, and
clear() lets it start compact again. A lookup
also stopped taking a reference to the identity of an exact str
or istr key, which made d[key] and key in d up
to 7% faster on the GIL build and up to 20% cheaper in instructions on the
free-threaded one -- by @asvetlov.
Related issues and pull requests on GitHub:
#1644, #1648, #1651.
Sped up getall(), del d[key], replacing a
key with d[key] = value and update() by 5
to 14 percent, and up to 26 percent in the pure-Python implementation,
when no key repeats, which each hash table started to track; the
tracking made add() up to 10 percent slower,
and building a small multidict from items up to 18 percent, both up to
16 percent in the pure-Python implementation -- by @asvetlov.
Related issues and pull requests on GitHub:
#1654.
Replaced the set of visited entries that getall(),
to_dict() and the items view's & kept with a
check that each match's entry index is above the previous one's, since
equal keys are first reached in insertion order. A
getall() of a key with many values needed up to
half fewer instructions -- by @asvetlov.
Sped up update() and
merge(): they stopped growing the table up front
when it could hold the argument even empty, as dict.update() does,
stopped holding extra references to the keys and values of a dict
argument, and walked the hash chain for each item without a function
call. An update of keys already present needed 13% to 29% fewer
instructions -- by @asvetlov.
Related issues and pull requests on GitHub:
#1747, #1755, #1757.
Made d[key] = value in the C extension keep the collector for
duplicate keys off the stack until a duplicate turns up, which made
replacing a key about 4-6% and adding a new one about 2-3% cheaper in
instructions -- by @asvetlov.
Related issues and pull requests on GitHub:
#1650.
Sped up building an istr from a str in the C
extension by copying the string directly instead of going through
str.__new__(); reading a str key of a
CIMultiDict for the first time and
popitem() became about 30% cheaper in
instructions -- by @asvetlov.
Related issues and pull requests on GitHub:
#1645.
Stopped a multidict with no watchers from testing for them on every
record made by del d[key], d[key] = value and pop(): each
operation checked once and ran a copy without the watcher code, which
recovered most of the cost the watchers C API had added to them. Changed
free-threaded builds to reserve versions for each multidict in blocks of
256, so a mutation advanced a counter in the object itself instead of
looking up a thread-local one, which cut 2 to 5 percent of the
instructions of every mutation -- by @asvetlov.
Stopped clearing each field of a hash table that is being freed, and read
its entry count once instead of after every release, cutting about 8% off
tearing down a table of 200 entries in clear() and on deallocation;
replaced Py_CLEAR() on success paths with Py_DECREF() or
Py_XDECREF(), which made building a multidict from items up to 1.7%
and comparing one with another mapping up to 4% faster
-- by @asvetlov.
Started emitting DeprecationWarning on access to multidict.upstr,
the alias for istr deprecated since 2.0, and removed it
from multidict.__all__
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1680.
Changed keys() and iteration over
MultiDict, CIMultiDict and their
proxies to yield each key once, as spelled in its first item, so that
CIMultiDict([('X-Foo', '1'), ('x-foo', '2')]) iterates over 'X-Foo'
alone and len(d.keys()) counts distinct keys; items(),
values() and len(d) kept every item
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1654.
Stopped calling a str subclass's own lower() for the
case-insensitive identity of a key in CIMultiDict and in
IStr_FromUnicode(); the identity came from str.lower() instead,
so computing it ran no user code -- by @asvetlov.
Related issues and pull requests on GitHub:
#1746.
Moved the project metadata from setup.cfg to pyproject.toml per
621, and removed the now-empty setup.cfg along with its obsolete
[bdist_wheel] universal flag. Adopted 639 license metadata:
declared the license as the SPDX expression Apache-2.0 and moved
license-files into the [project] table, which raised the
build-time requirement to setuptools >= 77.0. Built distributions
started carrying License-Expression in place of the legacy
License field -- by @aiolibsbot.
Related issues and pull requests on GitHub:
#1646.
Added the benchmarks/ scripts to the source distribution, so that the
test suite run from an unpacked sdist also covers the benchmark driver
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1624.
Stopped the wheel jobs that run under QEMU from running the pure-Python
half of the tests marked expensive, and the tests marked threaded
on GIL builds, which had grown enough since 7.0.0 to take the manylinux
riscv64 job past its two-hour timeout; native jobs still run them
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1762.
Added tools/check_inlining.py and an Inlining CI job, which
fail when GCC stops inlining a helper that a hot path depends on, or
makes an out-of-line copy of an inline function from the CPython headers
or from pythoncapi_compat.h such as Py_DECREF(), since each time
that happened a benchmark regressed on code the change never touched
-- by @asvetlov.
Stopped passing -Wno-conversion to the C extension build, which
had silently turned -Wconversion off, and added the explicit casts
it then required -- by @asvetlov.
Related issues and pull requests on GitHub:
#1611.
Dropped the macOS and Windows legs of the CI test matrix, since
cibuildwheel already ran the same test suite against the very wheels
those jobs installed. Moved every artifact the CI test jobs consume into
one first stage, the source distribution together with the pure-Python
and Linux binary wheels, and switched the remaining test jobs to that
source distribution instead of a repository checkout. The wheel builds
for the other platforms stopped gating the test matrix, the
windows-11-arm wheel build got limited to release tags again, and
x86_64 macOS wheels started being tested under Rosetta outside pull
requests only -- by @asvetlov.
Added CodSpeed benchmarks for mutating a multidict that a watcher is
attached to and for re-initializing a populated multidict, started
benchmarking the FT build alongside the GIL one, and split each build's
benchmark CI job into two parallel shards of similar length, selected by
a new benchmark_shard_2 pytest mark -- by @asvetlov.
Changed benchmarks/callgrind_driver.py to run its Callgrind children from a
fixed-length staging directory holding only the modules they import, so that
two checkouts of one commit at different paths no longer measure up to 3%
apart on allocating operations, and pinned mimalloc's arena address and purge
timer, which had moved free-threaded counts between two runs of one tree.
Stopped the driver from collecting whole-process counts, which cannot be
compared with bracketed ones, unless it is given --whole-process, and
made its checks before measuring, including --self-check, cover only
the implementations selected with --impl, so a pure-Python install can
be measured -- by @asvetlov.
Related issues and pull requests on GitHub:
#1623, #1624, #1637.
Added a test that runs every mutating method of both multidict classes
with keys, values and argument items whose finalizers mutate the
multidict, and checks the C implementation against the pure-Python one,
along with reusable helpers for writing such tests, and a threaded stress
test that re-initializes a shared multidict whose old values each have
such a finalizer while other threads run lock-free lookups and iterate it
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1621, #1631, #1633.
Made the ThreadSanitizer run reliable: added
tools/tsan_suppressions.txt for the faulthandler watchdog's
traceback dumps, used by the CI job and by the documented local command,
which also gained --no-cov; and changed the two pure-Python deadlock
tests for reciprocal view operations to fail only when their workers stop
making progress for 20 seconds, rather than when they have not finished
within 20 seconds; and rewrote the retry loop of the test for
reference leaks on memory errors so that it takes the same branches on
every Python version
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1655.
Renamed the hypothesis-freethreading CI job to hypothesis-ft
(Hypothesis FT), on a par with the GIL one, and removed the update
argument of the debug-only ASSERT_CONSISTENT() check, which stopped
doing anything once update() kept its doomed
entries whole -- by @asvetlov.
Moved the sanitizer build recipes and the performance measurement
procedure out of AGENTS.md into Claude Code skills under
.claude/skills/, and dropped its file layout table, so that less of
it is loaded into every agent session
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1692.
Reworked how the C extension spends GCC's inlining budget, which it sits
at as a single translation unit. Dropped the ALWAYS_INLINE macro and
inline from almost every helper, leaving the decision to the
compiler, and turned the hash table index lookup, store and probe helpers
into macros. Moved the table resize and reservation, clear(),
popitem(), the equality comparison, the iterator and view
constructors, the end of a lock-free read and other rarely run code out
of line, and made the proxy methods tail-call the shared ones. Marked the
error-raising helpers, __sizeof__, __reduce__() and the proxy
constructors' error paths cold, and kept NOINLINE only for the few
helpers GCC re-inlined without it. Moved argument parsing into the method
entry points and returned the keyword-bound arguments by value, which
dropped a stack protector canary from every get(), getone(),
getall(), add() and setdefault() call
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1613, #1614, #1661, #1664, #1667,
#1670, #1671, #1672, #1683, #1690,
#1691, #1693, #1694, #1695, #1697,
#1698, #1699, #1700, #1701, #1702,
#1703, #1704, #1705, #1707, #1709,
#1711, #1712, #1713, #1714, #1717,
#1718, #1719, #1720, #1722, #1723,
#1724, #1726, #1727, #1730, #1732,
#1733, #1734, #1735, #1737, #1739,
#1740, #1749, #1750, #1751.
Shrank the C extension by merging duplicated code: the bodies of
__init__(), extend(), update() and merge(), of pop()
and popone(), the per-class copies of the entry points, the update
loops and the tp_vectorcall slots, which read the class at run time
instead, the repeated tails of the table rebuilds and of getall() and
popall(), and the module setup, which became two table loops. The
set operators and comparisons of the keys and items views were rewritten
around shared helpers compiled for size. When they landed, the view
rewrite made the compiled extension about 6% smaller and reading the
class at run time about 15%, which left room for the compact tables; the
view set operators became up to 17% slower and update() and
merge() up to 3% slower -- by @asvetlov.
Related issues and pull requests on GitHub:
#1669, #1673, #1675, #1677, #1678,
#1679, #1706, #1715, #1716, #1725,
#1728, #1729, #1731.
Reorganized how the C extension's hash table lays out and reaches its
entries: split the entry into a key and value layout that every table
kind shares and a layout that adds the identity and its hash, computed
the entry size with sizeof, told a deleted entry by its key alone,
replaced the md_pos_t cursor and the entry stepping helpers with
typed, indexed access, and walked a table in one loop per table kind and,
for compact tables, per class. repr(), the views' | and -,
the update from another multidict and the comparison with a plain
mapping moved onto the shared walk over all entries
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1641, #1649, #1656, #1659, #1660,
#1662, #1663, #1665, #1674, #1676,
#1741, #1742, #1743, #1745, #1748.
Merged the C extension's two ways of dropping a hash table it no longer
used, freeing it at once on GIL builds and retiring it on free-threaded
ones, into one helper -- by @asvetlov.
Related issues and pull requests on GitHub:
#1618.
Upgraded pythoncapi_compat, and stopped copying the fast
path of PyUnstable_TryIncRef() into the free-threaded C
extension, where it read CPython's private reference count fields, by
calling the function itself instead -- by @asvetlov.
Removed dead code from the C extension: a self-update branch no caller
could reach, version guards that are always true on the supported Python
versions, unused helpers and single-use forwarding wrappers. Renamed the
header-local helpers to carry a leading underscore, moved
istr_canonical() next to the istrobject struct, and folded the
pure-Python update() match test back into one condition
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1635, #1668, #1689,
Note truncated.
Breaking change. istr can no longer be subclassed, on either backend. Code that subclassed it has to wrap or convert instead. The C extension also sta…
7.0.0 is a major release. It makes istr final, adds a
public C API for third-party extensions, and makes the C extension
substantially faster across the board.
Breaking change. istr can no longer be subclassed,
on either backend. Code that subclassed it has to wrap or convert instead.
The C extension also started rejecting an argument passed both
positionally and by name with TypeError, where it used to silently
misbehave.
Public C and Cython API. Other C extensions and Cython modules can now
create, read and mutate multidicts through a capsule, without going through
the Python-level API. The capsule comes with multidict_capi.h, a
cimport-able multidict/__init__.pxd and
multidict.get_include(). The API includes MultiDict_ForEach() and
a watchers API modeled on CPython's dict watchers. See the C API and
Cython API references.
Performance. On CodSpeed's GIL-build benchmarks against 6.9.1, 82
C-extension benchmarks got faster and none got slower. Rough figures:
CIMultiDict keyed by plain str: construction,add(), extend(), update() and item assignment got 2 to 3.8CIMultiDict keyed by istr, andMultiDict: lookups, insertion andgetall() and iteratingitems() got about 35% faster, and the view setpopitem() on a whole mapping went fromMultiDict and 4 to 5CIMultiDict with lowercase strMultiDict cost 2% to 7% more. With several threadsd[key] = value got about four times faster.The pure-Python backend is unchanged in speed.
Robustness. This release fixed several crashes and leaks in the C
extension, including use-after-free bugs when a key's lower() or a
value's __eq__ mutated a multidict, a crash at interpreter shutdown,
and table and reference leaks on the free-threaded build.
Fixed a segmentation fault in the C extension when a finalizer let the
same multidict be resized while del md[key] or
popall() was still scanning the table, and a
spurious RuntimeError from getall()
when a garbage collection triggered while building its result mutated
the multidict -- by @asvetlov.
Fixed a reference leak of the key and value in the C extension when
growing the hash table fails with MemoryError during
add(), update(),
merge(), setdefault()
or item assignment -- by @asvetlov.
Related issues and pull requests on GitHub:
#1507.
