avadim/fast-excel-reader

Fixed

• The cached string result of a formula is decoded like any other string. A calculated cell keeps
  its result in the sheet under the type str, and version 4.4.1 taught the reader to turn
  _xHHHH_ sequences back into the characters they stand for - but only for shared and inline
  strings, so formula results kept coming back escaped. A concatenation carrying a line break,
  =B1&C1 over cells holding one, returned the literal text _x000D_ instead of a CR. The type
  such a cell reports is unchanged: a numeric formula result is still cast to int or float, and
  the cell metadata still says str.

Fixed

• A number format the file declares itself is no longer mistaken for the builtin format its
  index stands for. The format numbers 14-22, 27-36, 45-47, 50-58 and 71-81 are reserved for
  builtin date formats, several of them localized, so a cell using one carries no format code
  of its own. But a file may register a code for such a number - XLS writers do it routinely,
  1C among them, and a workbook converted from XLS keeps doing it in XLSX - and it has then
  redefined that number. The reader typed such cells by the number alone, so an article number
  like 043, stored as 43 under the custom pattern 000, came back as a timestamp of
  1900-02-12 with no error raised. The declared pattern now decides, for the cell value as well
  as for format-category of the cell style; a number used without a declaration still means
  the builtin format it always did.

Fixed

• Characters stored as _xHHHH_ are decoded back to the character they stand for. XML cannot carry control characters (and normalizes CR), so a spreadsheet keeps them escaped — a cell holding a CR is stored as _x000D_, and Excel as well as the sibling fast-excel-writer writes it that way. The reader used to return the literal text _x000D_, so such a value did not survive a write/read round trip. Text that literally reads _xHHHH_ is stored as _x005F_xHHHH_ and still comes back unchanged; both forms are resolved in a single pass, so a decoded value is never decoded twice. Requires avadim/fast-excel-helper 1.4 or above.

Added

• Opening a workbook from a string or a stream. Excel::openString($content) reads a workbook
  held in memory (a database BLOB, an HTTP body, an S3/Flysystem read), and Excel::openStream($stream)
  reads it from an open stream resource (a URL via fopen(), php://memory, a Flysystem read stream).
  Both pick the reader by signature exactly like open(), so a string/stream carrying XLSX, XLS or CSV
  reads back identically to the same file on disk, and both accept the same $options as open().
  The content is copied to a temporary file (honouring Excel::setTempDir()), which is removed on
  script shutdown; the stream passed to openStream() is not closed. See
  docs/10-getting-started.md.
• Reading cell hyperlinks. Sheet::getHyperlinks() returns a sheet's hyperlinks keyed by the
  cell (or range) they are anchored to, each as ['link' => ..., 'location' => ..., 'display' => ..., 'tooltip' => ...]. External targets are resolved from the sheet relationships file; internal links
  expose their in-document location. A sheet without hyperlinks returns an empty array. See
  docs/16-sheet-metadata.md.
• Reading workbook document properties. Excel::getProperties() returns the core
  (docProps/core.xml) and application (docProps/app.xml) properties as one associative array with
  normalised keys — creator, lastModifiedBy, created, modified, title, application,
  company, ... Only the properties present in the file are returned; a workbook without a docProps
  part returns an empty array. See docs/12-advanced-reading.md.

Changed

• An empty (zero-byte) file passed to Excel::open() now fails up front with a clear "File ... is empty" error, instead of being silently opened as a zero-row CSV — open() guesses the format by signature and an empty file matches none. Excel::openCsv() is unchanged and still reads an empty file as a valid, zero-row CSV.

Added

• Reading CSV through Excel::open() — returns a one-sheet Csv\CsvBook exposing the same reading API as XLSX/XLS (key modes, read areas, withHeader(), readColumns(), ...). Format is chosen by signature (OLE2 → XLS, ZIP → XLSX, otherwise CSV); pass Excel::open($file, $options) to configure the CSV reader or ['format' => 'csv'] to force it. openCsv() is unchanged and still returns the low-level Csv\CsvReader (also via CsvBook::getReader()). CSV has no styles/dates/merged cells/images — those accessors return empty results. See docs/20-csv.md.
• Excel::isXlsx() — ZIP-signature check (PK\x03\x04).
• CSV binary guard — binary input (a NUL byte, or a high share of control characters) is rejected up front with a clear "file appears to be binary" error; UTF-16/UTF-32 text is exempt. New allow_binary option (CsvOptions::setAllowBinary()) turns it off.

Changed

• Excel::open(string $file, $options = []) takes an optional second argument (BC-safe); a file that is neither OLE2 nor ZIP is now read as CSV instead of being pushed into the XLSX reader.
• A ZIP that is not an XLSX (a DOCX/PPTX or a plain archive) now fails with "Not an XLSX workbook: the ZIP archive has no xl/workbook.xml" (with a DOCX/PPTX hint) instead of a cryptic "Internal file not found".

Fixed

• CSV: an empty file (or one whose length is a multiple of the read buffer) no longer emits a PHP warning and a phantom trailing row.
• CSV: CsvReader::rewind() now fully resets the read state, so re-reading after a partial pass no longer replays a stale buffer.

Full bilingual changelog: CHANGELOG.md · CHANGELOG.ru.md

Fixed

• Built-in date formats (numFmtId 14-22, such as the "short date" code 14) are now rendered
  deterministically. Previously the reader overwrote their patterns from the ambient ICU locale
  whenever ext-intl was loaded, so the same file produced different strings depending on the
  server locale and on whether the extension was installed (#53).
  These codes now resolve to fixed patterns regardless of environment. Formats that a file spells
  out explicitly are unaffected.

