fix(parsers/python): guard extraction so one bad file can't abort the whole repo parse

[gadievron]

What

Add a per-file guard so a single pathological Python file cannot abort the whole repo's extraction (mirrors the existing Zig/Go parsers).

Root cause

FunctionExtractor.process_file guarded only ast.parse (except SyntaxError); the extraction body and both caller loops (extract_from_scan, extract_all) were unguarded. A single file raising a non-SyntaxError — a RecursionError from a deeply-nested source, or any error in the tree walk — propagated out of the file loop and aborted the entire parse, losing all units collected so far.

How

• Add _process_file_guarded (try/except Exception -> log a WARNING to stderr + files_with_errors += 1) and route all three process_file call sites (extract_from_scan + both extract_all branches) through it. Mirrors parsers/zig/function_extractor.py:57-77.
• Move files_processed += 1 to the end of process_file, so a file that crashes mid-extraction is counted once as an error and never as processed (no double-count).
• except Exception (not BaseException) so KeyboardInterrupt/SystemExit still propagate; the failure is logged, never silently dropped.

Regression test

tests/parsers/python/test_function_extractor_robustness.py drives the real entry points (extract_from_scan = the production path used by parse_repository, and extract_all), not hand-built dicts. It covers a parse-time RecursionError, a post-parse extraction error (via monkeypatched process_function), and stats correctness.

RED on origin/master (98f5504):

$ python -m pytest tests/parsers/python/test_function_extractor_robustness.py -q
4 failed

GREEN on this branch:

$ python -m pytest tests/parsers/python/test_function_extractor_robustness.py -q
4 passed

Verification

Byte-identical unit-id set + statistics on click/flask/httpie/rich via both extract_from_scan and extract_all (no change on valid code); recovers every real unit when one file raises at parse OR post-parse. Python parser test suite unchanged vs base.

Compatibility / scope

• Based on current master; may need a rebase if Python parser: fix 10 finder bugs #112 / fix(parsers/python): extract block-scoped def/class units #125 / fix(parsers/python): segment-match path exclusion/classification + resolve relative-import anchors #90 (which also touch function_extractor.py) land first. Master already carries the block-scoped extraction shape, so the change merges clean today.
• The same latent unguarded-loop pattern in the JS / C / Ruby / PHP extractors is a tracked follow-up, out of scope here.

Author notes

• Implements: _process_file_guarded + the three re-routed call sites + the files_processed move.
• Input previously handled wrong: a repo containing one deeply-nested file (RecursionError in ast.parse) returned zero units for the entire repo; it now skips that file and returns the rest.
• Likely reviewer pushback: "a broad except swallows real errors" — it catches Exception (not BaseException), logs each failure to stderr with the file + exception type, and counts it in files_with_errors, matching the Zig/Go guard and the file's existing stderr-print convention.

🤖 Generated with Claude Code