Fixed a data race on the free-threaded build of the C extension where a
MultiDict's own version counter was read by
getversion() and written on every mutation with a plain,
non-atomic load and store. Both sides switched to a relaxed atomic
load/store, matching the pattern already used for the object's item
count -- by @asvetlov.
Related issues and pull requests on GitHub:
#1514.
Fixed the C extension discarding a MemoryError or
KeyboardInterrupt raised by a positional argument's __len__
while the size of an update was estimated. extend(), update(),
merge() and subclass construction started propagating it, matching the
pure-Python backend; only the TypeError from an unusable
__length_hint__ stayed ignored -- by @asvetlov.
Related issues and pull requests on GitHub:
#1517.
Ensured that canonical keys in pure-Python MultiDict and
CIMultiDict are converted to exact str instances
without invoking object.__str__() overrides, matching the C extension --
by @agustin18.
Related issues and pull requests on GitHub:
#1535.
Fixed the argument counts reported by the C extension when a method is
called with too many positional arguments; methods with an optional
second argument claimed to take exactly one, and methods requiring two
claimed to take a range. A method called with none of its two required
arguments now reports both as missing -- by @asvetlov.
Related issues and pull requests on GitHub:
#1545.
Fixed the C extension accepting an argument passed both positionally and
by name instead of raising TypeError. md.add("a", key="k", value="v") stored a wrong value, and md.get("a", "b", key="c")
silently ignored the surplus arguments -- by @asvetlov.
Related issues and pull requests on GitHub:
#1545.
Fixed a memory leak in the free-threaded build of the C extension.
clear() hands its hash table to a drain that
frees it once no lock-free reader is in flight, and a drain that finds
one leaves the table for the next drain to free. Deallocation ran no
drain of its own, so a multidict dropped after such a clear, with no
operation in between, never freed that table nor released the references
its entries still held. Builds with the GIL enabled emit byte-identical
code and are unaffected -- by @asvetlov.
Fixed a crash during interpreter shutdown in the C extension: a
multidict started keeping a reference to its own module, so the module
state it reads while being torn down could not be freed first
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1563.
Fixed retired hash tables lingering in the free-threaded build of the C
extension. A drain that finds a lock-free reader in flight puts the tables
it holds back for the next drain to run, and the reader it saw could
already have looked, so nothing was left to free them: they stayed until
the multidict was next used, and if it was dropped instead, the references
their entries held were never released. Builds with the GIL enabled emit
byte-identical code and are unaffected -- by @asvetlov.
Related issues and pull requests on GitHub:
#1565.
Fixed reference cycles through a multidict going uncollected in the
free-threaded build of the C extension. A hash table waiting to be freed
still owns the references of the entries left in it, and those were not
reported to the garbage collector, which then read the values as reachable
from outside the cycle and kept it alive. Builds with the GIL enabled emit
byte-identical code and are unaffected -- by @asvetlov.
Related issues and pull requests on GitHub:
#1566.
Fixed extend(), update() and merge() of the C extension
raising SystemError when given an object whose
__length_hint__() reported close to sys.maxsize together
with keyword arguments, and overflowing the table size computation for
such a hint; a hint too large to reserve for is ignored, as
list.extend() does -- by @asvetlov.
Related issues and pull requests on GitHub:
#1577.
Fixed a use-after-free in the C extension's MultiDict.__eq__
when a value's __eq__ mutated either multidict; comparing
exact str values also became faster -- by @asvetlov.
Related issues and pull requests on GitHub:
#1579.
Fixed a use-after-free in the C extension when a str subclass
key's lower() mutated the source while a
CIMultiDict was built from, extended, updated or
merged with a dict or a multidict of the other case
sensitivity. The extra references this takes are limited to the keys
whose lower() can run Python code, so other updates do not pay for
them -- by @asvetlov.
Added a public C and Cython API for third-party extensions. Its
MultiDict_ForEach() family hands a visitor each entry's key, value,
identity (the canonical key the mapping looks entries up by, lower-cased
on a CIMultiDict) and the identity's hash
-- by @Vizonex and @asvetlov.
Related issues and pull requests on GitHub:
#1478.
Changed the C extension to track the entries that
getall(), item assignment,
update(), merge()
and to_dict() have visited in a bitmap private
to the call, instead of temporarily marking entry hashes in the table
itself. Concurrent readers on free-threaded builds no longer have to
tolerate marked entries. Item assignment,
to_dict() and lookups of keys with many values
got faster, while update() with many repeated
keys got somewhat slower
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1505.
Replaced the C extension's per-key finder object with a single
callback-driven walk in a new _multilib/walk.h header. The entries a
walk has already reported were tracked by the walk itself rather than
by every caller, and the walk kept its position in registers instead of
reloading it from a struct on each step, so
getall() and the items and keys view set
operations got faster on keys with many values -- by @asvetlov.
Related issues and pull requests on GitHub:
#1522.
Sped up CIMultiDict operations keyed by a plain str in the C
extension. An all-ASCII key was lowered directly instead of through a
str.lower() call, and a key that is already lowercase became its own
identity, which saved both a copy and a rehash. Looking up a lowercase key
got about three times faster and inserting one about twice as fast, keys
that do carry uppercase gained 20% to 35%, and bulk operations on
already-lowercase keys, such as extend() and
construction from a dict, got up to three times faster
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1527.
Sped up building an istr in the C extension. The class gained its own
vectorcall, so the one-argument form stopped packing its argument into a
tuple for type_call() only to unpack it again with
PyArg_ParseTupleAndKeywords(). Building an istr from a
str got about 19% faster, and building one from an istr,
which hands back the original, about three times faster. In exchange
getall() costs nine more instructions per
call, since the added code moves where the compiler draws its inlining
line -- by @asvetlov.
Related issues and pull requests on GitHub:
#1537.
Sped up building and dropping multidicts in the C extension by reusing
freed hash tables instead of returning them to the allocator. Small
tables are kept in a bounded per-size-class pool held in the module
state, the way CPython reuses key objects for dict. Only on
builds with the GIL: a shared pool needs an atomic exchange to take a
table and another to put one back, which measured slower than the
per-thread allocator a free-threaded build already has
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1539.
Sped up copy(), constructing a multidict from
another one, and growing one past its current size in the C extension, by
not zeroing the parts of a hash table that are about to be overwritten. A
copied table is overwritten in full, and a resized one has the entries it
carries over written straight onto it -- by @asvetlov.
Related issues and pull requests on GitHub:
#1540.
Sped up items(),
keys(), values()
and iterating a multidict in the C extension, by reusing the view and
iterator objects they allocate instead of returning them to the
allocator -- by @asvetlov.
Related issues and pull requests on GitHub:
#1541.
Sped up constructing a MultiDict,
CIMultiDict, MultiDictProxy or
CIMultiDictProxy in the C extension, by reusing the
object itself as well as its hash table. Subclasses are unaffected and
keep allocating as before -- by @asvetlov.
Related issues and pull requests on GitHub:
#1542.
Sped up getall() in the C extension, by
shrinking the scratch buffer the walk carries on the stack from 4 KB to
1 KB. The buffer was large enough to stop the compiler inlining the walk
into its callers, which cost more than it saved -- by @asvetlov.
Related issues and pull requests on GitHub:
#1543.
Sped up lock-free reads on the free-threaded build of the C extension.
Every reader that leaves a MultiDict last used to
call into the retired-table drain, whose first act is a sequentially
consistent read-modify-write, even though the retired list is empty
almost every time; a plain load took over that case and the drain
itself moved out of line. key in md got about 10% cheaper,
md[key] about 8% and a missing md.get(key) about 5%, measured as
instruction counts on CPython 3.14. Builds with the GIL enabled emit
byte-identical code and are unaffected -- by @asvetlov.
Related issues and pull requests on GitHub:
#1546.
Added a watchers C API, so another C extension can be notified when a
MultiDict or CIMultiDict changes.
It follows CPython's PyDict_AddWatcher() family, with two
differences: the callback is passed two context pointers, one fixed per
registered callback and one per watched multidict, and events are
delivered once the operation that produced them has finished rather than
mid-mutation. The events are multi-value aware, and each one that concerns
a key carries the identity and its hash alongside it. Up to
MULTIDICT_MAX_WATCHERS (32) watchers can be registered, and a
failing callback is reported through sys.unraisablehook() with the
event and the multidict's type and address, as CPython does for dict
watchers -- by @asvetlov.
Related issues and pull requests on GitHub:
#1551.
Sped up lookups, insertions and deletions in the C extension by always
inlining the hash-table probe's initializer. Left to itself the compiler
emitted it out of line, so every probe opened with a call for five
stores. Free-threaded builds already inlined it and are unchanged
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1554.
Sped up every method call on CIMultiDict and
CIMultiDictProxy in the C extension by binding
their method descriptors to the class itself instead of inheriting
them from MultiDict and
MultiDictProxy; CPython only specializes a
method call when the descriptor belongs to the exact type of the
receiver, so every such call was taking the generic path
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1567.
Made popitem() in the C extension drop the
trailing deleted entries it walked over, so that popping every item
out of a mapping costs linear time rather than quadratic; the
pure-Python implementation already did this -- by @asvetlov.
Related issues and pull requests on GitHub:
#1568.
Sped up iterating over items() in the C
extension by handing out the same (key, value) tuple again when
the caller has already let go of it, as for key, value in d.items() does on every step; a tuple the caller keeps is left
alone and a new one is used from then on -- by @asvetlov.
Related issues and pull requests on GitHub:
#1569.
Sped up lookups on the free-threaded build by comparing the hash before
touching an entry's key and by matching a key that is the very same
object as the one looked up without taking a reference to it, so the
common hit costs two fewer atomic operations -- by @asvetlov.
Related issues and pull requests on GitHub:
#1570.
Sped up mutations on the free-threaded build when several threads
mutated multidicts at the same time: each thread reserved version
numbers in batches instead of bumping one counter shared by every
thread on each mutation, so with six threads a d[key] = value was
about four times faster -- by @asvetlov.
Sped up the C extension by keeping the table resize on the insert path and
the operations behind Python slots out of line, so they stopped crowding the
hot paths out of the compiler's inlining budget. On GIL builds, building a
multidict from 200 pairs got about 4% faster in instruction count and
inserting a new key about 2%; on free-threaded builds, key lookups got up to
5% faster -- by @asvetlov.
Made istr a final class to disallow subclassing in both
the C extension and pure-Python implementations -- by @agustin18.
Related issues and pull requests on GitHub:
#1535.
Documented how MultiDict and
CIMultiDict compare with dict on both the
default and the free-threaded builds, with per-operation measurements,
and how to run the benchmarks in a stable environment
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1532.
Documented what istr keys are worth on a
CIMultiDict, as instruction counts next to the
equivalent plain str keys. The advice to create them once and
reuse them was already there; the numbers behind it were not
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1547.
Documented that a few rows of the benchmarking reference are measured on
an empty or 20-item mapping rather than the 200-item one the section
describes -- by @asvetlov.
Related issues and pull requests on GitHub:
#1558.
Mentioned in the README and docs index that multidict is optimized for
both the GIL and free-threaded builds, and that its performance is typically
within 20-30% of dict for common operations, with a link to the
benchmarks page for the full breakdown -- by @asvetlov.
Related issues and pull requests on GitHub:
#1559.
Fixed the C extension's popone() and pop() docstrings, which
claimed the last occurrence of the key was removed; the first one is,
as on the pure-Python backend. Added a note on rejecting duplicate
values for keys that must carry exactly one, such as security-sensitive
HTTP headers -- by @asvetlov.
Related issues and pull requests on GitHub:
#1578.
Added a C API example of a watcher that invalidates a cache: it unwatches
from its own callback, watches again on the next rebuild, and gives up
on a multidict that keeps changing -- by @asvetlov.
Related issues and pull requests on GitHub:
#1586.
Refreshed the tables in the benchmarking reference for the 7.0.0 release,
and replaced the range quoted for the pure-Python backend with its average
slowdown -- by @asvetlov.
Related issues and pull requests on GitHub:
#1602.
Removed the C extension code path that existed only for free-threaded
Python 3.13, whose support was dropped in v6.8.0. The free-threaded
build started requiring CPython 3.14 or newer and failing with a clear
compiler error on older free-threaded interpreters; builds with the GIL
enabled were unaffected and still go down to Python 3.10
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1524.
Added CodSpeed benchmarks for getall() on a large table, update()
with duplicate keys, merge(), to_dict() and items view containment
with duplicate keys -- by @asvetlov.
Related issues and pull requests on GitHub:
#1506.
Added CodSpeed benchmarks for getall() on keys with many values, both
in a small table and in a table large enough to need a heap allocated
bitmap -- by @asvetlov.
Related issues and pull requests on GitHub:
#1509.
Moved the compiler hint macros used by the C extension out of
htkeys.h into compiler.h and dropped the HT_ prefix from
their names -- by @asvetlov.
Related issues and pull requests on GitHub:
#1510.
Moved the debug-only ASSERT_CONSISTENT macro and the
_md_check_consistency() and _md_dump() helpers used by the C
extension out of hashtable.h into their own debug.h header
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1513.
