use zlib-rs for gzip compression in rust code

[folkertdev]

What does this PR try to resolve?

This PR uses zlib-rs via the flate2 crate. It is used for (de)compressing gzip files (e.g. in cargo package).

Using zlib-rs is significantly faster, and produces output of roughly similar size. For the windows-bindgen crate
(a very large crate), the speedup is over 60%, or about 2 seconds.

> time cargo package -p windows-bindgen --no-verify
   Packaging windows-bindgen v0.61.0 (/home/folkertdev/rust/windows-rs/crates/libs/bindgen)
    Updating crates.io index
    Packaged 76 files, 31.2MiB (8.2MiB compressed)

________________________________________________________
Executed in    3.30 secs    fish           external
   usr time    3.19 secs  424.00 micros    3.19 secs
   sys time    0.05 secs   61.00 micros    0.05 secs

> time ~/rust/cargo/target/release/cargo package -p windows-bindgen --no-verify
   Packaging windows-bindgen v0.61.0 (/home/folkertdev/rust/windows-rs/crates/libs/bindgen)
    Updating crates.io index
    Packaged 76 files, 31.2MiB (8.3MiB compressed)

________________________________________________________
Executed in    1.25 secs    fish           external
   usr time    1.15 secs    0.00 micros    1.15 secs
   sys time    0.04 secs  589.00 micros    0.04 secs

How should we test and review this PR?

Generally CI/the test suite should handle correctness.

Something to look out for is zlib-rs producing larger binaries than before. So far we're seeing output sizes that are roughly the same (sometimes a bit better, sometimes a bit worse) as the status quo.

We've not specifically looked at decompression yet, mostly because we could not come up with a good command to benchmark. In general zlib-rs is much faster than stock zlib there too.

Additional information

For the time being, cargo still depends on stock zlib via e.g. curl and git. As far as I know it is the intention to eventually move away from these C dependencies, at which point the dependency on stock zlib also disappears.

[rustbot]

r? @epage

rustbot has assigned @epage.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[joshtriplett]

Thanks, @folkertdev!

Nominating for next week's Cargo meeting.

[epage]

We talked about how we want to approach these kind of changes in today's Cargo team meeting.

For gitoxide, its still under active development and each feature needs a lot of vetting before adoption, so we're taking an incremental, feature-at-at-ime, approach with an opt-in and opt-out windows to ensure as smooth of a path as possible.

For something like zlib-rs, it feels smaller and more mature. We likely don't need a a multi-release transition window. We do want it merged at the beginning of a release window to give as much runtime within the release as possible. We also appreciate how easy it would be to back this out during beta if need be. The main risk we see with zlib-rs is platform support.

For our -sys crates, we allow people building cargo to decide whether to use the vendored or system version of the package. The question came up if we should still support those cases in addition to the pure-Rust implementation. I don't feel that that model needs to be applied when adopting a pure-Rust variant of C code. We could use C implementations of json, blake3, toml, base64, etc and the question didn't come up of whether that should be made configurable.

As this is still early in the release Window, we can move forward with this.

[reneleonhardt]

Does this mean a final decision has been made or are alternatives like zstd still on the table? #2526

[weihanglo]

This was just a performance improvement. Had no discussion around #2526 when it got merged.