[nll] teach SCC about 'static
#53327
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
rust-highfive
added
the
S-waiting-on-review
|
The job Click to expand the log.I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact |
wesleywiser
commented
Aug 16, 2018
| None | ||
| }; | ||
|
|
||
| let normal_successors = self.graph.outgoing_regions(node).collect::<Vec<_>>(); |
There was a problem hiding this comment.
I had to .collect() here to work around a lifetime issue.
wesleywiser
commented
Aug 16, 2018
|
|
||
| impl<'s, 'graph> graph::GraphSuccessors<'graph> for AugmentedRegionGraph<'s> { | ||
| type Item = RegionVid; | ||
| type Iter = Box<dyn Iterator<Item = Self::Item>>; |
There was a problem hiding this comment.
I couldn't figure out how to name this type so I just boxed it. Should I hand roll a custom Iterator instead?
|
cc #53178 |
|
@bors try |
|
⌛ Trying commit 6e918df9c2afcdd4f71c364c27a0e7d5e3710279 with merge ed285b7a46c0949465c4c1af1d968de39cc1dbbc... |
|
☀️ Test successful - status-travis |
|
@rust-timer build ed285b7a46c0949465c4c1af1d968de39cc1dbbc |
|
Success: Queued ed285b7a46c0949465c4c1af1d968de39cc1dbbc with parent 5050349, comparison URL. |
kennytm
added
S-waiting-on-perf
|
(FYI perf is ready) |
|
It seems like the big win here is max-rss:
though instructions win too
|
nikomatsakis
force-pushed
the
wip_optimize_nll
branch
2 times, most recently
from
f8be3e8 to
c2ca559
Compare
|
OK, I pushed some commits fixing the nits that @wesleywiser was hitting |
nikomatsakis
approved these changes
Aug 22, 2018
|
@wesleywiser take a look at the changes I pushed; it seems like the only thing holding this up from landing is possible changes to the error messages? Can you run |
nikomatsakis
added
S-waiting-on-author
|
The job Click to expand the log.I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact |
wesleywiser
commented
Aug 22, 2018
| let static_successors: Range<RegionVid> = if node == self.static_region { | ||
| // `'static: 'a` for all regions `'a` -- so add an edge | ||
| // from `'static` to every other region. | ||
| 0.into() .. self.graph.constraint_graph.first_constraints.len().into() |
There was a problem hiding this comment.
0.into() could just be self.static_region
| LL | f(x) | ||
| | ^^^^ `x` escapes the function body here | ||
| LL | let f: fn(_) -> _ = foo; | ||
| | ^^^ cast requires that `'a` must outlive `'static` |
There was a problem hiding this comment.
Hmm, these errors do look less good, but they are also not wrong; I certainly see how they arise.
There was a problem hiding this comment.
This is a tricky case, but I think if we provide the suggested output in the other comment it would be slightly more understandable. Something closer to the ideal output in my mind would be:
error: unsatisfied lifetime constraints
--> $DIR/mir_check_cast_reify.rs:46:25
|
LL | fn foo<'a>(x: &'a u32) -> &'a u32
| --------------------------------- `foo` is not `'static` due to this definition
...
LL | fn bar<'a>(x: &'a u32) -> &'static u32 {
| -- --- -------- `foo` needs to be `'static` due to this return type
| | |
| | `x` is `'a` due to this
| lifetime `'a` defined here
...
LL | let f: fn(_) -> _ = foo;
| ^^^ this is `'a` but it must outlive `'static`
| | -- lifetime `'a` defined here | ||
| LL | //~^ ERROR unsatisfied lifetime constraints | ||
| LL | x | ||
| | ^ cast requires that `'a` must outlive `'static` |
There was a problem hiding this comment.
in this case I sort of prefer the error =)
There was a problem hiding this comment.
but I suppose it's mostly the crazy span; I think we are labeling the wrong point
There was a problem hiding this comment.
"cast requires..." is not a great label, but the spans are correct in my mind, although I would try and detect this kind of case and point at the return type:
error: unsatisfied lifetime constraints
--> $DIR/mir_check_cast_unsize.rs:19:5
|
LL | fn bar<'a>(x: &'a u32) -> &'static dyn Debug {
| -- --- -------- `x` must be `'static` due to this return type
| | |
| | `x` is `'a` due to this type definition
| lifetime `'a` defined here
LL | x
| ^ this is `'a` but it must outlive `'static`
|
cc @matthewjasper @davidtwco @estebank — expanding the definition of SCC here is messing with our error messages a bit. But then again these have always been heuristics. I am sort of inclined to land this PR "as is" and try to address in follow-up changes to the heuristics -- but I'm curious what y'all think. I'm not really sure (in any case) what those changes ought to be. The problem here remains that it's hard to separate out the outlives edges that users want to be shown; using the SCC rules seemed to do a decent job of screening out a lot of "junk" at the bottom of the graph, but it's imperfect, and there ought to be a better basis somehow. |
|
All of the cases where we've changed messages are for conflicting files, I'm not sure if there will be any changes after a rebase. |
|
I'll rebase tonight and we can see if that fixes the issues or not. |
@nikomatsakis I agree, lets merge and file tickets. From what I'm seeing there are two cases, and one is a special case of the other, which means that improving the more general case would make both better, and we can take our time to fix the latter. |
|
Actually, that's wrong. |
It is the |
|
Ok, I was totally off =) the region is not early-bound here but rather late-bound, and this is the code path that is returning invalid data: |
|
PS @varkor I'm sorry I cited you, your impeccable refactorings are definitely not the problem =) |
|
OK, so, I'm not sure the best fix, but the problem has to do with this code: rust/src/librustc_mir/borrow_check/nll/universal_regions.rs Lines 701 to 721 in 35a5541 Here we are walking over the hir-id for each late-bound region on the function. We are particularly interested in late-bound regions that are declared but not used -- in fact, we only care here about late-bound regions that have been give names (this code is needed because people may refer to those names elsewhere, and we used to get ICEs if those names were not used somewhere in the function). The problem is that the late-bound region here is anonymous; somehow the node-id we get for it winds up mapping to |
|
Thanks for your help @nikomatsakis! I'll rebase tonight and fix the merge conflicts with the tests. |
|
@bors p=1 RC milestone |
wesleywiser
force-pushed
the
wip_optimize_nll
branch
from
1e55c80 to
b1211e8
Compare
|
Rebased |
|
This is ready for review. |
kennytm
added
S-waiting-on-review
|
@bors r+ It would be nice to improve the heuristics, though. |
|
📌 Commit b1211e8 has been approved by |
bors
added
S-waiting-on-bors
bors
added a commit
that referenced
this pull request
Sep 7, 2018
[nll] teach SCC about `'static` r? @nikomatsakis I think this is right? I am seeing better performance on the `html5ever` benchmark but I'd like a perf run to quantify the exact speedup. There's a few ui tests failing due to changes in the error messages. The main issue seems to be that returns aren't being detected correctly? `mir_check_cast_unsize.rs` before: ``` error: unsatisfied lifetime constraints --> mir_check_cast_unsize.rs:17:46 | 17 | fn bar<'a>(x: &'a u32) -> &'static dyn Debug { | ________--____________________________________^ | | | | | lifetime `'a` defined here 18 | | //~^ ERROR unsatisfied lifetime constraints 19 | | x 20 | | //~^ WARNING not reporting region error due to nll 21 | | } | |_^ return requires that `'a` must outlive `'static` ``` `mir_check_cast_unsize.rs` after: ``` error: unsatisfied lifetime constraints --> mir_check_cast_unsize.rs:19:5 | 17 | fn bar<'a>(x: &'a u32) -> &'static dyn Debug { | -- lifetime `'a` defined here 18 | //~^ ERROR unsatisfied lifetime constraints 19 | x | ^ cast requires that `'a` must outlive `'static` ```
|
☀️ Test successful - status-appveyor, status-travis |
This was referenced Sep 7, 2018
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
r? @nikomatsakis
I think this is right? I am seeing better performance on the
html5everbenchmark but I'd like a perf run to quantify the exact speedup. There's a few ui tests failing due to changes in the error messages. The main issue seems to be that returns aren't being detected correctly?mir_check_cast_unsize.rsbefore:mir_check_cast_unsize.rsafter: