[ptweezy]

test_decompressor_inputbuf_1, _2, and _3 come from CPython's test_bz2. They expect part of a 100 to 300 byte input to be left over in the input buffer of IgzipDecompressor. But isal_inflate decompresses up to about 64 KiB ahead into a buffer of its own, so it reads these inputs in full. With gcov, the input buffer code in decompress() in igzip_libmodule.c runs in none of the three tests.

This PR uses 512 KiB of random data, which needs about as much input as output, and multiplies the offsets by 1,000. In the first test, the second call asks for 80,000 bytes instead of 2, because isal_inflate first returns the data in its own buffer and reads more input only after that. With gcov, the tests now do what their comments say: the first moves the contents of the input buffer to its start with memmove, the second appends to the buffer without moving or resizing it, and the third resizes it with PyMem_Realloc.

To test it, I made two changes to decompress(). Without the memmove, the first test fails. When the offset is dropped after PyMem_Realloc, the second test fails. All tests in tests/test_igzip_lib.py on develop pass with both changes. With an AddressSanitizer build of igzip_lib, the new tests pass. Together, the three tests take about 0.2 s.

This PR changes only tests, so it has no CHANGELOG entry.

Checklist

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

[rhpvorderman]

LGTM. Just a minor naming nitpick.

[rhpvorderman]

This is a bit unfortunate naming of the variables. This should be RANDOM_DATA as it is not text and COMPRESSED_RANDOM_DATA respectively.

[ptweezy]

Thanks, renamed them in 5ae2de7.