bpo-20082: fix misbehavior of buffered writes to raw files in append mode

[embray]

This is a patch for a bug I apparently wrote many years ago but never converted to a PR.

https://bugs.python.org/issue20082

[embray]

Hmm, this seems to suddenly be causing some failures I didn't expect :/

[ZackerySpytz]

This function should be checked for failure, so it should not return void.

[embray]

I'm not so sure. What is its failure state? It either sets self->appending = 1 or self->appending = 0. Neither case represents a failure or success. AFAICT all calls within it check for failure or success and are handled appropriately.

[serhiy-storchaka]

getattr() can raise an arbitrary exception, for example MemoryError or KeyboardInterrupt. They should not be cleared.

BTW, in 3.13 you can use PyObject_GetOptionalAttr() which automatically silences an AttributeError. It exists under other name as private API in older versions.

PyUnicode_FindChar() also can raise an exception, according to the specification, although it is less likely after getting rid of legacy strings.

[embray]

Resolved all the trivial code style issues. Remaining question is whether _bufferedwriter_set_append should have a return value (if so I'm not clear on what it should be).

I think the issue that was causing test failures is fixed.

[embray]

Hmm, now the CI systems are all consistently showing a new test failure, but I cannot reproduce it locally even with the exact same test command they're running.

Update: I can reproduce them now by configuring --with-pydebug.

[embray]

I think this is ready now for the next round of review. Not sure why the Azure pipeline failed but it doesn't seem related.

[github-actions]

This PR is stale because it has been open for 30 days with no activity.

[python-cla-bot]

All commit authors signed the Contributor License Agreement.