To reproduce, run test_zstd.CompressorTestCase.test_compress_flushblock or under UBSan:

>>> from compression.zstd import ZstdDecompressor, compress
>>> d = ZstdDecompressor()
>>> d.decompress(compress(b'Hi Emma!'))
b'Hi Emma!'
>>> d.unused_data
Modules/_zstd/decompressor.c:598:56: runtime error: applying non-zero offset 18446744073709551615 to null pointer
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior Modules/_zstd/decompressor.c:598:56 
b''
>>> 

ZstdDecompressor.unused_data computes self->input_buffer + self->in_begin without checking that input_buffer was ever allocated, so when a frame is fully consumed in a single decompress() call it leaves input_buffer NULL (and in_begin/in_end at their initial state). Reading .unused_data evaluates NULL + ..., which is UB. Although the resulting length of in_end - in_begin == 0 means it accidentally returns the correct b''.

The other unused_data implementations are not affected due to their design.

Also, CC @Rogdham

• Issue: Various UB warnings found by UBSan #148286