[BUG]Console.save_text clears recorded output when file write fails

[sanjay3941]

• I've checked the docs and closed issues for possible solutions.
• I can't find my issue in the FAQ.

Describe the bug

When using Console(record=True), calling save_text() with clear=True
appears to clear the recorded console output even when the file cannot be
successfully written.

Minimal Repoduction of the issue:

from rich.console import Console

console = Console(record=True)

console.print("IMPORTANT OUTPUT")

print("Before failed save:")
print(repr(console.export_text()))

try:
    console.save_text(
        r"Z:\this\path\should\not\exist\output.txt"
    )
except Exception as e:
    print("\nSave failed:")
    print(type(e).__name__, e)

print("\nAfter failed save:")
print(repr(console.export_text()))

The output:
Before failed save:
'IMPORTANT OUTPUT\n'

Save failed:
FileNotFoundError [Errno 2] No such file or directory: 'Z:\this\path\should\not\exist\output.txt'

After failed save:
''

The file was not successfully written, but the record output was also cleared.

i would expect the recorded output to remain available when the save fails since the output was not sucessfully persisted.

but i am not sure whether the current behavior is intentional or not .? if not could it be like the buffer could be preserved until the file write succeeds ?

Platform:
Windows 11
Python 3.13
Powershell and VS code terminal

[vardhans07]

Hi @willmcgugan / @sanjay3941 , this looks like a valid bug. Can I take this up and work on a fix?

[vardhans07]

Hi @willmcgugan / @sanjay3941 ,

I have resolved this issue and added regression tests for save_text, save_html, and save_svg. The branch is ready in my fork here: https://github.com/vardhans07/rich/tree/fix-console-save-buffer-preservation

Since repository interactions are currently limited, please let me know once PRs can be opened or feel free to pull directly from the branch!

Put together a fix for this and added regression tests. Here's a summary of what changed:

What was happening

save_text(), save_html(), and save_svg() were calling their export functions with clear=True right away. If the file write ran into an OSError (bad path, permission issues, etc.), the record buffer was already cleared in memory, causing permanent data loss.

Changes made

1. Deferred buffer clearing (rich/console.py):
   • save_text, save_html, and save_svg now call export_* with clear=False.
   • self._record_buffer only gets wiped inside _record_buffer_lock once the file write actually finishes without error.
2. Fixed export_svg():
   • Updated export_svg() so it properly checks the clear flag instead of clearing the buffer unconditionally.
3. Tests (tests/test_console.py):
   • Added unit tests for save_text, save_html, and save_svg checking that the buffer stays intact if an exception is thrown during saving.

[pacocartones]

Hey @willmcgugan — confirmed this is a real bug, with a root cause and a fix ready.

Root cause (rich/console.py, current master 9d8f9a3): save_text, save_html and save_svg all call their export_* helper with clear=clear before opening the target file. export_text runs del self._record_buffer[:] (L2220-2221) and only then does save_text do open(path, "w") (L2241). If the write raises (e.g. non-existent path → FileNotFoundError), the recording buffer is already gone. Same ordering in save_html (export clears at L2317-2318, open at L2349) and save_svg (L2471-2472 / L2641).

Reproduced on master:

Before failed save: 'IMPORTANT OUTPUT\n'
Save failed: FileNotFoundError ...
After failed save: ''

Proposed fix: export with clear=False, write the file, and only clear the buffer once the write succeeds (under _record_buffer_lock). This keeps the clear=True default semantics on the happy path but stops data loss when the write fails. Small regression tests for all three save_* methods pass, and the existing save/export tests still pass.

Would you be OK with me opening a PR for this? Per your AI_POLICY.md I'd flag it as AI-generated (Claude / Claude Code) and link back to this issue for your approval.

(Note: I see @vardhans07 above already has a branch for this — happy to defer to them if you'd rather they open it. Just let me know which you prefer.)

[vardhans07]

Hi @pacocartones @willmcgugan,

Thanks! I am opening the PR now from my branch with the fix and regression tests.

[vardhans07]

Hi @willmcgugan,

I have the fix and regression tests ready, but I am unable to open a PR directly due to repository interaction limits.

Please pull directly from my branch:
https://github.com/vardhans07/rich/tree/fix-console-save-buffer-preservation

Thanks!

3be3337

[CyberneticX-Tech]

I reproduced this and have a proposed fix ready. The approach is to export without clearing the record buffer, perform the file write, and only remove the exported records (by snapshot length) after a successful write. Any records added concurrently during the write are preserved. I've also added parametrized regression coverage for save_text, save_html, and save_svg.

Would this approach be acceptable?

Disclosure: Assisted by Antigravity AI pair programming agent per AI_POLICY.md.

[vardhans07]

Hi @willmcgugan,

Quick status update: My branch is fully complete and ready for review/merge:
https://github.com/vardhans07/rich/tree/fix-console-save-buffer-preservation

Summary of what is included:

• Buffer preservation fix: Wipes the buffer only after successful file writes across save_text, save_html, and save_svg under _record_buffer_lock.
• Regression tests: Full unit tests covering all three save methods on write errors (all tests passing locally).
• Project files: Updated CHANGELOG.md under [Unreleased] and added entry to CONTRIBUTORS.md.

Because repo interaction limits currently restrict non-collaborators from opening a direct PR, feel free to pull directly from my branch or let me know if access opens up. Thanks!

4df62fb