Deflate settings changed when replacing in-tree flate with flate2

[oyvindln]

In the pr that replaced the in-tree flate library, with flate2, there were some changes to how the compression was used, that may not have been intentional. I was suggested to make an issue after commenting there as the PR was already merged.

Is using the Compression::Default compression level intended? The flate functions were changed to use a faster compression level some time ago (see #37298), going back to a higher compression setting could cause some performance regressions. I think the previous settings would be equal to using the Compression::Fast setting in flate2.

Also, the old functions didn't use a zlib wrapper from what I can see (there is a flag for that), but the new code does. Don't know if this could cause some issues when trying to use libs from different versions, though at least it adds 6 bytes of overhead due to the header and checksum.

[Mark-Simulacrum]

cc @alexcrichton

Seems like going back to a lower compression level here is a good idea at least. I don't know if worrying about the zlib wrapper is too important though.

[oyvindln]

The zlib wrapper could cause problems if trying to compress data with the zlib wrapper using the functions that don't expect it, and vice versa. I don't know if rustc will read the rlib files from a different version though, so this may or may not be an issue. Changing this back to the old behaviour is just changing to use the reader/writers that operate on raw deflate streams so it would be very easy to change if it's needed.

[alexcrichton]

Ah yes no functional changes were intended! Should be easy to switch back to the deflate scheme and switch the compression level!