xlformat

xlformat applies an Excel number format to a value and returns the text Excel displays. It is pure Python with no runtime dependencies, and its output is tested against text captured from Excel 16 for Mac.

>>> from xlformat import format_value
>>> format_value(1234.5, "#,##0.00")
'1,234.50'

Documentation: https://xlformat.readthedocs.io.

Installation

pip install xlformat

Python 3.11 or later.

Usage

format_value returns the displayed text as a string.

>>> format_value(-1234.5, "#,##0.00;(#,##0.00)")
'(1,234.50)'
>>> format_value(44927.75, "yyyy-mm-dd hh:mm")
'2023-01-01 18:00'
>>> format_value(0.125, "# ?/?")
' 1/8'
>>> import datetime
>>> format_value(datetime.date(2023, 1, 1), "dddd, mmmm d, yyyy")
'Sunday, January 1, 2023'

Values may be int, float, Decimal, bool, str, None, date, datetime, time or timedelta. Pass date1904=True for workbooks that use the 1904 date system.

format_cell takes the same arguments and returns a FormattedCell with what Excel shows around the text: the section color, the default alignment, the position of a * fill, whether the value overflowed, and the text without any width applied.

>>> from xlformat import format_cell
>>> cell = format_cell(-1234.5, "[Red]0.00;[Blue]-0.00")
>>> cell.text, cell.color, cell.align
('-1234.50', 'Blue', 'right')

xlformat.DEFAULT_PALETTE maps color names such as Blue or Color12 to RRGGBB.

NumberFormat.parse compiles a format string once. Parsing is cached, and the compiled object has format and format_cell methods and facts about the pattern such as is_date. An invalid format string raises FormatError, but only when Excel itself would refuse to open a file containing it. Callers reading files usually catch it and fall back to General, which is what Excel's repair does.

Width

Without a width, the result is what copy and paste from Excel gives: * fills add nothing, _x pads add one space, and a value that cannot be shown is 255 # characters. Two parameters add a width, and they cannot be combined.

• width is a number of characters. It suits terminals and plain text. * fills repeat to the width, a number or date that does not fit becomes width # characters, General drops digits before it overflows, and text never overflows.
• column_width is the number from Excel's Column Width dialog, such as 8.43. The text is measured with the glyph widths of font, so the result matches the sheet character for character, including how many # and fill characters fit. font is a name or a (name, size) pair and defaults to ("Calibri", 11), the font openpyxl writes.

>>> format_value(1234567.891, "General", width=7)
'1234568'
>>> format_value(1000, "#,##0.0", width=6)
'######'
>>> format_value(1000, "#,##0.0", column_width=6)
'1,000.0'
>>> format_value(1234.5, "0.00*-", column_width=12, font=("Calibri", 14))
'1234.50---------'

The two modes differ on purpose. In Excel a comma or a point is narrower than a digit, so 1,000.0 fits a column 6 wide, while in character mode it is seven characters. A measure callable can replace the character count when neither mode fits, and width is then in its units.

Quirks

Three Excel behaviors are reproduced by default and can be switched off with quirks=Quirks.STRICT or by passing Quirks with individual fields set. QUIRKS.md has the evidence for each.

Scope and known differences

• Excel mode measures Calibri only, at 9, 10, 11, 12, 14 and 16 points. Other Calibri sizes are scaled from the nearest measured size and may differ from Excel by a character near the edge of a column. Any other font raises LookupError. Aptos Narrow, the Windows default, needs a capture on Windows.
• Character mode is its own contract. A # run is always width characters, and a value is judged by its character count, where Excel would fit or overflow some texts by glyph width.
• Expected outputs come from Excel 16 for Mac with the en-US locale. Month and day names are English; locale tags such as [$-409] are parsed but not applied. Windows Excel renders text with a different engine and may disagree on a borderline fit in Excel mode.
• Error values such as #N/A are not inputs. Excel displays them regardless of the format, so a caller shows them directly.
• Conditional formatting and cell styles beyond the format string are out of scope.

Every captured row passes; no known deviations remain. Fifteen color readings in the Excel captures are skipped because Excel's scripting interface misreports them; the sheet itself shows the color the package returns.

Related packages

xlformat is one of three packages from the sheetmill project, with sheetpeek, which renders workbooks to HTML, Markdown and terminal output, and xlformula, which evaluates formulas. Each installs and works on its own. A sheetmill meta-package that installs all three is planned.

Development

With uv, which installs the exact versions pinned in uv.lock:

uv sync --group docs
uv run pytest                 # unit tests and every captured Excel row, a few seconds
uv run pytest --tier=fast     # a per-format sample of the Excel captures
uv run ruff check . && uv run ruff format --check . && uv run mypy
uv run --group docs sphinx-build -n -W docs docs/_build/html
npx --yes pagefind@1.5.2 --site docs/_build/html   # search index for served copies

With pip 25.1 or later, which ignores uv.lock and installs the latest versions the dependency groups allow:

python -m venv .venv && source .venv/bin/activate
pip install -e . --group dev --group docs
pytest && ruff check . && ruff format --check . && mypy
sphinx-build -n -W docs docs/_build/html

CONTRIBUTING.md lists the checks a change must pass and how commits are written.

The documentation's search page uses Pagefind when the site is served over http or https, as on Read the Docs or with python -m http.server -d docs/_build/html. A copy opened straight from disk uses Sphinx's built-in search instead, because browsers do not let a file:// page fetch Pagefind's index.

The expected outputs under tests/captures/ were captured from Excel for Mac by the scripts in capture/: each suite is written as a workbook with openpyxl, opened in Excel through AppleScript, and read back as displayed text at each column width. Regenerating a suite needs Excel for Mac and is described in CAPTURING.md:

uv run python -m capture numeric_core

Most of the code is written with Claude Code, reviewed before each commit and tested against output captured from Excel; see CONTRIBUTING.md.

License

MIT.