[codeofwxz]

Related #60.

The tools open an inplace destination with wb before consuming the lazy subtitle parser. A parsing or processing error therefore leaves the original file empty. Encoding errors can also occur after truncation.

Keep the inplace output path unopened until the complete output has been composed and encoded. A shared write_output helper performs that final write; all seven writing tools use it, including mux's unchanged non-inplace path. Ordinary stdout and explicit output streams retain the existing encoding/fallback logic.

This protects the original input from parsing, conversion, and encoding failures. It does not provide atomic replacement or protection against write errors, process termination during writing, or power loss. The destination is still opened directly, preserving the existing inode, permissions, and link-following behavior. No runtime dependencies or new Python requirements are introduced.

Validation:

• Real subprocess tests cover the six inplace tools, successful stdout/file/inplace output equivalence, parsing and conversion errors, encoding failures, UTF-8/UTF-16 output, hard links, parameter conflicts, and ignore-parsing-errors. The initial regression set had 10 failures on the baseline; it passes after the change.
• Python 3.9.13 with the repository's test dependency ranges: 68 passed, 1 skipped; doctest run: 74 passed, 1 skipped.
• Python 3.14.4 with current pytest/Hypothesis: 68 passed, 1 skipped; the configured srt.py branch coverage target is 100%. This coverage target does not measure the CLI helpers.
• The skipped test creates a symbolic link, which this Windows environment does not permit. Hard-link behavior passed. Actual Python 2.7 and the full OS/version matrix were not run; a Python 2.7 grammar check accepted the helper, CLI scripts, and new test file.
• Black and the configured Bandit check pass. The configured Pylint command reports the same three diagnostics on the unchanged baseline (one removed-option warning and two positional-argument warnings). Pytype was not run.

All test subtitles are synthetic. Repository maintenance is currently infrequent; no maintainer response or acceptance is assumed.

[Copilot]

🟢 Approval recommended

All reviewed files are covered and no blocking issues were identified.

Pull request overview

This pull request prevents in-place files from being truncated before processing and encoding succeed.

Changes:

• Adds shared deferred-output handling.
• Updates all seven writing tools.
• Adds comprehensive CLI regression tests.

File summaries

File	Summary
tests/test_tools.py	Adds regression coverage for output modes, failures, encodings, and links.
srt_tools/utils.py	Implements deferred in-place writing.
srt_tools/srt-process	Uses the shared output helper.
srt_tools/srt-normalise	Uses the shared output helper.
srt_tools/srt-mux	Uses the shared output helper.
srt_tools/srt-lines-matching	Uses the shared output helper.
srt_tools/srt-linear-timeshift	Uses the shared output helper.
srt_tools/srt-fixed-timeshift	Uses the shared output helper.
srt_tools/srt-deduplicate	Uses the shared output helper.

Review details

• Files reviewed: 9/9 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.