-
Notifications
You must be signed in to change notification settings - Fork 12.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Create const blocks on the fly during typeck instead of waiting until writeback. #125806
Conversation
… writeback. This allows us to spawn a nested FnCtxt, creating a barrier that control flow statements cannot cross.
r? @fee1-dead rustbot has assigned @fee1-dead. Use |
let encl_body_owner_id = self.tcx.hir().enclosing_body_owner(expr.hir_id); | ||
|
||
// If this didn't hold, we would not have to report an error in | ||
// the first place. | ||
assert_ne!(encl_item_id.def_id, encl_body_owner_id); | ||
assert_ne!(encl_item_id.def_id, self.body_id); | ||
|
||
let encl_body = self.tcx.hir().body_owned_by(encl_body_owner_id); | ||
|
||
err.encl_body_span = Some(encl_body.value.span); | ||
err.encl_body_span = Some(self.tcx.def_span(self.body_id)); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This change is because enclosing_body_owner
does not yield const blocks anymore. I'm addressing this separately to remove this footgun, but the code here could have been written simpler anyway, so I just did that.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
enclosing_body_owner
being out of sync is somewhat sketch 💀 why doesn't it? it should be as much of a body as a closure is, for example.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think this change definitely improves things, but I'm still somewhat skeptical that we can't improve it more... Thoughts?
@@ -76,7 +76,12 @@ impl<'hir> LoweringContext<'_, 'hir> { | |||
ExprKind::Array(exprs) => hir::ExprKind::Array(self.lower_exprs(exprs)), | |||
ExprKind::ConstBlock(c) => { | |||
self.has_inline_consts = true; | |||
hir::ExprKind::ConstBlock(self.lower_expr(c)) | |||
self.with_new_scopes(e.span, |this| { | |||
let coroutine_kind = this.coroutine_kind.take(); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Kind of strange that coroutine_kind
is not handled in with_new_scopes
. Is there some other with_body_*
-like function that does the coroutine kind handling too?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have been looking into this in a branch, but didn't want to add more stuff to this PR.
we can get it all working by merging with_new_scopes
into lower_body
, but it needs some managing of the right spans to maintain diagnostic quality.
let encl_body_owner_id = self.tcx.hir().enclosing_body_owner(expr.hir_id); | ||
|
||
// If this didn't hold, we would not have to report an error in | ||
// the first place. | ||
assert_ne!(encl_item_id.def_id, encl_body_owner_id); | ||
assert_ne!(encl_item_id.def_id, self.body_id); | ||
|
||
let encl_body = self.tcx.hir().body_owned_by(encl_body_owner_id); | ||
|
||
err.encl_body_span = Some(encl_body.value.span); | ||
err.encl_body_span = Some(self.tcx.def_span(self.body_id)); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
enclosing_body_owner
being out of sync is somewhat sketch 💀 why doesn't it? it should be as much of a body as a closure is, for example.
let feed = self.tcx().create_def(self.body_id, kw::Empty, DefKind::InlineConst); | ||
feed.def_span(span); | ||
feed.local_def_id_to_hir_id(hir_id); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I feel like const blocks should be first-class nested bodies (liek closures) all the way from the AST-level (i.e. having a def id, being treated as a hir owner, etc) instead of having to synthesize a def-id here.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Perhaps yeah
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
though we'll def need to lazily create the DefId
of anon consts of method calls' const generic params, so that we can feed type_of. That's why I'm trying to figure out how to do this better.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
right now we are breaking query system invariants and are feeding anon const type_of
from typeck
, even though typeck
did not create the DefId
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also, why do we need to feed the type_of
query at all?
The type_of
for a const block should operate like a type_of
for a closure -- for closures, it calls typeck_results
on the parent and extracts it from the expr_ty
:
rust/compiler/rustc_hir_analysis/src/collect/type_of.rs
Lines 483 to 485 in 2a2c29a
Node::Expr(&Expr { kind: ExprKind::Closure { .. }, .. }) => { | |
tcx.typeck(def_id).node_type(hir_id) | |
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
though we'll def need to lazily create the DefId of anon consts of method calls' const generic params, so that we can feed type_of. That's why I'm trying to figure out how to do this better.
