linkcheck reports existing local files as broken when URI contains query or fragment

[SuYanqi]

Describe the bug

Linkcheck's local-file validation passes the raw URI (including query and fragment components) directly to a filesystem existence check. As a result, existing local files referenced with ?query or #fragment are incorrectly reported as broken.

For example, target.html exists, but the link target.html?view=full is checked as the literal filename target.html?view=full instead of stripping the query first.

How to Reproduce

import io
import shutil
from pathlib import Path
from sphinx.application import Sphinx

root = Path("/tmp/sphinx_linkcheck_test3")
shutil.rmtree(root, ignore_errors=True)
(root / "_build").mkdir(parents=True)
(root / "target.html").write_text("Target", encoding="utf-8")
(root / "conf.py").write_text("")
(root / "index.rst").write_text(
    "Test\n====\n\n`target <target.html?view=full>`_\n"
)

app = Sphinx(str(root), str(root), str(root / "_build"), str(root / "_doctree"),
             "linkcheck", status=io.StringIO(), warning=io.StringIO())
app.build(force_all=True)
print((root / "_build" / "output.txt").read_text())

Expected: index.rst:4: [working] target.html?view=full
Actual: index.rst:4: [broken] target.html?view=full:

The same issue occurs with fragments (target.html#section) and query-only URIs (?mode=full).

Environment Information

Sphinx: 9.0.4
Python: 3.11.13
Platform: Linux

Sphinx extensions

[]

Additional context

This regression was introduced by PR #7985.

The relevant code at sphinx/builders/linkcheck.py line 475:

if (hyperlink.docpath.parent / uri).exists():

The uri here is the raw string target.html?view=full — it should first strip ? and # suffixes before checking the filesystem.

[jayaddison]

Thanks for filing a bugreport - I have two questions about this:

• It is valid for Unix filenames to contain # and ? characters - so, for example, a filename of Who watches the watchers?.pdf is valid. Is it an acceptable risk that we might break some existing file links, or make it difficult to add certain local files?
• As far as I am aware, the querystring part of the URL is only relevant for webpages that are retrieved remotely -- in comparison to the fragment part that web browsers may use when opening local HTML files. Is support for both required?