daily-fix: --map-files misuse OSError before #1613 guard

Overview / Motivation

Auto-filed by the /daily Step C parked-candidate sweep (2026-07-28) from TWO same-bug parked candidates on the same file: task #1762 (ts 2026-07-28T15:16:59Z, fp eef8cf00a9f3, implementer round 1) and task #1764 (ts 2026-07-28T16:28:05Z, fp b1c4a980a273, code-reviewer round 1). Both describe the same failure — passing a payload/markdown file (not a newline-delimited path list) to select_step9c_tests.py --map-files crashes with an unhandled OSError: [Errno 36] File name too long from a per-line .exists() probe, instead of reaching the #1613 misuse diagnostic — with two complementary fixes (a content-shape pre-check and an OSError-tolerant existence helper).

Goal

Make --map-files misuse with a non-path-list payload reach the graceful #1613 diagnostic instead of a raw traceback: guard the per-line existence probes with an OSError/ValueError-tolerant helper, and/or pre-detect non-path-shaped lines and exit 2 with the existing FILE-of-paths message.

Workflow gap

• Bug observed: --map-files <source-or-markdown-file> crashes with OSError: [Errno 36] File name too long from (work_root / f).exists() when any content line exceeds NAME_MAX; the #1613 guard (which exit-2s cleanly for .py/.sh payloads) is defeated by its own probe — the existence check raises before the heuristic can report.
• Why it is a workflow gap: the #1613 misuse guard exists precisely for this input class; the operator gets a raw traceback instead of the designed diagnostic naming the FILE-of-paths contract.
• Confidence (emitters): medium (#1762) / high (#1764)
• verified-at-filing: grep -n '_safe_exists' scripts/select_step9c_tests.py → 0 hits (absence of the proposed helper), and code read confirms two unguarded per-line probe sites — if (work_root / f).exists(): (~L1434 region) and if any((work_root / f).exists() for f in files if not f.startswith("/")): (~L1568 region, immediately before the #1613 .py/.sh diagnostic) (2026-07-29 UTC). Landed-fix history check: git log --oneline --since='7 days ago' -- scripts/select_step9c_tests.py → no commit addressing the OSError path.

Proposed change (candidate diff sketch — refine in planning)

+ def _safe_exists(p: Path) -> bool:
+     try:
+         return p.exists()
+     except (OSError, ValueError):
+         return False  # unstat-able content line (e.g. >NAME_MAX) is not a repo path
- if any((work_root / f).exists() for f in files if not f.startswith("/")):
+ if any(_safe_exists(work_root / f) for f in files if not f.startswith("/")):
  (same guard at the L1434 sibling)

Optionally ALSO extend the #1613 guard with a cheap non-path-shape line check (e.g. len(line) > 512 or line.startswith(("- ", "# ", "**")) → exit 2 with the existing diagnostic), per the #1762 sketch — the planner picks the combination.

Scope / surfaces

• Primary target: scripts/select_step9c_tests.py (both .exists() probe sites + the #1613 guard)
• Add a regression test: --map-files fed a markdown payload exits 2 with the diagnostic, no traceback.

Constraints / invariants

• Exit-code contract of #1613 preserved (exit 2 + diagnostic naming the FILE-of-paths contract).
• Workflow-surface only; ruff passes; recursion guard applies to the spawned session.

Provenance

• workflow_fix_target: scripts/select_step9c_tests.py
• fingerprint: b1c4a980a273

target_file: scripts/select_step9c_tests.py bug_observed: --map-files <source-file> misuse crashes with an unhandled OSError: [Errno 36] File name too long from (work_root / f).exists() (L1568) when any content line of the mis-passed file exceeds NAME_MAX, instead of reaching the graceful #1613 "looks like a source file, not a path-LIST file" diagnostic (which fired correctly for a short-lined source file in the same session). why_workflow_gap: The #1613 misuse guard exists precisely for this input class but is defeated by its own probe — the existence check raises before the heuristic can report, so the operator gets a raw traceback instead of the designed diagnostic. proposed_change: Guard the per-line existence probes (L1434, L1568 — both instances of the class) with an OSError/ValueError-tolerant helper that treats unstat-able lines as non-paths. confidence: high related_task: #1764

(Second same-bug park, task #1762, fp eef8cf00a9f3, ts 2026-07-28T15:16:59Z — deduped into this filing; its routed-record on #1762 names this task. Its complementary content-shape pre-check sketch is included above for the planner.)