Added

• useLocaleFormats(?string $locale = null) — opt in to locale-dependent rendering of the built-in
  date codes (the previous behaviour, now explicit). Pass a locale for reproducible output, or nothing
  to use the process default locale. Requires ext-intl.

Full changelog: CHANGELOG.md · по-русски

Fixed

• XLS now reports formula text with the leading =, like XLSX always has. The f field of a formula
  cell has the same shape whichever format the file came from, so code no longer needs to special-case
  XLS. Formulas whose text cannot be recovered still report null, not "=".

Full changelog: CHANGELOG.md · по-русски

Added

• Reading of legacy XLS workbooks (Excel 97-2003, BIFF8). Excel::open() now chooses the reader
  from the file signature, so XLSX and XLS are opened the same way; the file extension is never
  consulted. Excel::openXls() opens a workbook explicitly and Excel::isXls() exposes the test.
  See docs/21-xls.md.
  • Values and types, with dates detected through number formats
  • Cell styles: fonts, fills, borders, alignment, number formats and palette colours
  • Formula text, including shared formulas
  • Embedded images
  • Multiple worksheets, hidden and very hidden sheets, merged cells, sheet dimensions
  • Encrypted workbooks and BIFF5/BIFF7 files are rejected with a message saying which
• AbstractBook and AbstractSheet, holding the format-independent half of the readers: read areas,
  key modes, result-mode flags, the row generator and every read* helper. XLSX and XLS share one
  implementation of the whole public reading API.
• withHeader() now accepts an optional list of column names: withHeader(['name', 'birthday']). The
  header row is still skipped, but the names come from the list instead of from its values. Names are
  positional - the first name goes to the first column of the read area - so no column letters are
  involved and the same call works on a sheet whose data does not start at A1. A shorter list renames
  only the columns it covers. Supported for XLSX, XLS and CSV. Calling it with no argument is
  unchanged. This mirrors the naming of writeHeader() in the sibling fast-excel-writer.

Fixed

• Sheet::readCellsWithStylesFrom() returned bare cell values instead of values with styles: it called
  readCells() rather than readCellsWithStyles(), and passed the style key into a bool parameter.
• Sheet::readCellsWithStyles($styleKey) never narrowed the result to the requested property. The key
  was looked up on the nested style, where properties sit inside their group, so 'fill-color' - the
  example in the method's own docblock - always returned the complete style instead.
• Complete cell styles - getCompleteStyleByIdx(), readCellsWithStyles() and everything built on
  them - died with Call to undefined method DOMText::getAttribute() on workbooks whose styles.xml
  is written with indentation, which several writers do.

Changed

• Reading XLSX is about 1.5 times faster. Values, types and peak memory are unchanged.
• The accessors that returned a concrete Sheet now return AbstractSheet, and the fluent setters on
  the workbook return AbstractBook. The objects handed back are unchanged, and a subclass may still
  narrow the return type back, so this only affects explicit type declarations in calling code.

Full changelog: CHANGELOG.md · по-русски

Bug fixes

Four defects in the sheet reading path, all in methods that had no test coverage.

• Sheet::readFirstRowCellsFrom() always threw a TypeError. It forwarded $columnKeys into readFirstRowCells(?bool $styleIdxInclude), so even a default call failed. Because the result is keyed by cell address, column keys cannot apply at all — renaming a column would corrupt the address — so the parameter was dropped, mirroring readCellsFrom(). The new signature is readFirstRowCellsFrom(string $areaRange, ?bool $styleIdxInclude = null). No working call can break, since every call to the old signature threw.
• Restricting columns while requesting numeric column keys returned only nulls. With a read area in place the row template was keyed by column letter while values were stored under numeric keys, and the values were then filtered out. Affected setReadArea() and setReadAreaColumns() combined with KEYS_COL_ZERO_BASED / KEYS_COL_ONE_BASED (and therefore KEYS_ZERO_BASED / KEYS_ONE_BASED). An explicit column name still takes precedence over the numeric key.
• Sheet::rewind() discarded its $columnKeys argument, although it is documented as an alias of reset().
• Sheet::firstCol() ignored the column bounds of the read area, reporting the first cell of the row as stored in the file. firstRow() was unaffected.

Tests

The suite grows from 84 to 262 tests (357 to 757 assertions), added ahead of the fixes as a regression net:

• 105 characterization snapshots covering every read* method, the full KEYS_* matrix, result-mode flags, read areas, styles, dates, metadata and degenerate inputs, compared as whole arrays with assertSame.
• Dedicated tests for generator semantics, read areas, result modes, merged cells, workbook delegation and streaming memory behaviour.
• RESULT_MODE_ROW, TRIM_STRINGS, TREAT_EMPTY_STRING_AS_EMPTY_CELL and KEYS_RELATIVE had no coverage at all before and are now tested.

Method coverage: Sheet 33.9% → 58.9%, Excel 55.6% → 69.4%.

Notes

• Reading XLSX files that use namespace-prefixed tags (<x:row>, <x:c>) still returns an empty result. This is a pre-existing limitation, now documented by a test rather than silently unnoticed.
• API reference under docs/ regenerated.

Full changelog: v3.1.0...v3.2.0

Add stat() to Excel and Sheet, refactor countActualDimension() to str…

…eaming

- Sheet::stat() returns rows/cols/cell counts in a single streaming pass,
cached in actualRows/actualCols/cellStat
- Excel::stat() aggregates per-sheet stats with workbook totals
- Rewrite countActualDimension() to a memory-bounded streaming scan that
glues tags across block boundaries and finds the last row via a tail window
- Document the full-scan cost in related methods' PHPDoc
- Add XlsxStatTest covering sheet/workbook stat and consistency with readCells