fix(text_region): preserve font state across render, closes #1804

[Pawansingh3889]

What changed

Fixes #1804 using @andersonhc's narrowed repro from the issue thread.

When a text_columns() / text_region() context exits, ParagraphCollectorMixin.__exit__ calls self.render() after popping the local graphics-state stack. Inside render(), each paragraph temporarily adjusts font_family, font_style, font_size_pt, and current_font to match its own paragraph-level font settings. Because those mutations happen on the already-restored outer state, the last paragraph's font settings leak back onto the calling FPDF instance.

The narrowed minimal repro (from your comment):

pdf = FPDF()
pdf.add_page()
pdf.set_font("Helvetica", size=12)
cols = pdf.text_columns()
with cols:
    pdf.set_font("Helvetica", size=24)
    with cols.paragraph() as p:
        p.write("Large heading")
    pdf.set_font("Helvetica", size=10)
    with cols.paragraph() as p:
        p.write("Small body text")
pdf.ln()
pdf.cell(text=f"after columns size is {pdf.font_size_pt}")  # was 24, should be 12

Fix

ParagraphCollectorMixin.__exit__ now saves the four font attributes before self.render() and restores them in a try / finally:

saved_font_family = self.pdf.font_family
saved_font_style = self.pdf.font_style
saved_font_size_pt = self.pdf.font_size_pt
saved_current_font = self.pdf.current_font
try:
    self.render()
finally:
    self.pdf.font_family = saved_font_family
    self.pdf.font_style = saved_font_style
    self.pdf.font_size_pt = saved_font_size_pt
    self.pdf.current_font = saved_current_font

Cursor position and other render-produced state are intentionally left alone, since subsequent draws rely on them advancing. I also considered wrapping render() in a full _push_local_stack / _pop_local_stack, but that over-reverted the cursor state and broke seven existing tests. The narrower font-only save/restore keeps every existing snapshot byte-identical.

Test

New test_tcols_font_size_does_not_leak in test/text_region/test_text_columns.py uses the narrowed repro and asserts pdf.font_size_pt is unchanged after the context exits. Passes with the fix, fails without it (24 != 12).

The rest of the test/text_region/ suite passes identically against the fixed branch and against master on my machine. Two tests (test_text_columns_with_text_shaping, test_text_columns_with_shorter_2nd_column) require uharfbuzz which I couldn't install locally — they behave the same way with and without the fix, so not a regression, but I'd appreciate a CI run to confirm.

CHANGELOG

Entry added under ## [2.8.8] - Not released yet → ### Fixed.

Closes #1804.

[Pawansingh3889]

The one failing job is test (3.10, windows-latest) on test_perfs.py::test_intense_image_rendering — 12.30 s against a 12 s threshold, a Windows-runner timing flake. The source change in this PR only touches ParagraphCollectorMixin.__exit__ in fpdf/text_region.py, no image-rendering path. Every other test-matrix cell passes, including check-reference-pdf-files. Could you kick off a re-run when convenient?

[andersonhc]

Thank you for the contribution.

Before merging this, I think there is one more context-manager scenario we should cover in this PR.

__exit__() restores font_family, font_style, font_size_pt, and current_font, but it does not restore/invalidate current_font_is_set_on_page. Because of that, after leaving text_columns(), the Python-side font state can say 12pt while the PDF content stream still has the 24pt font selected. The next pdf.cell() may then skip emitting a corrective Tf operator and render with the wrong size.

A regression test that shows this is:

def test_text_columns_restore_page_font_after_context(tmp_path):
    pdf = FPDF()
    pdf.add_page()
    pdf.set_font("Helvetica", size=12)

    with pdf.text_columns() as cols:
        pdf.set_font("Helvetica", size=24)
        cols.write("Large heading")
        pdf.set_font("Helvetica", size=10)
        cols.write("Small body text")

    pdf.cell(text="after columns")
    assert_pdf_equal(
        pdf,
        HERE / "text_columns_restore_page_font_after_context.pdf",
        tmp_path,
    )

[Pawansingh3889]

Apologies for the noisy CI on the c84b7fa amend. Two existing tests in test_text_columns.py (test_tcols_balance and test_text_columns_with_shorter_2nd_column) regressed because the current_font_is_set_on_page = False invalidation correctly forces a Tf operator re-emission for text rendered after a text_columns context, exactly the bug you flagged. The old reference PDFs had captured the buggy state where that Tf was missing.

Pushed b46ea413 regenerating both reference PDFs. Byte deltas are tiny (-8B and +3B), consistent with Tf operator reorganization and no other rendering change.

One caveat: test_text_columns_with_shorter_2nd_column uses set_text_shaping(True), so its bytes depend on the local uharfbuzz version. I regenerated it with uharfbuzz 0.54.1 (current latest). If CI's version differs and the hash still mismatches, I'll iterate, or you can regen on your end and I'll rebase.

The original PR description's claim that every existing snapshot stays byte-identical is no longer accurate after this amend; the flag-invalidation legitimately changes those two snapshots. Want me to edit the description, or leave it for the merge commit message to clarify?