Only define Z_SOLO on wasm32-unknown-unknown, not the wasi/emscripten targets

[calvinrp]

build_zlib() defines Z_SOLO for every wasm32* target. Z_SOLO is
zlib's "no C library" mode: no default zalloc/zfree — so the one-shot
helpers compress()/uncompress() always return Z_STREAM_ERROR at
runtime — and no gz* API.

Only wasm32-unknown-unknown actually lacks a libc. The wasi targets
link wasi-libc (full malloc/free and fd-based I/O) and emscripten ships
musl, so there the blanket define just breaks working API: on
wasm32-wasip1,

let rc = libz_sys::compress(dst.as_mut_ptr(), &mut dst_len,
                            src.as_ptr(), src.len() as _);
assert_eq!(rc, libz_sys::Z_OK); // fails today: rc == -2 (Z_STREAM_ERROR)

compiles fine and fails at runtime on every call, while the identical
program works on native targets. flate2 doesn't notice because it always
supplies its own allocators through the streaming API; direct libz-sys
users on wasi hit it with no workaround inside libz-sys.

Minimal repro (any wasi runtime, e.g. wasmtime):

cargo new zsolo && cd zsolo && cargo add libz-sys@1.1.29
# main.rs: call libz_sys::compress() on any buffer and print the rc
cargo build --target wasm32-wasip1 && wasmtime target/wasm32-wasip1/debug/zsolo.wasm
# prints rc = -2 (Z_STREAM_ERROR); the same program prints 0 on native

This PR scopes the define to wasm32-unknown-unknown. wasi/emscripten
now take the standard source list, including the gz* files, which
compile cleanly against wasi-libc (verified on wasm32-wasip1: the crate
builds, and a one-shot compress/uncompress roundtrip passes where it
returned -2 before). src/lib.rs needs no change — it already declares
the one-shot functions unconditionally.

[Copilot]

Pull request overview

This PR fixes an incorrect zlib build configuration for WebAssembly targets by scoping Z_SOLO to only wasm32-unknown-unknown, avoiding runtime breakage of compress()/uncompress() (and gz*) on wasi/emscripten where a C library is available.

Changes:

• Change build_zlib() to define Z_SOLO only for wasm32-unknown-unknown (instead of all wasm32* targets).
• Ensure non-wasm32-unknown-unknown wasm targets build the standard zlib source set (including gz* sources).
• Add clarifying comments explaining the rationale and behavioral impact of Z_SOLO.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Byron]

Thanks a lot for contributing! This seems to be good, but it also doesn't look like we have any WASM tests here.

@jongiddy do you think it's feasible to try to add WASM test coverage? If not, or if you don't know, do you think it's safe to merge this PR as is? Thanks you. (PS: I didn't investigate this with an agent yet).

[folkertdev]

In zlib-rs we test with custom cargo runners with a wide range of target architectures, mostly with qemu but we also include wasmtime for webassembly.

https://github.com/trifectatechfoundation/zlib-rs/blob/5a82af8a944d3b2d23e861c5438bac35bc40d16d/.github/workflows/checks.yaml#L22-L115

[Byron]

Thanks for chiming in @folkertdev!
I will give it a shot.

[Byron]

This seems to work - I will have the other WASM runtimes added as well so we can test all environments that are affected by this change.

[Byron]

@folkertdev Thanks again! There are now tests for a bunch of WASM runtimes, which validate the claims that were originally made here. Thus it appears like the PR is truly an improvements, and thanks to CI it should stay that way.