Added CodSpeed benchmarks for CIMultiDict keyed by plain str instead
of istr, covering already-lowercase, mixed-case and non-ASCII keys
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1521.
Documented that the pure-Python test leg needs --no-c-extensions
alongside MULTIDICT_NO_EXTENSIONS=1; the environment variable picks
the backend that gets imported, but only the flag deselects the
C-extension half of the test matrix -- by @asvetlov.
Related issues and pull requests on GitHub:
#1523.
Documented how to run the Hypothesis property tests, which a default
pytest run excludes via the -m "not hypothesis" marker
expression in pytest.ini and whose pin make install-dev does
not install -- by @asvetlov.
Related issues and pull requests on GitHub:
#1525.
Added tools/codegen_diff.py, which reports the functions GCC emits
differently before and after a change to the C extension. Addresses,
RIP-relative displacements and branch offsets inside a symbol are
normalized away first, so a single instruction-length change no longer
makes every later function look modified, while constants and structure
offsets are left alone so that a real change is not folded away. A tree
that fails to build is reported by name rather than compared
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1529.
Fixed benchmarks/istr.py to import pyperf under its current
name; it had been failing to start since the package was renamed from
perf -- by @asvetlov.
Related issues and pull requests on GitHub:
#1531.
Rewrote the benchmark scripts around a single registry of benchmarked
operations, and added a Callgrind driver that records deterministic
instruction counts per operation. The measurements of destructive
operations such as pop() and clear() stopped counting the cost
of rebuilding the mapping -- by @asvetlov.
Related issues and pull requests on GitHub:
#1532.
Moved the bulk extend(), update() and merge() paths of the C
extension out of hashtable.h into their own bulk_update.h
header, leaving hashtable.h to the table itself and the single-key
operations -- by @asvetlov.
Related issues and pull requests on GitHub:
#1533.
Moved the two list-reading helpers behind itemsview.__contains__() and
the extend()/update()/merge() sequence parser out of
hashtable.h and into a new multidict/_multilib/unpack.h, together with
a shared unpack_pair() that both call to read a (key, value)
pair out of an exact two-element tuple or list. C extension only; the pure
Python implementation has no header layout to mirror
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1534.
Added a MULTIDICT_NO_FREELIST=1 build option that stops the C
extension reusing freed blocks. A reused block never reaches free(),
so AddressSanitizer can neither poison it nor report a use-after-free on
it; an instrumented run wants a second pass with this set, as
AGENTS.md describes.
Also added a private _freelist_clear() to the C extension, for tests
that inject an allocation failure: a reused block lets an operation run
without calling the allocator at all, which would put the recovery path
out of reach -- by @asvetlov.
Related issues and pull requests on GitHub:
#1539.
Added a visitor to the public C API test harnesses that mutates the multidict
it is walking, covering the reentrancy guard in MultiDict_ForEach() from
both multidict._testcapi and multidict._testcyapi
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1544.
Added benchmarks for the shapes where an allocation is most of the work:
an empty mapping, a 20-item one built or copied, a view or an iterator
taken on its own, and a proxy. The allocation an operation makes is a
fixed cost, so at the existing 200-item size it was divided across 200
entries and all but vanished; d.items() measured 1.7% faster where
it is 46% faster -- by @asvetlov.
Related issues and pull requests on GitHub:
#1556.
Sorted docs/spelling_wordlist.txt case-insensitively, and documented
in AGENTS.md that it is to be re-sorted rather than appended to. The
list had grown in append order, which made a word hard to find and put
every PR adding one on the same last line -- by @asvetlov.
Related issues and pull requests on GitHub:
#1557.
Stopped asking every performance pull request to regenerate the tables in
the benchmarking reference, since each run shifts every row slightly and the
churn hides the rows that moved, and moved the refresh to once per release.
Added RELEASE.md, which documents the release procedure, including
that refresh -- by @asvetlov.
Related issues and pull requests on GitHub:
#1564.
Renamed the C extension's internal helpers in the _multilib headers
so that a leading underscore marks exactly the functions used only
within their own header; type slots and method callbacks keep their
names -- by @asvetlov.
Related issues and pull requests on GitHub:
#1571.
Spelled out in AGENTS.md that every verb in a news fragment is
past tense, and added a bad and good example pair plus a matching
"Things not to do" entry -- by @asvetlov.
Related issues and pull requests on GitHub:
#1588.
Changed the release workflow to rewrite the Sphinx roles of the changelog
section as Markdown before publishing it as the GitHub Release body, so
that class, method and exception references, contributor handles and
documentation links render there instead of showing as raw markup
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1596.
Changed the internal htkeys_build_indices() helper in the C
extension to return void; it cannot fail, so both callers were
checking for an error that never occurred -- by @asvetlov.
Related issues and pull requests on GitHub:
#1516.
Routed every field of the C extension's hash table that more than one
thread can reach through a named accessor in the new
_multilib/freethreading.h header, instead of each call site choosing
between an atomic and a plain access behind #ifdef Py_GIL_DISABLED.
Each accessor kept the operation and memory ordering its call sites
already used, so neither build changed behavior -- by @asvetlov.
Related issues and pull requests on GitHub:
#1518.
Removed get_mod_state_by_cls(), an unused helper in the C extension
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1520.
Documented why md_clear() keeps clearing entry fields with
Py_CLEAR() on the GIL build rather than routing them through
_multilib/freethreading.h: Py_CLEAR() skips the store when a
field is already NULL, so no accessor there is a substitute for it
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1526.
Collapsed the two #ifdef Py_GIL_DISABLED arms that fill in a freshly
inserted hash table entry in the C extension into one. Both builds started
writing the entry's fields in the order the free-threaded build needs, and
publishing the value without first loading and releasing the one already
there, since a fresh entry holds none -- by @asvetlov.
Related issues and pull requests on GitHub:
#1528.
Merged the C extension's deferred-decref accumulator into reflist_t,
so one collector of strong references, in the _multilib/reflist.h
header, came to serve both the callers that build a result list and the
ones that postpone a decref until a mutation finishes. Its storage became an
inline array that spills into a chain of heap blocks, which also made
getall() a little faster on keys with many
values, since the collected values stopped being copied to a bigger
buffer as they accumulate -- by @asvetlov.
Related issues and pull requests on GitHub:
#1530.
Dropped two module state lookups in the C extension that had already been
answered elsewhere. MultiDict.__init__() and CIMultiDict.__init__()
switched to reading the state pointer that tp_new() stored on the object instead
of resolving it from the type a second time, and an istr stopped
carrying a state pointer of its own, which was written on every creation
and never read. Constructing a subclass of either container got about 1%
cheaper, and every istr is eight bytes smaller
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1536.
Narrowed the critical section of the single-key methods (add(),
__setitem__(), __delitem__(), setdefault(), pop(),
popone(), popall() and getall()): a key is converted to its
identity and hashed before the lock is taken, rather than while a
free-threaded build holds the multidict locked -- by @asvetlov.
Related issues and pull requests on GitHub:
#1538.
Changed MultiDict_ForEach() to walk every entry with a linear scan
instead of the resumable iterator cursor, which its caller does not need while
it holds the critical section for the whole walk -- by @asvetlov.
Related issues and pull requests on GitHub:
#1544.
Sped up the argument parsing of the C extension. The all-positional form
of every method taking key started binding its arguments inline instead of
calling into the keyword machinery, and keyword names were matched against
interned parameter names by identity before falling back to a string
comparison -- by @asvetlov.
Related issues and pull requests on GitHub:
#1549.
Stopped building the implicit None default of MultiDict.get()
on every call in the C extension, and produced it only when the key is
actually missing -- by @asvetlov.
Related issues and pull requests on GitHub:
#1549.
Laid out the common branch in line in the single-key lookup probe loop and in
the key to identity conversion: the deleted-slot check and the str
subclass check were marked as the rare cases, so the compiler stopped putting
the frequent path behind a taken branch -- by @asvetlov.
Related issues and pull requests on GitHub:
#1550.
Stopped building the implicit None default of
MultiDict.setdefault() on every call in the C extension -- by
@asvetlov.
Related issues and pull requests on GitHub:
#1553.
Reordered the fields of the C extension's multidict object so the ones a
lookup reads share a cache line, and dropped its separate module
reference in favor of the one the module state already holds, which
made every multidict 16 bytes smaller -- by @asvetlov.
Related issues and pull requests on GitHub:
#1584.
Dropped the C extension's multidict_tp_traverse() and
multidict_tp_clear() wrappers; md_traverse() and md_clear()
now serve as the type's GC slots directly -- by @asvetlov.
Related issues and pull requests on GitHub:
#1589.
Removed a redundant NULL check and a stray PyErr_Format()
argument from the C constructors, and an always true condition from the
pure Python getall() -- by @asvetlov.
Related issues and pull requests on GitHub:
#1590.
Merged the duplicated __init__() bodies of the C MultiDict and
CIMultiDict into one helper -- by @asvetlov.
Related issues and pull requests on GitHub:
#1591.
Replaced the allocate, set state and take the module reference sequence
repeated across the C constructors with one helper -- by @asvetlov.
Related issues and pull requests on GitHub:
#1592.
Renamed internal helpers whose names no longer matched what they did,
and fixed misspelled docstring identifiers in the C extension
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1593.
Renamed the C extension's functions that are used only as type slots
after the slot they fill, as <type>_tp_<slot>, <type>_nb_<slot>,
<type>_sq_<slot> or <type>_mp_<slot> -- by @asvetlov.
Related issues and pull requests on GitHub:
#1594.
Replaced the custom tp_new slots that blocked direct instantiation of
the C view and iterator types with Py_TPFLAGS_DISALLOW_INSTANTIATION
-- by @asvetlov.
Related issues and pull requests on GitHub:
#1595.
Replaced deprecated instrumentation codspeed mode with simulation -- by :user: asvetlov
Fixed the C extension reading freed memory on free-threaded builds when a
list handed to :py:meth:~multidict.MultiDict.update,
:py:meth:~multidict.MultiDict.extend, :py:meth:~multidict.MultiDict.merge
or the :py:class:~multidict.MultiDict and :py:class:~multidict.CIMultiDict
constructors, a [key, value] item inside any iterable handed to them, or a
list tested with in against :py:meth:~multidict.MultiDict.items, is
changed by another thread; a call that catches the list shrinking under it
now raises :py:exc:RuntimeError -- by :user:rodrigobnogueira.
Related issues and pull requests on GitHub:
#1437.
Fixed a data race on the free-threaded build where a retired hash table's
reader count used relaxed atomics, letting a lock-free get()/getone()/
__getitem__() read race a concurrent free of that table. The reader-exit
decrement and the drain's free check now use release/acquire ordering
instead -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1481.
Fixed a free-threaded build bug where two threads calling update(),
merge(), or __setitem__() on the same key at the same time could lose
the key entirely instead of just racing on which value wins. A decref of the
replaced value could transiently suspend the writer's critical section,
letting a second writer for the same key observe the first writer's
in-progress entry as absent and, once both settled, mistake it for a stale
duplicate and delete it. Every such decref is now deferred until the writer
has released its critical section, so the window can no longer open.
setdefault() had an unrelated instance of the same blind spot (it could
insert a duplicate rather than recognizing an in-flight key), fixed alongside
it -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1483.
Fixed a free-threaded build bug where getall() and the items()/
keys()/values() equality path could raise KeyError or report a
present, never-deleted key as missing. A concurrent update()/extend()/
__setitem__() call can have its critical section transiently suspended
(a decref triggering a blocking allocator call) while an entry is marked as
part of its own bookkeeping; a reader landing in that window used to treat
the mark as "not found" instead of "still there, in flight" -- by
:user:asvetlov.
Related issues and pull requests on GitHub:
#1484.
Fixed a reference leak in the C extension where operand | md.items()
and md.items() - operand leaked one key and one value reference per
element of operand, letting a large operand grow memory without bound
(:gh:GHSA-54p9-h82j-f925 <aio-libs/multidict/security/advisories/GHSA-54p9-h82j-f925>)
-- by :user:asvetlov.
The issue was reported by :user:waydeshi.
Related commits on GitHub:
:commit:350b4a0.
Fixed a segmentation fault on the standard (non-free-threaded) C extension
build when a value type's __del__ released the GIL (for example by
calling time.sleep()) while update(), merge(), __setitem__(),
__delitem__(), pop(), popone(), or popall() was dropping a
replaced or removed value. Py_BEGIN_CRITICAL_SECTION compiles to a no-op
on this build, so nothing else was stopping a second thread from mutating the
very same MultiDict concurrently once the GIL was released mid-mutation.
Every such decref is now deferred until the mutation has fully finished, the
same technique already used to close the analogous free-threaded-build race
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1489.
Fixed the C extension reading freed memory while iterating a
:py:class:~multidict.CIMultiDict whose keys are plain :py:class:str.
Converting such a key to :py:class:~multidict.istr could run Python code
(a :py:class:str subclass's __str__ or __del__) or, on free-threaded
builds, suspend the iterator's critical section, after which the iterator read
the entry again even though a concurrent mutation could already have freed it.
As part of the fix, :py:meth:~multidict.MultiDict.copy and re-initializing
from another multidict now assign a new version in the C extension instead of
reusing the source's, matching the pure Python implementation
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1496.
Fixed a use-after-free on the free-threaded build where a lock-free
get(), [] or in could read a hash table that a concurrent
resize had just retired and another reader was freeing
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1497.
Removed a redundant include and a duplicated exclude line from
MANIFEST.in; sdist contents are unchanged -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1478.
Added .claudeignore file -- by :user:asvetlov
Related issues and pull requests on GitHub:
#1479.
Scaled up the pure-Python pop(), popitem(), __delitem__(),
add() and item-insertion benchmarks to do more work per measurement.
Repeated CodSpeed runs on the same commit showed these particular
benchmarks flagged as dominated by syscalls, understating their real
cost and adding noise to the reported values; a larger working set
amortizes that overhead -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1485.
Replaced deprecated instrumentation codspeed mode with simulation -- by :user:asvetlov
Related issues and pull requests on GitHub:
#1493.
Reorganized the mutating benchmarks (item insertion, update(),
add() of the same key, pop(), popitem(), clear(),
__delitem__() and __setitem__()) to copy a fresh multidict and
apply the operation in a loop, like the add() and extend()
benchmarks already do. The insertion, update() and clear()
benchmarks previously mutated a single multidict shared across
rounds, so only the first round measured the intended operation; the
rest did a single copy per round, letting per-round overhead dominate
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1494.
The repr() and view inequality benchmarks were updated to repeat their
operation in a loop, like the other benchmarks, and the CodSpeed benchmark
job was moved to Python 3.14 -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1498.
Dropped -I from the AddressSanitizer test command in AGENTS.md
and in the CI job. It implies -E, which made Python ignore
PYTHONMALLOC=malloc, so small hash tables were still served from
pymalloc arenas where use-after-free went undetected
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1499.
The CI/CD workflow was updated to stop superseded runs of the same pull request
when a new commit is pushed; runs on master, release branches, tags,
the merge queue, and the daily schedule are never interrupted
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1500.
The release job was changed to upload distributions and their signatures
to the GitHub Release one file at a time, skipping assets that were
already attached and retrying after a pause, so that a parallel upload
burst no longer tripped the GitHub secondary rate limit
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1503.
Corrected several comments in the free-threaded C extension that
attributed critical-section suspension to a blocking PyMem_Malloc()
call; allocation alone never suspends an acquired critical section, and
the real risk at those sites is a decref running
a finalizer or weakref callback. Also dropped a retry loop in
md_clone_from_ht() that guarded against the same, non-existent
allocation-triggered suspension -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1486.
Changed the free-threaded build's deferred decref buffer, used by
update() and __setitem__(), to a chain of fixed-size blocks
with a large inline first block instead of a small inline array that
was reallocated on growth -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1501.
Protected repr() of MultiDict , MultiDictProxy , and their views in the C extension with a critical section, avoiding data races on the free-threaded
Protected repr() of MultiDict, MultiDictProxy, and their views
in the C extension with a critical section, avoiding data races on the
free-threaded build of CPython -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1431.
Guarded repr() of MultiDictProxy in the C extension and of
KeysView in both the C extension and the pure-Python
implementation against infinite recursion on self-referential
containers, matching the existing guard on MultiDict,
ItemsView, and ValuesView -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1431.
Protected MultiDict.update(), .extend(), .merge(), .clear(),
.copy(), and the MultiDict/CIMultiDict constructors in the C
extension with a critical section, using the two-object form when a
second multidict, multidict proxy, or plain dict instance is
involved, avoiding data races and a segmentation fault on the
free-threaded build of CPython. .clear() now also publishes the
empty table before releasing any entry's references, so a concurrent
caller can never observe a partially-cleared multidict even if releasing
a value runs arbitrary Python code that suspends the held critical
section -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1433.
Protected MultiDict.add(), __setitem__/__delitem__,
get()/getone()/__getitem__, __contains__, getall(),
setdefault(), pop()/popone()/popall()/popitem(),
__eq__, iteration, and the &/|/-/^/in/
isdisjoint() operations on .keys() and .items() views in the
C extension with a critical section, avoiding data races and
use-after-free crashes on the free-threaded build of CPython
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1438.
Fixed a data race on the free-threaded build where MultiDictObject.used
was written non-atomically while len() read it with a relaxed atomic
load -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1441.
Fixed several pre-existing use-after-free races on the free-threaded build: _md_resize()
and md_clone_from_ht() could allocate a new hash table, have their
critical section transiently suspended during that allocation, and then use
a keys-table pointer or size a concurrent operation had already invalidated;
_md_del_at() (plain del/pop()) could likewise leave self in
an inconsistent state across a suspending decref -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1441.
Fixed a data race on the free-threaded build where md->keys was
published with a plain store while lock-free readers loaded it
atomically, and a false-negative race in get()/__contains__
where a hash temporarily marked by a concurrent replace could make a
present key look absent -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1441.
Fixed several more pre-existing use-after-free and data-corruption races on
the free-threaded build, this time in the __setitem__()/update()/
extend()/merge() replace path: a decref suspending the critical
section mid-replace could leave a stale table pointer in use, misplace a
temporarily-marked entry during a concurrent resize, or let one thread's
duplicate-tracking mark get overwritten by an unrelated key's entry sharing
the same hash bucket -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1441.
Fixed an unbounded memory leak on the free-threaded build where a retired
hash table could sit on md->retired for the rest of the object's
lifetime under sustained concurrent read traffic instead of being freed
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1443.
Fixed a data race on the free-threaded build of the C extension where
state->global_version, the version counter shared by every
multidict instance and used to derive getversion(), was bumped with
a plain increment instead of an atomic op. Two threads mutating
different instances at the same time could step on each other's update
and hand out a duplicate (or non-monotonic) version number
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1444.
Fixed a data race on the free-threaded build of the C extension where
getall()/popall() and update()/extend()/merge()
temporarily marked and unmarked a matching entry's hash with a plain,
non-atomic store, while a lock-free __contains__/get() on another
thread could load that same field with an atomic operation and no lock at
all -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1448.
Fixed a data race in the pure-Python MultiDict/CIMultiDict
fallback where concurrent mutating calls (add(), __setitem__,
__delitem__, setdefault(), pop()/popone(),
popall(), popitem(), update(), extend(), merge(),
clear(), and re-__init__) could corrupt the shared hash table,
or silently drop a concurrent write, on a regular, GIL-enabled
interpreter. Pure-Python bytecode is not atomic under the GIL, so two
threads could interleave mid hash-table insert or deletion, leaving
the index table pointing at stale or out-of-range entries and causing
an infinite probe loop, an AttributeError, or a lost mutation.
Each MultiDict/CIMultiDict instance now holds its own lock for
the duration of these operations, with two-object operations locking
both instances in a fixed order to avoid deadlock. The C extension
already serializes these operations with a critical section and was
not affected
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1449.
Fixed a data race on the free-threaded build of the C extension where
to_dict() cleared every entry's temporary "seen" mark with a plain,
non-atomic store, while a lock-free __contains__/get() on another
thread could load that same field with an atomic operation and no lock at
all -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1450.
Fixed a data race on the free-threaded build of the C extension where
update()/extend()/merge() read self's module state without
holding self's lock, while a concurrent __init__() on the same,
already-published multidict could rewrite that same field under lock.
Rather than routing every read through a lock-free lookup, the field is
now written only once, at object construction, and never touched again by
__init__(), so it is safe to read unlocked anywhere -- by
:user:asvetlov.
Related issues and pull requests on GitHub:
#1452.
Fixed items() set algebra (&, |, -, in, isdisjoint()) so it no
longer corrupts or hangs when a compared value's __eq__() re-enters the
same :class:~multidict.MultiDict, for example by calling getall() on
it. The comparison now runs only after the internal hash-chain walk has
been fully materialized and restored, so the callback can no longer
observe entries the walk still has marked -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1453.
Fixed a data race on the free-threaded build of the C extension where
popall()/popone()/__delitem__/add() rewrote a hash table
index slot with a plain, non-atomic store, while a lock-free
get()/getone()/__getitem__/__contains__ on another thread
walked that same index array with a plain, non-atomic load and no lock at
all -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1454.
Stopped folding the case of :class:~multidict.istr keys in the
case-sensitive :class:~multidict.MultiDict, so both the C and the pure
Python implementations now keep such a key exactly as given
-- by :user:youdie006.
Related issues and pull requests on GitHub:
#1457.
Fixed the C extension's update() (and merge(), which shares the same
code path) silently failing to invalidate an in-progress keys()/
items()/values() iterator when the call only overwrote an
already-present key's value in place. Adding a new key already invalidated
iterators correctly; only the in-place overwrite branch was missing the
version bump -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1458.
Fixed the pure-Python MultiDict/CIMultiDict iterator guard so that
mutating the mapping before the first next() call on an items(),
keys(), or values() iterator (or on iter(md)) now reliably raises
RuntimeError, matching the C extension. Previously a mutation that
happened before the iterator was ever advanced, such as clear() or a
del/popone() that removed the only remaining entry, could make the
iterator silently raise StopIteration instead -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1459.
Fixed a use-after-free race on the free-threaded build of the C extension
where a retired hash table could be freed while a lock-free
get()/getone()/__getitem__/__contains__ reader on another
thread was still walking it: the coarse "no reader in flight" gate
reaching zero did not reliably mean every such reader had also reached
its own per-table exit, so a table whose own reader count is still
nonzero is now deferred for a later retry instead of freed
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1461.
Added a :meth:~multidict.MultiDict.to_dict method returning a plain
:class:dict that maps every key to the list of all its values. Unlike
dict(md), which keeps only the first value per key, and unlike a
{k: md.getall(k) for k in md} comprehension, which emits one entry per
spelling of a case-insensitive key, it groups by key identity
-- by :user:rodrigobnogueira.
Related issues and pull requests on GitHub:
#783.
MultiDict, CIMultiDict, MultiDictProxy and CIMultiDictProxy in the
C extension implemented the vectorcall calling convention for construction.
This sped up MultiDict(...)-style calls by avoiding an intermediate
positional-arguments tuple and keyword-arguments dictionary in the common case; the gained burst is ~10.
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1436.
Made MultiDict.__contains__(), .get(), .getone() and
__getitem__() lock-free on CPython 3.14+'s free-threaded build, instead
of taking a critical section like the rest of the C extension's API (on
3.13, which lacks the public API these need to safely read an entry
concurrently, they still take the critical section, exactly as before). A
resize/shrink/clear no longer frees the outgoing hash table immediately; it
is deferred until no lock-free reader could still be walking it -- by
:user:asvetlov.
Related issues and pull requests on GitHub:
#1441.
Made the pure-Python MultiDict/CIMultiDict implementation safe to
use from multiple threads under a free-threaded (no-GIL) build. Every
method that touches an instance's hash table now takes that instance's own
lock, two-object operations (update(), extend(), merge(),
__eq__(), copying) lock both objects in a fixed, deadlock-safe order,
and the version counter shared by every instance is guarded separately. On
a regular (GIL-enabled) interpreter this adds no overhead: the locked
methods are the exact same function objects as before -- by
:user:asvetlov.
Related issues and pull requests on GitHub:
#1447.
Sped up adding many values for the same key to a
:class:~multidict.MultiDict or :class:~multidict.CIMultiDict, each
added value no longer gets slower than the one before it
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1466.
Added the plural form "fallbacks" to the docs spell checker's allowed
word list so :file:CHANGES.rst builds cleanly -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1439.
Added race tests that iterate and extend a :class:~multidict.MultiDict
from several threads on a free-threaded build, so a regression in the C
extension's locking fails a free-threaded CI leg instead of going unnoticed
-- by :user:rodrigobnogueira.
Related issues and pull requests on GitHub:
#1317.
Added AddressSanitizer/UndefinedBehaviorSanitizer and
ThreadSanitizer CI jobs. The former runs the suite under a
regular CPython with MULTIDICT_ASAN_BUILD=1; the latter builds a
free-threaded CPython instrumented with --with-thread-sanitizer
and runs the suite against it via MULTIDICT_TSAN_BUILD=1 -- by
:user:asvetlov.
Related issues and pull requests on GitHub:
#1455.
Added Hypothesis-based property and stateful fuzz tests covering
:class:~multidict.MultiDict/:class:~multidict.CIMultiDict semantics,
views, iterators, and threaded stress scenarios, run against both the
C-extension and pure-Python backends -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1460.
Added core dump collection to the AddressSanitizer and ThreadSanitizer
CI jobs: a crash now uploads the core file alongside the crashing
interpreter binary and the compiled extension as a downloadable
artifact, for offline debugging with a matching symbol-carrying binary
instead of raw hex offsets -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1462.
Added benchmarks for adding many values for the same key and for
creating a multidict with many items -- by :user:bdraco.
Related issues and pull requests on GitHub:
#1465.
Skipped abstractmethods from coverage leaks report, they are empty placeholders that are never executed -- by :user:asvetlov
Related issues and pull requests on GitHub:
#1470.
Add benchmarks for __setitem__ -- by :user:asvetlov
Related issues and pull requests on GitHub:
#1471.
The AddressSanitizer and ThreadSanitizer CI jobs were changed to run pytest
with --capture=no. Previously a sanitizer abort could exit the process
before pytest flushed its per-test output buffer, silently dropping the
sanitizer report from the job log -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1472.
Moved the hypothesis pin out of requirements/pytest.txt into its own
requirements/pytest-hypothesis.txt, installed only by the dedicated
Hypothesis CI jobs. hypothesis now ships a Rust extension with no
prebuilt wheel for some emulated architectures (e.g. musllinux i686),
which broke wheel-testing jobs that never run the Hypothesis-marked tests
in the first place -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1475.
Replaced PyIter_Next() calls with the newer PyIter_NextItem() API in
the C extension, so the iterator-exhausted and error cases are told apart by
the return code instead of an ambiguous NULL result -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1440.
Removals and backward incompatible breaking changes
A segmentation fault that could be triggered when getting an item is now fixed
-- by :user:Vizonex.
Related issues and pull requests on GitHub:
#1310.
Fixed reference leak in iterators, views and istr
-- by :user:Vizonex.
Related issues and pull requests on GitHub:
#1311.
Fixed the pure-Python :class:~multidict.MultiDict constructor and
:py:meth:~multidict.MultiDict.extend,
:py:meth:~multidict.MultiDict.update, and
:py:meth:~multidict.MultiDict.merge methods over-allocating their
internal hash table when called with both a positional argument and
keyword arguments, because keyword arguments were counted twice in the
size estimate -- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1338.
Fixed __repr__ of :class:~multidict.MultiDict,
:class:~multidict.CIMultiDict, their proxies, and the keys/items views
producing invalid output when keys contained quote characters --
keys are now formatted with :func:repr so the result is a valid Python
string literal -- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1342.
Fixed a segfault when calling :py:meth:~multidict.MultiDict.add with only one of its two required arguments supplied by keyword, e.g. d.add(key="k"). Extra keyword arguments passed to the lookup and removal methods are also now rejected with :exc:TypeError instead of being silently ignored.
-- by :user:devdanzin
Related issues and pull requests on GitHub:
#1376.
Fixed a segfault when constructing a multidict iterator type directly, e.g. type(iter(md.keys())).__new__(...). Such an iterator had a NULL internal pointer that next() dereferenced. The iterator types now forbid direct instantiation, the same way the view types were fixed in #1163.
-- by :user:devdanzin
Related issues and pull requests on GitHub:
#1377.
Fixed a segfault when using a :py:class:~multidict.MultiDict or :py:class:~multidict.CIMultiDict created via __new__ without calling __init__ (for example a subclass that does not call super().__init__()). The internal state was left as NULL pointers that the first method call dereferenced. tp_new now initializes the object to a valid empty mapping.
-- by :user:devdanzin
Related issues and pull requests on GitHub:
#1378.
Fixed two crashes in the C extension caused by holding a raw pointer into a hash table across an operation that could reshape it. Updating a multidict from itself (e.g. d.extend(d)) freed the very table being iterated -- a use-after-free; extend(self) now doubles the contents and update(self)/merge(self) are no-ops. Building a multidict from a list of pairs whose case-insensitive key .lower() shrinks that list read past the end of the list; the length is now re-checked on every iteration.
-- by :user:devdanzin
Related issues and pull requests on GitHub:
#1379.
Fixed three reference/resource leaks on error paths in the C extension: the items-view __contains__ leaked the first element of a candidate pair when reading the second one raised; __repr__ leaked its PyUnicodeWriter when the multidict was mutated during iteration; and the internal iteration helper leaked the identity reference when key materialization failed under low memory.
-- by :user:devdanzin
Related issues and pull requests on GitHub:
#1381.
Stopped several feature-detection fallbacks in the C extension from swallowing every exception. When probing an argument (arg.items()/arg.keys()) or measuring it (len(other)) fails, the code now clears only the expected :exc:TypeError/:exc:AttributeError and lets everything else -- notably :exc:MemoryError and :exc:KeyboardInterrupt -- propagate, matching the already-correct sites elsewhere in the module.
-- by :user:devdanzin
Related issues and pull requests on GitHub:
#1382.
Fixed a segmentation fault when extending a multidict with itself
-- by :user:cananoo and :user:asvetlov.
Related issues and pull requests on GitHub:
#1398.
Fixed a memory leak when MultiDict and CIMultiDict instances are
initialized more than once -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1412.
Fixed a reference leak in items-view set operations
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1413.
Fixed a memory leak when destroying C-extension multidict instances
and proxies -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1415.
Fixed a reference leak from repeated __init__ calls on
MultiDictProxy and CIMultiDictProxy instances -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1419.
Fixed missing decref for :data:None on error in :meth:~multidict.MultiDict.setdefault
-- by :user:asvetlov
Related issues and pull requests on GitHub:
#1421.
Added :py:func:reversed support to the keys, values, and items views of
:class:~multidict.MultiDict, :class:~multidict.CIMultiDict, and their
proxies in both the C-extension and pure-Python implementations
-- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#448.
Optimized key identity comparison for C Extension.
Now it uses fast path that is equal to :c:func:PyUnicode_Equal from
Python 3.14+ but without redundant type checks. It gives ~15% speed-up on benchmarks with many key comparisons.
-- by :user:asvetlov
Related issues and pull requests on GitHub:
#1406.
Changed both the pure-Python and C implementations to mark a temporarily
removed hash table entry by setting the high bit of its hash instead of
overwriting it with a sentinel value, so restoring the entry no longer
recomputes the hash; :meth:~multidict.MultiDict.getall is faster ~10% now in C version
-- by :user:asvetlov
Related issues and pull requests on GitHub:
#1426.
Dropped support for Python 3.9 as it has reached end of life.
Related issues and pull requests on GitHub:
#1316.
Dropped support for free-threaded Python 3.13 -- by :user:ngoldbaum.
Related issues and pull requests on GitHub:
#1326.
Fixed CIMultiDictProxy documentation to state it inherits from
MultiDictProxy (not MultiDict), and fixed missing word in
istr section -- by :user:veeceey.
Related issues and pull requests on GitHub:
#1298.
Clarified that istr preserves the original casing and
compares as a regular str; case-insensitive matching is
handled by CIMultiDict -- by :user:gyanu2507.
Related issues and pull requests on GitHub:
#1397.
Fixed broken RST markup in items() docstrings
-- by :user:veeceey.
Related issues and pull requests on GitHub:
#1299.
Dropped support for free-threaded Python 3.13 -- by :user:ngoldbaum.
Related issues and pull requests on GitHub:
#1326.
Wheels for iOS and Android (CPython 3.13+) are now published on
release-tag builds, so downstreams on those platforms no longer need a
local compiler to install multidict; these wheels are build-verified
only, since running the test suite on mobile needs a simulator or an
emulator -- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1337.
Added support for building and shipping riscv64 wheels
-- by :user:justeph.
Related issues and pull requests on GitHub:
#1293.
The setuptools build dependency lower bound has been restored to be
>= 47 after an incorrect automated increase in :pr:1315
-- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1400.
Added support for collecting code coverage of isolated multidict tests
-- by :user:Vizonex.
Related issues and pull requests on GitHub:
#1314.
Switched to mirrors-clang-format and enabled clang-format in pre-commit.ci
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1318.
Added a release-tag-gated CI job that cross-compiles the C extension
for iOS and Android through cibuildwheel -- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1337.
Added an AGENTS.md orientation file at the repository root, covering the
pull request template, the CHANGES/ news fragment conventions, the
draft-PR / human-review workflow, and the dual pure-Python and C-extension
parity rule, so LLM contributors land changes that match project style
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1339.
Added init, installable, instantiation, parametrization,
parametrized, postfix, and unparseable to the docs spelling list
so news fragments and docs can use these words without failing the spell
check build -- by :user:pctablet505.
Related issues and pull requests on GitHub:
#1343.
Documented three common agent failure modes in
:file:AGENTS.md: the docs spell check
(make doc-spelling) catches unknown words in news fragments
before CI does; coverage runs over the test tree too, so
unreachable defensive raise guards and one-sided cleanup
branches in tests will surface as uncovered on the codecov
patch report; and branch creation is restricted on
aio-libs/multidict, so PRs must be pushed from a fork
-- by :user:bdraco.
Trimmed the CI test matrix to drop redundant Py_DEBUG jobs on
macOS and Windows and to gate the windows-11-arm wheel build to
release-time only, cutting CI wall-clock by roughly a third without
reducing code coverage -- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1346.
Added a CLAUDE.md at the repository root that imports
:file:AGENTS.md via Claude Code's @-syntax, so the
project's LLM contributor rules load automatically when
working in Claude Code
-- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1347.
Enabled the I (import sorting) rule in
[tool.ruff.lint] so ruff check covers import order on top
of the existing UP (pyupgrade) group, and updated the
ruff-check pre-commit hook to run with
--fix --exit-non-zero-on-fix --show-fixes so import-order
fixes apply on commit. Sorted imports in four tests/ files
to match the new rule -- by :user:bdraco.
Related issues and pull requests on GitHub:
#1348.
Enabled the PLC0415 (import-outside-top-level) rule in
[tool.ruff.lint]. Function-scoped imports must now opt in
with # noqa: PLC0415 plus a comment explaining the reason
(typically avoiding a heavy optional dependency at import time).
This catches a pattern that LLM contributors frequently introduce.
There are currently no opt-outs in the tree, so the rule is
enforced everywhere ruff check runs
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1349.
Switched the cibuildwheel build frontend to build[uv] so
that uv provisions every build and test virtual environment
in the wheel matrix. Test-dependency installation in particular
drops from a multi-second pip install per ABI to a roughly
sub-second uv resolve
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1350.
Overrode CIBW_BUILD_FRONTEND=build for the odd-arch wheel
matrix; the upstream manylinux/musllinux images for
those arches do not ship uv
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1352.
Allowed re-running the deploy job after a partial release failure: the
Make Release step now skips when the GitHub Release already exists,
and the PyPI publish step uses skip-existing so dists that were
already uploaded on a prior attempt do not break the retry
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1353.
Switched the aarch64 and armv7l wheel builds to GitHub's native ARM
runners. The aarch64 wheels now build without QEMU emulation, and
armv7l runs on aarch64 hosts so its 32-bit ARM execution is far
cheaper than the previous aarch64-on-x86_64 path
-- by :user:bdraco.
Related issues and pull requests on GitHub:
#1354.
Documented design principles for pure-Python :class:~multidict.istr implementation
-- by :user:asvetlov
Related issues and pull requests on GitHub:
#1360.
Pinned coverage to the ctrace measurement core so the test suite runs
on Python 3.14. Coverage picks sysmon there, which cannot record the
dynamic contexts pytest-cov switches at runtime, and the resulting warning
became an error under the suite's filterwarnings setting
-- by :user:rodrigobnogueira.
Related issues and pull requests on GitHub:
#1393.
Dependabot has been restricted to the requirements subdirectory to
avoid unintended updates outside dependency requirement files
-- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1400.
Enabled automatic upgrades from ruff after each python version that is dropped
-- by :user:Vizonex.
Related issues and pull requests on GitHub:
#1325.
Explicitly marked empty_htkeys as const.
Moved the structure to .rodata linker section.
Unexpected modification of the structure will lead to segfault instead
of corrupting the data and fail fast.
.rodata is mount as read-only-mapped, it is shared well across
multiple threads without cache misses.
The change is pretty trivial, it is made for the sake of correctness,
not for speedup.
-- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1405.
Slightly reorganized multidict creating process.
md_init() creates an empty multidict.tp_alloc slot is used in object cloning instead of bare :c:func:PyType_GenericNew call.-- by :user:asvetlov
Related issues and pull requests on GitHub:
#1420.
Dropped dead code, _md_del_at() and _md_del_at_for_upd() never fails
-- by :user:asvetlov
Related issues and pull requests on GitHub:
#1423.
Fixed a missing space in the :exc:ValueError message
raised when constructing a multidict from a sequence
-- by :user:veeceey.
Related issues and pull requests on GitHub:
#1295.
Renamed the benchmark script from benchmarks/becnhmark.py to
benchmarks/benchmark.py so it matches the invocation documented in
:doc:benchmark -- by :user:aiolibsbot.
Related issues and pull requests on GitHub:
#1335.
Fixed slow memory leak caused by identity by adding Py_DECREF to identity value before leaving md_pop_one on success -- by :user: Vizonex .
Fixed slow memory leak caused by identity by adding Py_DECREF to identity value before leaving md_pop_one on success
-- by :user:Vizonex.
Related issues and pull requests on GitHub:
#1284.
(2026-01-25)
Fixed slow memory leak caused by identity by adding Py_DECREF to identity value before leaving md_pop_one on success -- by Vizonex.
Related issues and pull requests on GitHub: 1284.
Updated tests and added CI for CPython 3.14 -- by :user: kumaraditya303 .
Updated tests and added CI for CPython 3.14 -- by :user:kumaraditya303.
Related issues and pull requests on GitHub:
#1235.
Updated tests and added CI for CPython 3.14 -- by :user:kumaraditya303.
Related issues and pull requests on GitHub: #1235.
(2025-10-05)
Updated tests and added CI for CPython 3.14 -- by kumaraditya303.
Related issues and pull requests on GitHub: 1235.
Fixed MutliDict & CIMultiDict memory leak when deleting values or clearing them -- by :user: Vizonex
Fixed MutliDict & CIMultiDict memory leak when deleting values or clearing them
-- by :user:Vizonex
Related issues and pull requests on GitHub:
#1233.
The type preciseness coverage report generated by MyPy <https://mypy-lang.org>__ is now uploaded to Coveralls <https://coveralls.io/github/aio-libs/multidict>__ and
will not be included in the Codecov views <https://app.codecov.io/gh/aio-libs/multidict>__ going forward
-- by :user:webknjaz.
Added memory leak test for popping or deleting attributes from a multidict to prevent future issues or bogus claims.
-- by :user:Vizonex
Related issues and pull requests on GitHub:
#1233.
Fixed MutliDict & CIMultiDict memory leak when deleting values or clearing them
-- by :user:Vizonex
Related issues and pull requests on GitHub: #1233.
The type preciseness coverage report generated by MyPy <https://mypy-lang.org>__ is now uploaded to Coveralls <https://coveralls.io/github/aio-libs/multidict>__ and
will not be included in the Codecov views <https://app.codecov.io/gh/aio-libs/multidict>__ going forward
-- by :user:webknjaz.
Related issues and pull requests on GitHub: #1122, #1231.
Added memory leak test for popping or deleting attributes from a multidict to prevent future issues or bogus claims.
-- by :user:Vizonex
Related issues and pull requests on GitHub: #1233.
Fixed inconsistencies generated by the C implementation of _md_shrink() which might later lead to assertion failures and crash -- by :user: Romain-Gei
Fixed inconsistencies generated by the C implementation of _md_shrink() which might later lead to assertion failures and crash -- by :user:Romain-Geissler-1A.
Related issues and pull requests on GitHub:
#1229.
(2025-06-30)
Fixed inconsistencies generated by the C implementation of _md_shrink() which might later lead to assertion failures and crash -- by Romain-Geissler-1A.
Related issues and pull requests on GitHub: 1229.
Fixed a memory corruption issue in the C implementation of _md_shrink() that could lead to segmentation faults and data loss when items were deleted f
Fixed a memory corruption issue in the C implementation of _md_shrink() that could lead to segmentation faults and data loss when items were deleted from a :class:~multidict.MultiDict. The issue was an edge case in the pointer arithmetic during the compaction phase -- by :user:bdraco.
Fixed format string compilation errors in debug builds on 32-bit platforms by using portable %zd format specifiers for Py_ssize_t values instead of %ld -- by :user:bdraco.
Re-enabled 32-bit Linux wheel builds that were disabled by default in cibuildwheel 3.0.0 -- by :user:bdraco.
Fixed setuptools deprecation warning about the license specification -- by :user: asvetlov .
If :meth:multidict.MultiDict.extend, :meth:multidict.MultiDict.merge, or :meth:multidict.MultiDict.update raises an exception, now the multidict internal state is correctly restored.
Patch by :user:asvetlov.
Related issues and pull requests on GitHub:
#1215.
Fixed setuptools deprecation warning about the license specification -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1216.
Fix compiler warnings and convert them to errors -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1217.
If multidict.MultiDict.extend, multidict.MultiDict.merge, or multidict.MultiDict.update raises an exception, now the multidict internal state is correctly restored. Patch by asvetlov.
Related issues and pull requests on GitHub: #1215.
Fixed setuptools deprecation warning about the license specification -- by asvetlov.
Related issues and pull requests on GitHub: #1216.
Fix compiler warnings and convert them to errors -- by asvetlov.
Related issues and pull requests on GitHub: #1217.
Added :meth: multidict.MultiDict.merge which copies all items from arguments if its key not exist in the dictionary -- by :user: asvetlov .
Added :meth:multidict.MultiDict.merge which copies all items from arguments if its key
not exist in the dictionary -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#292.
Stopped reallocating memory for the internal htkeys_t structure when inserting new items if the
multidict has deleted items and it could be collapsed in-place. Removal of
malloc()/free() improves the performance slightly.
The change affects C implementation only, pure Python code is not changed.
Patch by :user:asvetlov.
Related issues and pull requests on GitHub:
#1200.
C implementation of :class:multidict.MultiDict.getall now is slightly faster if it returns nothing -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1212.
Replaced docstring for :meth:multidict.MultiDict.update to don't use RST/markdown markup.
Related issues and pull requests on GitHub:
#1203.
Improved documentation for :meth:multidict.MultiDict.extend and :meth:multidict.MultiDict.update -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1205.
When building wheels, the source distribution is now passed directly
to the cibuildwheel invocation -- by :user:webknjaz.
Related issues and pull requests on GitHub:
#1199.
Set up PYTHONHASHSEED for benchmarks execution to make measured times stable -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1202.
Added multidict.MultiDict.merge which copies all items from arguments if its key not exist in the dictionary -- by asvetlov.
Related issues and pull requests on GitHub: #292.
Stopped reallocating memory for the internal htkeys_t structure when inserting new items if the multidict has deleted items and it could be collapsed in-place. Removal of malloc()/free() improves the performance slightly.
The change affects C implementation only, pure Python code is not changed.
Patch by asvetlov.
Related issues and pull requests on GitHub: #1200.
C implementation of multidict.MultiDict.getall now is slightly faster if it returns nothing -- by asvetlov.
Related issues and pull requests on GitHub: #1212.
Replaced docstring for multidict.MultiDict.update to don't use RST/markdown markup.
Related issues and pull requests on GitHub: #1203.
Improved documentation for multidict.MultiDict.extend and multidict.MultiDict.update -- by asvetlov.
Related issues and pull requests on GitHub: #1205.
When building wheels, the source distribution is now passed directly to the cibuildwheel invocation -- by webknjaz.
Related issues and pull requests on GitHub: #1199.
Set up PYTHONHASHSEED for benchmarks execution to make measured times stable -- by asvetlov.
Related issues and pull requests on GitHub: #1202.
Fixed a bug in C implementation when multidict is resized and it has deleted slots.
Fixed a bug in C implementation when multidict is resized and it has deleted slots.
The bug was introduced by multidict 6.5.0 release.
Patch by :user:asvetlov.
Related issues and pull requests on GitHub:
#1195.
A pair of code formatters for Python and C have been configured in the pre-commit tool.
Related issues and pull requests on GitHub:
#1123.
Shorted fixture parametrization ids.
For example, test_keys_view_xor[case-insensitive-pure-python-module] becomes test_keys_view_xor[ci-py] -- by :user:asvetlov.
Related issues and pull requests on GitHub:
#1192.
The :file:reusable-cibuildwheel.yml workflow has been refactored to
be more generic and :file:ci-cd.yml now holds all the configuration
toggles -- by :user:webknjaz.
Related issues and pull requests on GitHub:
#1193.
(2025-06-24)
Fixed a bug in C implementation when multidict is resized and it has deleted slots.
The bug was introduced by multidict 6.5.0 release.
Patch by asvetlov.
Related issues and pull requests on GitHub: 1195.
A pair of code formatters for Python and C have been configured in the pre-commit tool.
Related issues and pull requests on GitHub: 1123.
Shorted fixture parametrization ids.
For example, test_keys_view_xor[case-insensitive-pure-python-module] becomes test_keys_view_xor[ci-py] -- by asvetlov.
Related issues and pull requests on GitHub: 1192.
The reusable-cibuildwheel.yml workflow has been refactored to be more generic and ci-cd.yml now holds all the configuration toggles -- by webknjaz.
Related issues and pull requests on GitHub: 1193.
Replace internal implementation from an array of items to hash table. algorithmic complexity for lookups is switched from O(N) to O(1).
Replace internal implementation from an array of items to hash table.
algorithmic complexity for lookups is switched from O(N) to O(1).
The hash table is very similar to :class:dict from CPython but it allows keys duplication.
The benchmark shows 25-50% boost for single lookups, x2-x3 for bulk updates, and x20 for
some multidict view operations. The gain is not for free:
:class:~multidict.MultiDict.add and :class:~multidict.MultiDict.extend are 25-50%
slower now. We consider it as acceptable because the lookup is much more common
operation that addition for the library domain.
Related issues and pull requests on GitHub:
#1128.
Builds have been added for arm64 Windows
wheels and the reusable-build-wheel.yml
template has been modified to allow for
an os value (windows-11-arm) which
does not end with the -latest postfix.
Related issues and pull requests on GitHub:
#1167.
(2025-06-17)
Note
The release was yanked because of 1195, multidict 6.5.1 should be used instead.
Replace internal implementation from an array of items to hash table. algorithmic complexity for lookups is switched from O(N) to O(1).
The hash table is very similar to dict from CPython but it allows keys duplication.
The benchmark shows 25-50% boost for single lookups, x2-x3 for bulk updates, and x20 for some multidict view operations. The gain is not for free: ~multidict.MultiDict.add and ~multidict.MultiDict.extend are 25-50% slower now. We consider it as acceptable because the lookup is much more common operation that addition for the library domain.
Related issues and pull requests on GitHub: 1128.
Builds have been added for arm64 Windows wheels and the reusable-build-wheel.yml template has been modified to allow for an os value (windows-11-arm) which does not end with the -latest postfix.
Related issues and pull requests on GitHub: 1167.
Fixed a segmentation fault when calling meth: multidict.MultiDict.setdefault with a single argument -- by :user: bdraco .
Fixed a segmentation fault when calling :py:meth:multidict.MultiDict.setdefault with a single argument -- by :user:bdraco.
Related issues and pull requests on GitHub:
#1160.
Fixed a segmentation fault when attempting to directly instantiate view objects
(multidict._ItemsView, multidict._KeysView, multidict._ValuesView) -- by :user:bdraco.
View objects now raise a proper :exc:TypeError with the message "cannot create '...' instances directly"
when direct instantiation is attempted.
View objects should only be created through the proper methods: :py:meth:multidict.MultiDict.items,
:py:meth:multidict.MultiDict.keys, and :py:meth:multidict.MultiDict.values.
Related issues and pull requests on GitHub:
#1164.
:class:multidict.MultiDictProxy was refactored to rely only on
:class:multidict.MultiDict public interface and don't touch any implementation
details.
Related issues and pull requests on GitHub:
#1150.
Multidict views were refactored to rely only on
:class:multidict.MultiDict API and don't touch any implementation
details.
Related issues and pull requests on GitHub:
#1152.
Dropped internal _Impl class from pure Python implementation, both pure Python and C
Extension follows the same design internally now.
Related issues and pull requests on GitHub:
#1153.
Fixed a segmentation fault when calling multidict.MultiDict.setdefault with a single argument -- by bdraco.
Related issues and pull requests on GitHub: #1160.
Fixed a segmentation fault when attempting to directly instantiate view objects (multidict._ItemsView, multidict._KeysView, multidict._ValuesView) -- by bdraco.
View objects now raise a proper TypeError with the message "cannot create '...' instances directly" when direct instantiation is attempted.
View objects should only be created through the proper methods: multidict.MultiDict.items, multidict.MultiDict.keys, and multidict.MultiDict.values.
Related issues and pull requests on GitHub: #1164.
multidict.MultiDictProxy was refactored to rely only on multidict.MultiDict public interface and don't touch any implementation details.
Related issues and pull requests on GitHub: #1150.
Multidict views were refactored to rely only on multidict.MultiDict API and don't touch any implementation details.
Related issues and pull requests on GitHub: #1152.
Dropped internal _Impl class from pure Python implementation, both pure Python and C Extension follows the same design internally now.
Related issues and pull requests on GitHub: #1153.
Fixed building the library in debug mode.
Fixed building the library in debug mode.
Related issues and pull requests on GitHub: #1144.
Fixed custom PyType_GetModuleByDef() when non-heap type object was passed.
Related issues and pull requests on GitHub: #1147.
Added the ability to build in debug mode by setting :envvar:MULTIDICT_DEBUG_BUILD in the environment -- by :user:bdraco.
Related issues and pull requests on GitHub: #1145.
Fixed a segmentation fault when creating subclassed class:~multidict.MultiDict objects on Python < 3.11 -- by bdraco.
Fixed a segmentation fault when creating subclassed ~multidict.MultiDict objects on Python < 3.11 -- by bdraco.
The problem first appeared in 6.4.0
Related issues and pull requests on GitHub: #1141.
No change release of 6.4.0 since the attestations failed to upload to GitHub
No change release of 6.4.0 since the attestations failed to upload to GitHub
Fixed a memory leak creating new ~multidict.istr objects -- by bdraco.
Fixed a memory leak creating new ~multidict.istr objects -- by bdraco.
The leak was introduced in 6.3.0
Related issues and pull requests on GitHub: #1133.
Fixed reference counting when calling multidict.MultiDict.update -- by bdraco.
The leak was introduced in 4.4.0
Related issues and pull requests on GitHub: #1135.
Switched C Extension to use heap types and the module state.
Related issues and pull requests on GitHub: #1125.
Started building armv7l wheels -- by bdraco.
