fix(file-ops): keep spaced filenames in files_only search results

[TabooHarmony]

What does this PR do?

Fixes a regression from #39858 where search_files in files_only mode silently drops any matching file whose name contains a space.

#39858 added _split_tool_diagnostics to separate rg/grep error text from match output by line shape (since _exec merges stderr into stdout). The bare-path shape regex (^[^\s:][^\s]*$) forbids whitespace, so a files-only output line like My Document.md or sub/Meeting Notes.txt fails the regex, carries no rg:/grep: prefix, and gets folded into the "diagnostics" bucket — stripped from the result. No error is raised, so the loss is silent.

Content and count modes are unaffected: their path:line:content / path:count shapes satisfy the regex via the :N anchor. Only files_only, with its bare unanchored path lines, is hit.

The fix makes _split_tool_diagnostics mode-aware. In files_only mode it tracks rg's multi-line regex parse error: block and strips a line only while inside an active block. A regex parse error aborts the search, so no real matches are interleaved with its continuation lines — every other files_only line is treated as a real path and preserved verbatim, including names with spaces and (as a bonus) names that happen to start with error: or with leading whitespace. The hard-error path (rg:/grep: prefixed lines and the parse-error block) is still classified as diagnostics, so genuine failures are surfaced unchanged.

Related Issue

No separate issue — the regression and its repro are self-contained below.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• tools/file_operations.py:
  • _split_tool_diagnostics takes an output_mode argument (default "content", backward compatible). In files_only mode it classifies by parse-error-block tracking instead of the whitespace-hostile content shape regex.
  • Both call sites in _search_with_rg / _search_with_grep pass output_mode.
• tests/tools/test_search_error_guard.py:
  • test_files_only_preserves_spaced_paths — spaced names survive.
  • test_files_only_still_drops_regex_parse_error — hard error still classified as diagnostics in files_only mode.
  • test_files_only_keeps_path_named_like_error_summary — a file named error: notes.md is preserved.
  • test_files_only_keeps_indented_path_outside_error_block — a path with leading whitespace is preserved outside a parse-error block.

How to Test

Reproduce the drop on main:

from tools.file_operations import ShellFileOperations
from tools.environments.local import LocalEnvironment
import os, tempfile

root = tempfile.mkdtemp(); os.makedirs(f"{root}/sub")
for n in ["normal.py", "My Document.md", "sub/Meeting Notes.txt", "sub/clean.txt"]:
    open(f"{root}/{n}", "w").write("NEEDLE\n")

ops = ShellFileOperations(LocalEnvironment(cwd=root), cwd=root)
print(sorted(ops._search_with_rg("NEEDLE", root, None, 50, 0, "files_only", 0).files))

Before: only normal.py and sub/clean.txt (two spaced names dropped, no error). After: all four files returned. Reproduces on both _search_with_rg and _search_with_grep.

Run the tests:

pytest tests/tools/test_search_error_guard.py::TestSplitToolDiagnostics -q

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(file-ops):)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix
• I've run the relevant tests and they pass
• I've added tests for my changes
• I've tested on my platform: Linux

Documentation & Housekeeping

• N/A — no docs/config/architecture changes

[teknium1]

Thanks for isolating the regression. The premise is confirmed on current main: tools/file_operations.py:401 rejects whitespace in bare files_only paths, while both backend paths call the splitter at tools/file_operations.py:2304 and tools/file_operations.py:2432.

Problems

• The new tests in tests/tools/test_search_error_guard.py call _split_tool_diagnostics directly. They do not verify either production backend forwards output_mode or returns a spaced file path through its files_only parsing path.

Suggested changes

• Add a parameterized end-to-end test using the existing _METHODS and _search helpers in tests/tools/test_search_error_guard.py:63-73; create spaced filenames and assert each available backend returns them in files_only mode.

This is an automated hermes-sweeper review.

[TabooHarmony]

Addressed the review feedback and rebased onto current main. Added a parameterized end-to-end regression test using the existing _METHODS and _search helpers. It creates filenames containing spaces and verifies that both available production backends return them in files_only mode.

The new test fails against current main for both grep and ripgrep, and passes with this fix. Relevant file-operation and search tests pass locally, aside from the existing unreadable-file cases that cannot reproduce under root because root can read chmod(000) files.