I don't understand why this needs to be entangled with inline const exprs. Perhaps we should just distinguish these two rather than treating them the same.
Just gonna revert #124650 and start over |
…r-errors Revert: create const block bodies in typeck via query feeding as per the discussion in rust-lang#125806 (comment) It was a mistake to try to shoehorn const blocks and some specific anon consts into the same box and feed them during typeck. It turned out not simplifying anything (my hope was that we could feed `type_of` to start avoiding the huge HIR matcher, but that didn't work out), but instead making a few things more fragile. reverts the const-block-specific parts of rust-lang#124650 `@bors` rollup=never had a small perf impact previously fixes rust-lang#125846 r? `@compiler-errors`
…r-errors Revert: create const block bodies in typeck via query feeding as per the discussion in rust-lang#125806 (comment) It was a mistake to try to shoehorn const blocks and some specific anon consts into the same box and feed them during typeck. It turned out not simplifying anything (my hope was that we could feed `type_of` to start avoiding the huge HIR matcher, but that didn't work out), but instead making a few things more fragile. reverts the const-block-specific parts of rust-lang#124650 ``@bors`` rollup=never had a small perf impact previously fixes rust-lang#125846 r? ``@compiler-errors``
…r-errors Revert: create const block bodies in typeck via query feeding as per the discussion in rust-lang#125806 (comment) It was a mistake to try to shoehorn const blocks and some specific anon consts into the same box and feed them during typeck. It turned out not simplifying anything (my hope was that we could feed `type_of` to start avoiding the huge HIR matcher, but that didn't work out), but instead making a few things more fragile. reverts the const-block-specific parts of rust-lang#124650 ```@bors``` rollup=never had a small perf impact previously fixes rust-lang#125846 r? ```@compiler-errors```
…errors Revert: create const block bodies in typeck via query feeding as per the discussion in rust-lang#125806 (comment) It was a mistake to try to shoehorn const blocks and some specific anon consts into the same box and feed them during typeck. It turned out not simplifying anything (my hope was that we could feed `type_of` to start avoiding the huge HIR matcher, but that didn't work out), but instead making a few things more fragile. reverts the const-block-specific parts of rust-lang#124650 `@bors` rollup=never had a small perf impact previously fixes rust-lang#125846 r? `@compiler-errors`
…errors Revert: create const block bodies in typeck via query feeding as per the discussion in rust-lang#125806 (comment) It was a mistake to try to shoehorn const blocks and some specific anon consts into the same box and feed them during typeck. It turned out not simplifying anything (my hope was that we could feed `type_of` to start avoiding the huge HIR matcher, but that didn't work out), but instead making a few things more fragile. reverts the const-block-specific parts of rust-lang#124650 `@bors` rollup=never had a small perf impact previously fixes rust-lang#125846 r? `@compiler-errors`
Revert: create const block bodies in typeck via query feeding as per the discussion in rust-lang/rust#125806 (comment) It was a mistake to try to shoehorn const blocks and some specific anon consts into the same box and feed them during typeck. It turned out not simplifying anything (my hope was that we could feed `type_of` to start avoiding the huge HIR matcher, but that didn't work out), but instead making a few things more fragile. reverts the const-block-specific parts of rust-lang/rust#124650 `@bors` rollup=never had a small perf impact previously fixes rust-lang/rust#125846 r? `@compiler-errors`
Revert: create const block bodies in typeck via query feeding as per the discussion in rust-lang/rust#125806 (comment) It was a mistake to try to shoehorn const blocks and some specific anon consts into the same box and feed them during typeck. It turned out not simplifying anything (my hope was that we could feed `type_of` to start avoiding the huge HIR matcher, but that didn't work out), but instead making a few things more fragile. reverts the const-block-specific parts of rust-lang/rust#124650 `@bors` rollup=never had a small perf impact previously fixes rust-lang/rust#125846 r? `@compiler-errors`
This allows us to spawn a nested FnCtxt, creating a barrier that control flow statements cannot cross.
fixes #125670
The issues were caused by #124650 where I failed to account for the lack of changing scopes for const blocks