Yoshua Wuyts (Apr 09 2024 at 22:12): Yoshua Wuyts (Apr 09 2024 at 22:12):Published this today on the Rust blog: https://blog.rust-lang.org/2024/04/09/updates-to-rusts-wasi-targets.html - FYI for people here tracking Rust toolchain work
Jeff Parsons (Apr 10 2024 at 02:00):Purely out of curiosity, do you know if anybody is blogging/Tweeting/Tooting about the implementation details of Rust toolchain work as they go? I'm particularly interested in how much y'all expect to build upon wasi-libc vs. implement directly against the WASI ABI, and how the nascent async work is likely to interact with this — e.g. everything stays on forks for now, a separate wasm32-wasip3 target ASAP, or somehow allowing tinkering with async on the wasip2 target?
Jeff Parsons (Apr 10 2024 at 02:00)::pray::man_bowing:
Joel Dice (Apr 10 2024 at 14:20):FWIW, I very much expect Rust's std will continue to build upon wasi-libc. It can't use the WASI ABI directly without sacrificing interoperability with C code, since wasi-libc is the one responsible for mapping POSIX-style file descriptors to WASI resource handles. If Rust's std tried to do that itself, it would end up with a mapping that's incompatible with wasi-libc's , preventing Rust and C code from passing descriptors back and forth.
Yoshua Wuyts (Sep 06 2024 at 17:43):Oh, someone on Reddit asked a really good question about how to migrate existing cfg's targeting target_os = "wasi" to continue working on the new target: https://www.reddit.com/r/rust/comments/1f9qbpv/comment/llqyh9i/
I responded, but I'm wondering if perhaps we should file a PR to the announcement post to include this info?
Yoshua Wuyts (Sep 06 2024 at 17:44):This seems something other folks will start running into as well
Alex Crichton (Sep 06 2024 at 17:48):if backwards compat is needed then target_os = "wasi", not(target_env = "p2") will suffice
Alex Crichton (Sep 06 2024 at 17:48):that's, however, not necessarily future-compatible with a p3, so there's a tradeoff
Yoshua Wuyts (Sep 06 2024 at 17:49):Alex Crichton said:
if backwards compat is needed then
target_os = "wasi", not(target_env = "p2")will suffice
wait, can you explain why that's better than passing target_env = "p1" ?
Yoshua Wuyts (Sep 06 2024 at 17:49):I assume there's a reason, but not sure I get it?
Alex Crichton (Sep 06 2024 at 17:50):oh the difference is that target_env = "p1" was added recently, something like 1.80.0 IIRC
Yoshua Wuyts (Sep 06 2024 at 17:50):oh like that!
Alex Crichton (Sep 06 2024 at 17:50):so if a crate wants to support rustc 1.79.0 and prior with p1 you can't rely on target_env = "p1"
Alex Crichton (Sep 06 2024 at 17:50):the "official" way to select p1, however, is target_env = "p1" for sure
Yoshua Wuyts (Sep 06 2024 at 17:53):wait, can't we just tell people to do
Yoshua Wuyts (Sep 06 2024 at 17:53):target_os = "wasi", not(target_env = "p2"), not(target_env = "p3")
for like, Ultimate Backwards Compat
Yoshua Wuyts (Sep 06 2024 at 17:54):(at the cost of readability I guess)
Yoshua Wuyts (Sep 06 2024 at 17:56):(updated the comment to include the line about backwards-compat)
Alex Crichton (Sep 06 2024 at 18:29):oh I wouldn't recommend that
Alex Crichton (Sep 06 2024 at 18:30):mostly b/c target_env = "p3" doesn't actually exist yet
Alex Crichton (Sep 06 2024 at 18:30):and it also doesn't include a theoretical target_env = "p4"
Last updated: Dec 23 2024 at 13:07 UTC