Bug report

Bug description:

When io.StringIO fails with MemoryError while converting its accumulated writes to the internal buffer (first readline(), iteration, partial read(), truncate() or a write that is not at the end), the text is dropped but the object keeps its old length. Later calls then index outside a 2-character buffer: they crash or return unrelated memory as text. getvalue() and a full read() can leave the object in the same state.

import io, resource
resource.setrlimit(resource.RLIMIT_DATA, (128 << 20, 128 << 20))
f = io.StringIO()
f.write("x" * (32 << 20))
f.seek(0)
try:
    f.readline()
except MemoryError:
    print("MemoryError")
print(f.seek(0, 2))
f.seek(0)
f.readline()

On a debug build of main this prints MemoryError, then 33554432, then dies with a segmentation fault in the second readline(). Expected: the object still holds its text or is empty, and the second call raises MemoryError or returns normally.

email.feedparser can reach this: FeedParser.close() after a feed() that raised MemoryError on a very long first line. gh-157242 was a similar state-after-MemoryError bug in bytearray.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-160324: Fix io.StringIO state after MemoryError #160325
• [3.15] gh-160324: Keep io.StringIO state after MemoryError (GH-160325) #161041
• [3.14] gh-160324: Keep io.StringIO state after MemoryError (GH-160325) #161042