Re: cleaner/better zlib sources?

Linus Torvalds [off-list ref] wrote:

I looked at git profiles yesterday, and some of them are pretty scary. We 
spend about 50% of the time under some loads in just zlib uncompression, 
and when I actually looked closer at the zlib sources I can kind of 
understand why. That thing is horrid.

Yes.  This is actually one of the motivations behind pack v4.
We don't store the "important bits" of commits and trees in zlib
compressed format at all; allowing us to completely bypass the
inflate() penalty you describe.

We're already much faster on the linux-2.6 kernel tree, and that's
*with* converting the pack raw data into text, then reparsing
that text into a struct commit* or a struct name_entry using the
current code.  We're also planning on reworking those parsers to
parse the raw pack data, allowing us to save some very unnecessary
raw->string->raw conversion time.

But Nico and I still looking to use zlib for commit messages and
blob content, so any improvements to inflate (or its replacement)
would still be most helpful.

-- 
Shawn.