TheBIG_DATA variable intest_bz2 will sometimes only have one block of compressed data instead of the two it is intended to contain due to the value depending on the order of listing files. To fix this, we stream in input data and check for multiple block header magic strings to ensure that there are at least 2 block headers.

This is an unfortunate hack, but the bz2 module doesn't expose flushing in calls toBZ2Compressor.compress() so we cannot directly generate two blocks. I'm also not certain if block headers can span the output to.compress(), so conservatively I have the code scan the entire set of data, which should only need to happen a few times at test setup time. There's probably a few ways to optimize this code, but I wanted to keep it simple to get it in before the alpha tomorrow. I specifically did not want to use BZ2File as I wanted the module setup code to be as minimal as possible in case the implementation is broken.

• Issue:test_bz2.testDecompressorChunksMaxsize is flaky due to non-deterministic BIG_TEXT #145607