Related issues and pull requests on GitHub: #1127.
Resolved a memory leak by ensuring proper reference count decrementation -- by asvetlov and bdraco.
Resolved a memory leak by ensuring proper reference count decrementation -- by :user:asvetlov and :user:bdraco.
Related issues and pull requests on GitHub: #1121.
(2025-04-03)
Resolved a memory leak by ensuring proper reference count decrementation -- by asvetlov and bdraco.
Related issues and pull requests on GitHub: 1121.
Fixed keys not becoming case-insensitive when multidict.CIMultiDict is created by passing in a multidict.MultiDict -- by bdraco.
Fixed keys not becoming case-insensitive when multidict.CIMultiDict is created by passing in a multidict.MultiDict -- by bdraco.
Related issues and pull requests on GitHub: #1112.
Fixed the pure Python version mutating the original multidict.MultiDict when creating a new multidict.CIMultiDict from an existing one when keyword arguments are also passed -- by bdraco.
Related issues and pull requests on GitHub: #1113.
Prevented crashing with a segfault when repr is called for recursive multidicts and their proxies and views.
Related issues and pull requests on GitHub: #1115.
The change additionally disables the deprecated GitHub Annotations integration in Codecov_.
Set operations for KeysView and ItemsView of case-insensitive multidicts and their proxies are processed in case-insensitive manner.
Related issues and pull requests on GitHub: #965.
Rewrote multidict.CIMultiDict and it proxy to always return multidict.istr keys. istr is derived from str, thus the change is backward compatible.
The performance boost is about 15% for some operations for C Extension, pure Python implementation have got a visible (15% - 230%) speedup as well.
Related issues and pull requests on GitHub: #1097.
Fixed a crash when extending a multidict from multidict proxy if C Extensions were used.
Related issues and pull requests on GitHub: #1100.
Implemented a custom parser for METH_FASTCALL | METH_KEYWORDS protocol -- by asvetlov.
The patch re-enables fast call protocol in the multidict C Extension.
Speedup is about 25%-30% for the library benchmarks for Python 3.12+.
Related issues and pull requests on GitHub: #1070.
The C-extension no longer pre-allocates a Python exception object in lookup-related methods of ~multidict.MultiDict when the passed-in key is not found but default value is provided.
Namely, this affects MultiDict.getone(), MultiDict.getall(), MultiDict.get(), MultiDict.pop(), MultiDict.popone(), and MultiDict.popall().
Additionally, the ~multidict.MultiDict comparison with regular dictionaries is now about 60% faster on Python 3.13+ in the fallback-to-default case.
Related issues and pull requests on GitHub: #1078.
Implemented __repr__() for C Extension classes in C.
The speedup is about 2.5 times.
Related issues and pull requests on GitHub: #1081.
Made C version of multidict.istr pickleable.
Related issues and pull requests on GitHub: #1098.
Optimized multidict creation and extending / updating if C Extensions are used.
The speedup is between 25% and 70% depending on the usage scenario.
Related issues and pull requests on GitHub: #1101.
multidict.MultiDict.popitem is changed to remove the latest entry instead of the first.
It gives O(1) amortized complexity.
The standard dict.popitem removes the last entry also.
Related issues and pull requests on GitHub: #1105.
Started running benchmarks for the pure Python implementation in addition to the C implementation -- by bdraco.
Related issues and pull requests on GitHub: #1092.
The the project-wide Codecov_ metric is no longer reported via GitHub Checks API. The combined value is not very useful because one of the sources (MyPy) cannot reach 100% with the current state of the ecosystem. We may want to reconsider in the future. Instead, we now have two separate “runtime coverage” metrics for library code and tests. They are to be kept at 100% at all times. And the “type coverage” metric will remain advisory, at a lower threshold.
The default patch metric check is renamed to “runtime” to better reflect its semantics. This one will also require 100% coverage. Another “typing” patch coverage metric is now reported alongside it. It's considered advisory, just like its project counterpart.
When looking at Codecov_, one will likely want to look at MyPy and pytest flags separately. It is usually best to avoid looking at the PR pages that sometimes display combined coverage incorrectly.
The change additionally disables the deprecated GitHub Annotations integration in Codecov_.
Finally, the badge coloring range now starts at 100%.
-- by webknjaz
Related issues and pull requests on GitHub: #1093.
Synchronized pythoncapi_compat.h with the latest available version.
Related issues and pull requests on GitHub: #1063.
Moved registering ABCs for C Extension classes from C to Python.
Related issues and pull requests on GitHub: #1083.
Refactored the internal pair_list implementation.
Related issues and pull requests on GitHub: #1084.
Implemented views comparison and disjoints in C instead of Python helpers.
The performance boost is about 40%.
Related issues and pull requests on GitHub: #1096.
Removals and backward incompatible breaking changes
Fixed in checks throwing an exception instead of returning False when testing non-strings.
Related issues and pull requests on GitHub: #1045.
Fix a leak when the last accessed module in PyInit__multidict init is not released.
Related issues and pull requests on GitHub: #1061.
Implemented support for the free-threaded build of CPython 3.13 -- by lysnikolaou.
Related issues and pull requests on GitHub: #1015.
Started publishing wheels made for the free-threaded build of CPython 3.13 -- by lysnikolaou.
Related issues and pull requests on GitHub: #1015.
Used stricter typing across the code base, resulting in improved typing accuracy across multidict classes. Funded by an NLnet grant.
Related issues and pull requests on GitHub: #1046.
Covered the unreachable code path in multidict._multidict_base._abc_itemsview_register() with typing -- by skinnyBat.
Related issues and pull requests on GitHub: #928.
Added support for Python 3.13 -- by bdraco.
Related issues and pull requests on GitHub: #1002.
Removed Python 3.7 support -- by bdraco.
Related issues and pull requests on GitHub: #997.
Added tests to have full code coverage of the multidict._multidict_base._viewbaseset_richcmp() function -- by skinnyBat.
Related issues and pull requests on GitHub: #928.
The deprecated ::set-output workflow command has been replaced by the $GITHUB_OUTPUT environment variable in the GitHub Actions CI/CD workflow definition.
Related issues and pull requests on GitHub: #940.
codecov-action has been temporarily downgraded to v3 in the GitHub Actions CI/CD workflow definitions in order to fix uploading coverage to Codecov. See this issue for more details.
Related issues and pull requests on GitHub: #941.
In the GitHub Actions CI/CD workflow definition, the Get pip cache dir step has been fixed for Windows runners by adding shell: bash. See actions/runner#2224 for more details.
Related issues and pull requests on GitHub: #942.
Interpolation of the pip cache keys has been fixed by adding missing $ syntax in the GitHub Actions CI/CD workflow definition.
Related issues and pull requests on GitHub: #943.
Upgraded the C-API macros that have been deprecated in Python 3.9 and later removed in 3.13 -- by iemelyanov.
Related issues and pull requests on GitHub: #862, #864, #868, #898.
Reverted to using the public argument parsing API PyArg_ParseTupleAndKeywords under Python 3.12 -- by charles-dyfis-net and webknjaz.
The effect is that this change prevents build failures with clang 16.9.6 and gcc-14 reported in #926. It also fixes a segmentation fault crash caused by passing keyword arguments to MultiDict.getall() discovered by jonaslb and hroncok while examining the problem.
Related issues and pull requests on GitHub: #862, #909, #926, #929.
Fixed a SystemError: null argument to internal routine error on a MultiDict.items().isdisjoint() call when using C Extensions.
Related issues and pull requests on GitHub: #927.
On the Contributing docs page, a link to the Towncrier philosophy has been fixed.
Related issues and pull requests on GitHub: #911.
Stopped marking all files as installable package data -- by webknjaz.
This change helps setuptools understand that C-headers are not to be installed under lib/python3.{x}/site-packages/.
Related commits on GitHub: 31e1170.
Started publishing pure-python wheels to be installed as a fallback -- by webknjaz.
Related commits on GitHub: 7ba0e72.
Switched from setuptools' legacy backend (setuptools.build_meta:__legacy__) to the modern one (setuptools.build_meta) by actually specifying the the [build-system] build-backend option in pyproject.toml -- by Jackenmen.
Related issues and pull requests on GitHub: #802.
Declared Python 3.12 supported officially in the distribution package metadata -- by hugovk.
Related issues and pull requests on GitHub: #877.
The test framework has been refactored. In the previous state, the circular imports reported in #837 caused the C-extension tests to be skipped.
Now, there is a set of the pytest fixtures that is set up in a parametrized manner allowing to have a consistent way of accessing mirrored multidict implementations across all the tests.
This change also implemented a pair of CLI flags (--c-extensions / --no-c-extensions) that allow to explicitly request deselecting the tests running against the C-extension.
-- by webknjaz.
Related issues and pull requests on GitHub: #98, #837, #915.
Updated the test pins lockfile used in the cibuildwheel test stage -- by hoodmane.
Related issues and pull requests on GitHub: #827.
Added an explicit void for arguments in C-function signatures which addresses the following compiler warning:
warning: a function declaration without a prototype is deprecated in all versions of C [-Wstrict-prototypes]
-- by hoodmane
Related issues and pull requests on GitHub: #828.
An experimental Python 3.13 job now runs in the CI -- webknjaz.
Related issues and pull requests on GitHub: #920.
Added test coverage for the and, or, sub, and xor operators in the multidict/_multidict_base.py module. It also covers NotImplemented and "~typing.Iterable-but-not-~typing.Set" cases there.
-- by a5r0n
Related issues and pull requests on GitHub: #936.
The version of pytest is now capped below 8, when running MyPy against Python 3.7. This pytest release dropped support for said runtime.
Related issues and pull requests on GitHub: #937.
Fixed a type annotations regression introduced in v6.0.2 under Python versions <3.10. It was caused by importing certain types only available in newer versions. (#798)
Declared the official support for Python 3.11 — by mlegner. (#872)
Revert #644, restore type annotations to as-of 5.2.0 version. (#688)
Restored back MultiDict, CIMultiDict, MultiDictProxy, and CIMutiDictProxy generic type arguments; they are parameterized by value type, but the key type is fixed by container class.
MultiDict[int] means MutableMultiMapping[str, int]. The key type of MultiDict is always str, while all str-like keys are accepted by API and converted to str internally.
The same is true for CIMultiDict[int] which means MutableMultiMapping[istr, int]. str-like keys are accepted but converted to istr internally. (#682)
Use METH_FASTCALL where it makes sense.
MultiDict.add() is 2.2 times faster now, CIMultiDict.add() is 1.5 times faster. The same boost is applied to get*(), setdefault(), and pop*() methods. (#681)
Fixed type annotations for keys of multidict mapping classes. (#644)
Support Multidict[int] for pure-python version. __class_getitem__ is already provided by C Extension, making it work with the pure-extension too. (#678)
Dropped Python 3.6 support (#680)
#659
Added support Python 3.10
Started shipping platform-specific wheels with the musl tag targeting typical Alpine Linux runtimes.
Started shipping platform-specific arm64 wheels for Apple Silicon. (#629)
Fixed pure-python implementation that used to raise "Dictionary changed during iteration" error when iterated view (.keys(), .values() or .items()) was created before the dictionary's content change. (#620)
Supported GenericAliases (MultiDict[str]) for Python 3.9+ #553
Synchronized the declared supported Python versions in setup.py with actually supported and tested ones. #552
Provided x86 Windows wheels #550
Provided wheels for aarch64, i686, ppc64le, s390x architectures on Linux as well as x86_64. #500
Provided wheels for Python 3.9. #534
Dropped Python 3.5 support; Python 3.6 is the minimal supported Python version.
#503
Removals and backward incompatible breaking changes
Covered the unreachable code path in
multidict._multidict_base._abc_itemsview_register()
with typing -- by :user:skinnyBat.
Related issues and pull requests on GitHub: #928.
Added support for Python 3.13 -- by :user:bdraco.
Related issues and pull requests on GitHub: #1002.
Removed Python 3.7 support -- by :user:bdraco.
Related issues and pull requests on GitHub: #997.
Added tests to have full code coverage of the
multidict._multidict_base._viewbaseset_richcmp() function
-- by :user:skinnyBat.
Related issues and pull requests on GitHub: #928.
The deprecated <https://hynek.me/til/set-output-deprecation-github-actions/>_
::set-output workflow command has been replaced
by the $GITHUB_OUTPUT environment variable
in the GitHub Actions CI/CD workflow definition.
Related issues and pull requests on GitHub: #940.
codecov-action <https://github.com/codecov/codecov-action>_
has been temporarily downgraded to v3
in the GitHub Actions CI/CD workflow definitions
in order to fix uploading coverage to
Codecov <https://app.codecov.io/gh/aio-libs/multidict>.
See this issue <https://github.com/codecov/codecov-action/issues/1252>
for more details.
Related issues and pull requests on GitHub: #941.
In the GitHub Actions CI/CD workflow definition,
the Get pip cache dir step has been fixed for
Windows runners by adding shell: bash.
See actions/runner#2224 <https://github.com/actions/runner/issues/2224>_
for more details.
Related issues and pull requests on GitHub: #942.
Interpolation of the pip cache keys has been
fixed by adding missing $ syntax
in the GitHub Actions CI/CD workflow definition.
Related issues and pull requests on GitHub: #943.
Upgraded the C-API macros that have been deprecated in Python 3.9 and later removed in 3.13 -- by @iemelyanov💰.
Upgraded the C-API macros that have been deprecated in Python 3.9 and later removed in 3.13 -- by @iemelyanov💰.
Related issues and pull requests on GitHub: #862, #864, #868, #898.
Reverted to using the public argument parsing API PyArg_ParseTupleAndKeywords() under Python 3.12 -- by @charles-dyfis-net💰 and @webknjaz💰.
The effect is that this change prevents build failures with clang 16.9.6 and gcc-14 reported in #926. It also fixes a segmentation fault crash caused by passing keyword arguments to MultiDict.getall() discovered by @jonaslb💰 and @hroncok💰 while examining the problem.
Related issues and pull requests on GitHub: #862, #909, #926, #929.
Fixed a SystemError: null argument to internal routine error on a MultiDict.items().isdisjoint() call when using C Extensions.
Related issues and pull requests on GitHub: #927.
On the Contributing docs age, a link to the Towncrier philosophy has been fixed.
Related issues and pull requests on GitHub: #911.
Stopped marking all files as installable package data -- by @webknjaz💰.
This change helps setuptools understand that C-headers are not to be installed under lib/python3.{x}/site-packages/.
Related commits on GitHub: 31e1170.
Started publishing pure-python wheels to be installed as a fallback -- by @webknjaz💰.
Related commits on GitHub: 7ba0e72.
Switched from setuptools' legacy backend (setuptools.build_meta:__legacy__) to the modern one (setuptools.build_meta) by actually specifying the the [build-system] build-backend option in pyproject.toml -- by @Jackenmen💰.
Related issues and pull requests on GitHub: #802.
Declared Python 3.12 supported officially in the distribution package metadata -- by @hugovk💰.
Related issues and pull requests on GitHub: #877.
The test framework has been refactored. In the previous state, the circular imports reported in #837 caused the C-extension tests to be skipped.
Now, there is a set of the pytest fixtures that is set up in a parametrized manner allowing to have a consistent way of accessing mirrored multidict implementations across all the tests.
This change also implemented a pair of CLI flags (--c-extensions / --no-c-extensions) that allow to explicitly request deselecting the tests running against the C-extension.
-- by @webknjaz💰
Related issues and pull requests on GitHub: #98, #837, #915.
Updated the test pins lockfile used in the cibuildwheel test stage -- by @hoodmane💰.
Related issues and pull requests on GitHub: #827.
Added an explicit void for arguments in C-function signatures which addresses the following compiler warning:
warning: a function declaration without a prototype is deprecated in all versions of C [-Wstrict-prototypes]
-- by @hoodmane💰
Related issues and pull requests on GitHub: #828.
An experimental Python 3.13 job now runs in the CI -- @webknjaz💰.
Related issues and pull requests on GitHub: #920.
Added test coverage for the bitwise and, bitwise or, sub, and xor operators in the multidict/_multidict_base.py module. It also covers NotImplemented and "Iterable-but-not-Set" cases there.
-- by @a5r0n💰
Related issues and pull requests on GitHub: #936.
The version of pytest is now capped below 8, when running MyPy against Python 3.7. This pytest release dropped support for said runtime.
Related issues and pull requests on GitHub: #937.
Upgraded the C-API macros that have been deprecated in Python 3.9 and later removed in 3.13 -- by iemelyanov.
Related issues and pull requests on GitHub: 862, 864, 868, 898.
Reverted to using the public argument parsing API PyArg_ParseTupleAndKeywords under Python 3.12 -- by charles-dyfis-net and webknjaz.
The effect is that this change prevents build failures with clang 16.9.6 and gcc-14 reported in 926. It also fixes a segmentation fault crash caused by passing keyword arguments to MultiDict.getall() discovered by jonaslb and hroncok while examining the problem.
Related issues and pull requests on GitHub: 862, 909, 926, 929.
Fixed a SystemError: null argument to internal routine error on a MultiDict.items().isdisjoint() call when using C Extensions.
Related issues and pull requests on GitHub: 927.
On the Contributing docs page, a link to the Towncrier philosophy has been fixed.
Related issues and pull requests on GitHub: 911.
Stopped marking all files as installable package data -- by webknjaz.
This change helps setuptools understand that C-headers are not to be installed under lib/python3.{x}/site-packages/.
Related commits on GitHub: 31e1170.
Started publishing pure-python wheels to be installed as a fallback -- by webknjaz.
Related commits on GitHub: 7ba0e72.
Switched from setuptools' legacy backend (setuptools.build_meta:__legacy__) to the modern one (setuptools.build_meta) by actually specifying the the [build-system] build-backend option in pyproject.toml -- by Jackenmen.
Related issues and pull requests on GitHub: 802.
Declared Python 3.12 supported officially in the distribution package metadata -- by hugovk.
Related issues and pull requests on GitHub: 877.
The test framework has been refactored. In the previous state, the circular imports reported in 837 caused the C-extension tests to be skipped.
Now, there is a set of the pytest fixtures that is set up in a parametrized manner allowing to have a consistent way of accessing mirrored multidict implementations across all the tests.
This change also implemented a pair of CLI flags (--c-extensions / --no-c-extensions) that allow to explicitly request deselecting the tests running against the C-extension.
-- by webknjaz.
Related issues and pull requests on GitHub: 98, 837, 915.
Updated the test pins lockfile used in the cibuildwheel test stage -- by hoodmane.
Related issues and pull requests on GitHub: 827.
Added an explicit void for arguments in C-function signatures which addresses the following compiler warning:
warning: a function declaration without a prototype is deprecated in all versions of C [-Wstrict-prototypes]
-- by hoodmane
Related issues and pull requests on GitHub: 828.
An experimental Python 3.13 job now runs in the CI -- webknjaz.
Related issues and pull requests on GitHub: 920.
Added test coverage for the and, or, sub, and xor operators in the multidict/_multidict_base.py module. It also covers NotImplemented and "~typing.Iterable-but-not-~typing.Set" cases there.
-- by a5r0n
Related issues and pull requests on GitHub: 936.
The version of pytest is now capped below 8, when running MyPy against Python 3.7. This pytest release dropped support for said runtime.
Related issues and pull requests on GitHub: 937.
Fixed a type annotations regression introduced in v6.0.2 under Python versions <3.10. It was caused by importing certain types only available in newer
Declared the official support for Python 3.11 — by mlegner. ((#872))
mlegner. ((#872))Revert (#644), restore type annotations to as-of 5.2.0 version. ((#688))
Restored back MultiDict, CIMultiDict, MultiDictProxy, and CIMutiDictProxy generic type arguments; they are parameterized by value type, but the key ty
Restored back MultiDict, CIMultiDict, MultiDictProxy, and
CIMutiDictProxy generic type arguments; they are parameterized by value type, but the
key type is fixed by container class.
MultiDict[int] means MutableMultiMapping[str, int]. The key type of
MultiDict is always str, while all str-like keys are accepted by API and
converted to str internally.
The same is true for CIMultiDict[int] which means MutableMultiMapping[istr, int]. str-like keys are accepted but converted to istr internally. ((#682))
Use METH_FASTCALL where it makes sense.
Use METH_FASTCALL where it makes sense.
MultiDict.add() is 2.2 times faster now, CIMultiDict.add() is 1.5 times faster.
The same boost is applied to get*(), setdefault(), and pop*() methods. ((#681))
__class_getitem__ is already provided by C Extension, making it work with the pure-extension too. ((#678))Started shipping platform-specific wheels with the musl tag targeting typical Alpine Linux runtimes.
musl tag targeting typical Alpine Linux runtimes..keys(), .values() or .items()) was created before the dictionary's content change. ((#620)_)Started shipping platform-specific wheels with the musl tag targeting typical Alpine Linux runtimes.
musl tag targeting typical Alpine Linux runtimes..keys(), .values() or .items()) was created before the dictionary's content change. ((#620)_)Nothing published for this version
Started shipping platform-specific wheels with the musl tag targeting typical Alpine Linux runtimes.
musl tag targeting typical Alpine Linux runtimes..keys(), .values() or .items()) was created before the dictionary's content change. ((#620)_)Nothing published for this version
Nothing published for this version
Nothing published for this version
Support GenericAliases (MultiDict[str]) for Python 3.9+ #553 _
GenericAliases (MultiDict[str]) for Python 3.9+
#553 <https://github.com/aio-libs/multidict/issues/553>_setup.py with actually supported and tested ones.
#552 <https://github.com/aio-libs/multidict/issues/552>_Supported GenericAliases (MultiDict[str]) for Python 3.9+ 553
Synchronized the declared supported Python versions in setup.py with actually supported and tested ones. 552
Fox a bug in release process, upload x86 Windows wheels
Fox a bug in release process, upload x86 Windows wheels
Provide x86 Windows wheels #550 _
#550 <https://github.com/aio-libs/multidict/issues/550>_Provide wheels for aarch64, i686, ppc64le, s390x architectures on Linux as well as x86_64. #500 _
aarch64, i686, ppc64le, s390x architectures on Linux
as well as x86_64.
#500 <https://github.com/aio-libs/multidict/issues/500>_#534 <https://github.com/aio-libs/multidict/issues/534>_#503 <https://github.com/aio-libs/multidict/issues/503>_Provided wheels for aarch64, i686, ppc64le, s390x architectures on Linux as well as x86_64. 500
Provided wheels for Python 3.9. 534
Dropped Python 3.5 support; Python 3.6 is the minimal supported Python version.
503
- Fixed an issue with some versions of the wheel dist failing because of being unable to detect the license file. #481
Fixed an issue with some versions of the wheel dist failing because of being unable to detect the license file. #481
- Fixed creating and updating of MultiDict from a sequence of pairs and keyword arguments. Previously passing a list argument modified it inplace, and
Fixed creating and updating of MultiDict from a sequence of pairs and keyword arguments. Previously passing a list argument modified it inplace, and other sequences caused an error. #457
Fixed comparing with mapping: an exception raised in the len() method caused raising a SyntaxError. #459
Fixed comparing with mapping: all exceptions raised in the getitem() method were silenced. #460
MultiDict.iter fix memory leak when used iterator over multidict instance. #452 _
MultiDict.iter fix memory leak when used iterator over multidict instance.
#452 <https://github.com/aio-libs/multidict/issues/452>_- Implement __sizeof__ function to correctly calculate all internal structures size. #444
Implement sizeof function to correctly calculate all internal structures size. #444
Expose getversion() function. #451
Fix crashes in popone / popall when default is returned. #450
Corrected the documentation for MultiDict.extend() #446
- Fixed crashing when multidict is used pyinstaller #432
Fixed crashing when multidict is used pyinstaller #432
Fixed typing for CIMultiDict.copy() #434
Fixed a memory leak in MultiDict.copy() #443
- CIMultiDictProxy.copy() fixed to return a case-insensitive multidict._multidict.CIMultiDict() object. #427
CIMultiDictProxy.copy() fixed to return a case-insensitive multidict._multidict.CIMultiDict() object. #427
Made CIMultiDict subclassable again. #416
Fixed a regression, multidict.MultiDict can be constructed from an arbitrary iterable of pairs again. #418
CIMultiDict.add() may be called with keyword arguments now. #421
Mention MULTIDICT_NO_EXTENSIONS environment variable in docs. #393
Document the fact that istr preserves the casing of argument untouched but uses internal lower-cased copy for keys comparison. #419
- Replace Cython optimization with pure C #249
Replace Cython optimization with pure C #249
Implement length_hint() for iterators #310
Support the MultiDict[str] generic specialization in the runtime. #392
Embed pair_list_t structure into MultiDict Python object #395
Embed multidict pairs for small dictionaries to amortize the memory usage. #396
Support weak references to C Extension classes. #399
Add docstrings to provided classes. #400
Merge multidict._istr back with multidict._multidict . #409
Explicitly call tp_free slot on deallocation. #407
Return class from class_getitem to simplify subclassing #413
- Fix PyPI link for GitHub Issues badge. #391
Fix PyPI link for GitHub Issues badge. #391
- Preserve the case of istr strings. #374
Fix GC object tracking. #314
Preserve the case of istr strings. #374
Generate binary wheels for Python 3.8.
Nothing published for this version
Nothing published for this version
Fixes another memory leak introduced by 4.5.0 release.
Fixes another memory leak introduced by 4.5.0 release.
aiohttp client API leaked memory with multidict 4.5.0-4.5.1, the release fixes it.
The release fixes a very serious memory leak introduced by 4.5.0 release.
The release fixes a very serious memory leak introduced by 4.5.0 release.
Upgrade from 4.5.0 to 4.5.1 is highly recommended.
Multidict views ported from Cython to C extension
Changes ------------- * Restore Python 3.4 support
Allow to install the library on systems without compilation toolset
Fix type annotations (#283)
Allow to install the library on systems without compilation toolset (#281)
- Rewrite C implementation to use C pair list.
Rewrite C implementation to use C pair list.
Fix update order when both arg and kwargs are used.
One column per quarter.