Repository navigation
rustc 1.100 - #664
rustc 1.100#664Firestar99 wants to merge 7 commits into
Conversation
a09109c to
307eb46
Compare
There was a problem hiding this comment.
nightly-2026-09-06 is the first broken version (#project-const-generics > nightly-2026-09-06 generic const regression), nightly-2026-09-05 is the last one that works
307eb46 to
5d56590
Compare
|
Moved it back by a day. If you're curious how I update toolchains:
|
5d56590 to
b400300
Compare
defc9ac to
92d789f
Compare
92d789f to
5c7b834
Compare
|
Blocker for merge: Not sure if we want to release 1.99 on it's own or merge this and then release. |
|
I hope you release 1.99.0 separately. I'll not be able to upgrade to anything newer for now, so having a choice would be nice. I can use it from Git too, but release is cleaner. |
|
The latest nightly (includes rust-lang/rust#165070) works again with const generics, so 1.101.0 will probably be fine |
| fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>, | ||
| callee: Self::Value, | ||
| fn_val: Self::Value, | ||
| _return_slot: ReturnSlot<Self::Value>, |
There was a problem hiding this comment.
Shouldn't ignore this, and instead assert that it is Direct or whatever value we force.
| } | ||
|
|
||
| fn type_f16b(&self) -> Self::Type { | ||
| bug!("f16b not supported") |
There was a problem hiding this comment.
Interestingly enough, SPV_KHR_bfloat16 adds support somewhat elegantly (an optional extra operand to OpTypeFloat which can indicate non-IEEE encodings).
It might not be difficult to support (but we can do that separately).
There was a problem hiding this comment.
Let's wait until someone is actually asking for them
| # HACK(eddyb) these only exist to unify features across dependency trees, | ||
| # in order to avoid multiple separate instances of `rustc_codegen_spirv`. | ||
| ahash = { version = "0.8.11", features = ["no-rng"] } | ||
| bytemuck = { workspace = true, features = ["aarch64_simd"] } | ||
| log = { version = "0.4.22", features = ["std"] } | ||
| regex = { version = "1", features = ["perf"] } | ||
| rustix = { version = "1.0.8", features = ["all-apis"] } | ||
|
|
There was a problem hiding this comment.
Could this cause a regression in build times? I don't remember the exact combination of build commands that would cause a rebuild without this, wish I had automated it.
There was a problem hiding this comment.
I've noticed that some of these dependencies are completely unused within the entire tree, so I didn't see any point in keeping any of them. See difftests having a lot fewer dependencies, the main Cargo.lock won't see them due to eg. wgpu from examples depending on these. So yes, you may experience more rebuilds from building compiletests and the wgpu example, but for anyone using rust-gpu this is a net benefit as they don't need to compile as many crates, just like our difftests.
Also we should probably migrate to workspace dependencies.
5c7b834 to
8cc47ed
Compare
8cc47ed to
15e676d
Compare
…not being ignorable
15e676d to
00c82c6
Compare
nightly-2026-09-05inspirv-std-nightlycrate to enableadt_const_paramsuse #636 (comment), so we have one commit doing exactly that