What I had in mind was UB changing over time more than tarballs getting corrupted (eg. strict aliasing-related optimizations).
There's practically no chance for a corrupted gzip-compressed tarball to decompress cleanly (even if by chance it inflates without error, there's a CRC32 of the uncompressed payload at the end of the gzip stream).
I don't want to keep posting this[1] in every C discussion, but when it's relevant, it's relevant. TL;DR A particular version of a compiler interpreted undefined behavior in such a way as to elide code that specifically tested for an error case to exit early because the error implied undefined behavior, so the test could never be reached without undefined behavior. If that sounds like circular logic, that's because it is.
The error happened to exist in a fairly specific version of clang shipped by apple, but not in any of the major release versions:
I was not able to reproduce this bug on other platforms, perhaps because the compiler optimization is not applied by other compilers. On Linux, I tested GCC 4.9, 5.4, and 6.3, as well as Clang 3.6, 3.8, and 4.0. None were affected. Nevertheless, the compiler behavior is technically allowed, thus it should be assumed that it can happen on other platforms as well.
The specific compiler version which was observed to be affected is:
