Repository navigation
Conversation
Docutils 1.0 represents native doctest blocks as literal_block nodes with the pycon and doctest classes, and stores colspec colwidth values as strings. Recognise the new doctest node when moving doctest blocks out of block quotes, assigning a highlight language, marking translatable nodes and collecting doctests. Convert colwidth values to integers in the text, LaTeX and Texinfo writers, and create string colwidths in autosummary tables on Docutils 0.22 and later.
sylvesterkaczmarek
left a comment
sylvesterkaczmarek
left a comment
There was a problem hiding this comment.
I think this loses valid proportional widths. Docutils’ colspec.propwidth() can return a float, but this helper truncates it, so 2.5* becomes 2 and 0.5* becomes 0. That changes table ratios and can create zero-width columns. Could we preserve the proportion, or scale fractional values to integers, instead of int()-truncating it?
…ctest-colwidth-14680
Docutils 1.0 allows a colspec's proportional width to be fractional. Truncating with int() distorted the ratios between columns and turned any width below 1 into a zero-width column. Read the width exactly and scale every colspec in the tgroup by the lowest common denominator of the table, so the proportions are preserved. Tables that already use integer widths are unaffected.
sylvesterkaczmarek
left a comment
sylvesterkaczmarek
left a comment
There was a problem hiding this comment.
Rechecked current f3d1d2c. Fractional colwidths are now parsed exactly and scaled with a common denominator across sibling columns, so values such as 2.5*, 0.5* and unequal fractional widths keep their proportions instead of truncating or collapsing to zero. The added regression matrix covers the cases I raised. The two remaining red matrix jobs fail in an unrelated Python-domain rendering test, not the table-width code. My concern is resolved.
Master (d19cbb5) now handles Docutils 1.0 doctest blocks and string colwidths itself, but converts colwidth with int(), which raises on fractional or starred widths such as '2.5*'. Keep master's doctest handling and use _get_colwidth() in the LaTeX, Texinfo and text writers.
sylvesterkaczmarek
left a comment
sylvesterkaczmarek
left a comment
There was a problem hiding this comment.
Thanks for refreshing this against current master, Tejas. I checked the latest diff at 050c93e: the shared scaling still preserves the tested fractional and starred column-width ratios, and the Docutils/platform CI matrix is green. My earlier concern remains resolved. Happy to approve the refreshed head.
…ctest-colwidth-14680 # Conflicts: # CHANGES.rst
|
Thanks for checking the refreshed head again, @sylvesterkaczmarek. CI is green on 30d111f. @AA-Turner, could you take a look when you have time? |
The new Docutils 1.0 changes how some document information is stored.
Doctest blocks now come in as a different type of node.
Table column widths are also stored as text instead of numbers.
Because of this, Sphinx could no longer find some doctest blocks correctly.
Tables also started causing problems in text, LaTeX, and Texinfo output.
The fix allows Sphinx to understand both the old and new doctest node types.
It also converts the table column widths from text back into numbers.
This keeps Sphinx working correctly with the new Docutils version.
The changes were tested with both Docutils 0.22 and 1.0.
Claude Code wrote the code, tests, and changelog; I reviewed everything, wrote the description, and submitted the change.