fix(svg): keep the stripped width/height when detecting a percentage

[Anai-Guo]

Problem

SVGObject.extract_shape_info() calls .strip() on the width/height attributes as bare statements:

self.width = None
if width_str is not None:
    width_str.strip()                     # <-- result discarded
    if width_str.endswith("%"):
        self.width = Percent(width_str[:-1])
    else:
        self.width = resolve_length(width_str)

Python strings are immutable, so .strip() returns a new string and the original is unchanged — the whitespace is still there on the next line.

A trailing space then defeats the percentage check: "100% ".endswith("%") is False, so the value falls through to resolve_length(), which parses the unit as % and rejects it as a relative length:

>>> SVGObject('<svg xmlns="http://www.w3.org/2000/svg" width="100% " height="100% " viewBox="0 0 10 10"></svg>')
ValueError: 100%  uses unsupported relative length %

Thirteen lines below, in the same method, viewbox already does it correctly:

viewbox = viewbox.strip()

Scope

Only percentages are affected. unit_splitter (\s*(?P<value>[-+]?[\d\.]+)\s*(?P<unit>%|[a-zA-Z]*)) tolerates leading whitespace and .match() ignores anything trailing, so absolute lengths already survive the missing strip. Measured on the current code:

width	before	after
"100%"	Percent(100.0)	Percent(100.0)
"100% "	ValueError	Percent(100.0)
" 100% "	ValueError	Percent(100.0)
"100" / "100 " / "100px "	fine	fine (unchanged)

Leading-only whitespace (" 100%") happens to work already, because endswith("%") still holds and float(" 100") accepts the space — so 3 of the 4 whitespace cases in the new test fail without the fix.

Fix

Keep the stripped value, which is what the code was already trying to do:

-            width_str.strip()
+            width_str = width_str.strip()
...
-            height_str.strip()
+            height_str = height_str.strip()

Test

Added test_document_shape_info_percent_surrounded_by_whitespace to TestSVGObject, parametrized over "100% ", " 100%", " 100% ", "100%\n" (a newline is easy to hit when the attribute is written across lines in hand-authored SVG).

Verified both directions rather than just the happy one:

# with the patch reverted, new test only
3 failed, 1 passed        (" 100%" passes -- see the scope table above)

# with the patch applied
test/svg/  ->  315 passed

black (26.3.1, the revision pinned in .pre-commit-config.yaml) reports both files unchanged.

Note on the base commit and the CHANGELOG

My fork cannot be synced to master — the GitHub API refuses merge-upstream for this token because the sync would touch .github/workflows/codeql.yml, which needs the workflow OAuth scope — so this branch is based on my fork's own tip. I checked the drift explicitly: fpdf/svg.py and test/svg/test_svg.py are byte-identical between that commit and current master, so the diff is exactly what it looks like.

CHANGELOG.md is the one file that has moved since (the [2.8.9] section gained its ### Fixed heading and two entries). Writing my entry against the older copy either conflicts or auto-merges into a duplicated ### Fixed heading, so I have deliberately left CHANGELOG.md out rather than commit something that merges wrong. The entry I would add is:

* SVG `width`/`height` attributes holding a percentage surrounded by whitespace (_e.g._ `width="100% "`) no longer raise `ValueError: uses unsupported relative length %`; the value is now stripped before the percentage is detected, as the code already intended - _cf._ [PR #1936](https://github.com/py-pdf/fpdf2/pull/1936)

Happy to add it (or rebase the whole branch) on request.

---

🤖 Generated with Claude Code