GeezerVison

Since Python 3.12, GzipFile collects writes in an io.BufferedWriter. IGzipFile inherits writelines() from io.IOBase, which calls write() once for each line, so each line runs the Python code of IGzipFile.write() before it reaches that buffer.

With this PR, IGzipFile.writelines() passes the lines to the buffer's own writelines() on Python 3.12 and later, which calls write() in C for each line. Closed files, read mode, and a write() that a subclass, mock.patch, monkeypatch or an instance attribute replaced still go through io.IOBase.writelines(), so their errors and per-line write() calls stay the same. Python 3.11 has no write buffer, so nothing changes there.

I measured on Python 3.14 in Docker on Linux, writing to an io.BytesIO, and took the median of 5 interleaved rounds of the best of 5 runs:

A run of the same change on Python 3.13 gave about the same speedup.

Errors for bad lines are the same as before, because they already came from the buffer's write(), and the compressed output is identical to writing each line with write(). One edge case changes: if the iterable closes the file partway through, the remaining lines go into the write buffer and are lost, where write() raised ValueError before.

The new tests compare writelines() with a write() loop for lists, iterators, empty input, mixed bytes-like objects and long lines. They also check that the TypeErrors match gzip.GzipFile, that the lines before an error raised by the iterable are kept, the errors for closed and read-only files, that replaced write() methods are still called, and, on 3.12 and later, that the lines reach the buffer in one call. The test suite passes on this branch on Linux with CPython 3.11 and 3.14.

Checklist

• Pull request details were added to CHANGELOG.rst
• Documentation was updated (if needed)