ICE when existential type is exported from one crate and used in another #53443
Comments
|
Oops. Yea. I assumed this would work since |
oli-obk
added
I-ICE
|
Is the fix just to have rust/src/librustc/middle/resolve_lifetime.rs Line 630 in b559042 where the If you think the fix for this ICE is just that one |
|
I think they are closely related. We should probably run {
intravisit::walk_ty(self, ty);
return;
}in case |
|
My real code's existential types do not have lifetimes, though they do have pub existential type GetResponse: Future<Output = crate::Result<crate::Mod>> + 'static;
pub existential type DownloadResponse: Stream<Item = crate::Result<crate::reqwest_async::Chunk>> + 'static;Adding The concrete type for the first one is compiler-created because it's the return type of an async block. pub fn get(&self, mod_name: &factorio_mods_common::ModName) -> GetResponse {
...
async {
....
}
}I believe this is sugar for The second one's concrete type is my own Stream impl that also does not have lifetimes. It'll take me some time to get a repro because something is broken with the rustc I built locally (it started hitting #53469 even though it was fine before, not sure why). If you want you can find the real code at https://github.com/Arnavion/fac-rs/tree/3a6afed4. The existential types are in the |
Rust + factorio? <3
the issue you are hitting is that a |
|
@oli-obk The #53469 I started hitting was because of two different versions of libsyntax_pos in stage1 builds, as I described here. So with stage2 builds I don't have that problem, So I've got a repro for the issue where
extern crate a;
struct Bar(a::Foo);
fn main() {
}ICE |
|
Your suggestion did fix that error by the way. --- a/src/librustc/middle/resolve_lifetime.rs
+++ b/src/librustc/middle/resolve_lifetime.rs
@@ -623,7 +623,13 @@ impl<'a, 'tcx> Visitor<'tcx> for LifetimeContext<'a, 'tcx> {
}
hir::TyKind::Path(hir::QPath::Resolved(None, ref path)) => {
if let Def::Existential(exist_ty_did) = path.def {
- let id = self.tcx.hir.as_local_node_id(exist_ty_did).unwrap();
+ let id = match self.tcx.hir.as_local_node_id(exist_ty_did) {
+ Some(id) => id,
+ None => {
+ intravisit::walk_ty(self, ty);
+ return;
+ },
+ };
// Resolve the lifetimes in the bounds to the lifetime defs in the generics.
// `fn foo<'a>() -> impl MyTrait<'a> { ... }` desugars to |
|
With the above patch, there is a new ICE when using the field inside an inherent method. extern crate a;
struct Bar(a::Foo);
impl Bar {
fn zero(&self) -> &a::Foo {
&self.0
}
}
fn main() {
}ICE |
|
For me another variant of this issues is blocker:
#![feature(existential_type)]
pub trait View {
type Tmp: Iterator<Item = u32>;
fn test(&self) -> Self::Tmp;
}
pub struct X;
impl View for X {
existential type Tmp: Iterator<Item = u32>;
fn test(&self) -> Self::Tmp {
vec![1,2,3].into_iter()
}
}and extern crate a;
use a::View;
fn main() {
let v = a::X;
v.test();
}ICE
|
|
Is there any progress on this? |
|
@Arnavion how do you feel about making a PR with your patch that has a partial fix? It looks like that would be enough to unblock me from using existential types in my project. |
|
A bunch of stuff changed with oli-obk's recent change so the second patch won't apply directly. I'll retest and put up a fork. |
|
For the first ICE (function that returns existential type in crate A called from crate B), the patch (on top of 46880f4) is diff --git a/src/librustc_resolve/build_reduced_graph.rs b/src/librustc_resolve/build_reduced_graph.rs
index 25a7ff9cd3..02a28a1e2f 100644
--- a/src/librustc_resolve/build_reduced_graph.rs
+++ b/src/librustc_resolve/build_reduced_graph.rs
@@ -736,6 +736,9 @@ impl<'a, 'cl> Resolver<'a, 'cl> {
Def::Macro(..) => {
self.define(parent, ident, MacroNS, (def, vis, DUMMY_SP, expansion));
}
+ Def::Existential(..) => {
+ self.define(parent, ident, TypeNS, (def, vis, DUMMY_SP, expansion));
+ }
_ => bug!("unexpected definition: {:?}", def)
}
}The second ICE (existential type exported from crate A used as struct member in crate B) is already fixed. The third ICE (existential type exported from crate A used in return type of a function in crate B) still exists. And, while I'm at it, @andreytkachenko's ICE also still exists. |
|
@Arnavion thanks so much for following up on this! Do you want to open a PR, since you've done all the work? 😁 |
|
The patch is just guesswork on my part, and the whole thing is still broken anyway, so there's no point to me making a PR. If you want to try it you can apply that patch locally. |
|
Just encountered an ICE and searching led me here. Is this just another instance of the same bug or should I report it separately? Run |
|
I'm running into this with ICE |
|
@jonhoo Yes. If you see the "Easy playground repro" link in the OP, that happens because the doctest is a separate crate. |
Sadly these do not work yet due to the ICE from rust-lang/rust#53443
|
Sorry about taking so long, it was just a few small fixes in the end. I think #57836 fixes all the ICEs mentioned in this issue. |
|
Thanks. I confirmed this fixed all the issues with my original project too. |
Sadly these do not work yet due to the ICE from rust-lang/rust#53443
a/Cargo.tomla/src/lib.rsb/Cargo.tomlb/src/main.rsICE
rust/src/librustc_resolve/build_reduced_graph.rs
Line 741 in b559042
Resolver::build_reduced_graph_for_external_crate_defneeds an arm forDef::Existential? @oli-obkEdit: Easy playground repro
The text was updated successfully, but these errors were encountered: