Wrap native map to work around 2**24 element limit #50310
Draft
Conversation
This file contains hidden or 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
typescript-bot
added
Author: Team
For Uncommitted Bug
|
@typescript-bot perf test this |
|
Heya @weswigham, I've started to run the perf test suite on this PR at 3448841. You can monitor the build here. Update: The results are in! |
|
@weswigham Here they are:
CompilerComparison Report - main..50310
System
Hosts
Scenarios
TSServerComparison Report - main..50310
System
Hosts
Scenarios
Developer Information: |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
@typescript-bot perf test this (just in case the assertion affects things) |
|
Heya @weswigham, I've started to run the perf test suite on this PR at e65cf19. You can monitor the build here. Update: The results are in! |
|
@weswigham Here they are:
CompilerComparison Report - main..50310
System
Hosts
Scenarios
TSServerComparison Report - main..50310
System
Hosts
Scenarios
Developer Information: |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
@typescript-bot perf test this again lmao |
|
Heya @weswigham, I've started to run the perf test suite on this PR at a54799f. You can monitor the build here. Update: The results are in! |
|
@weswigham Here they are:
CompilerComparison Report - main..50310
System
Hosts
Scenarios
TSServerComparison Report - main..50310
System
Hosts
Scenarios
Developer Information: |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
@ahejlsberg I've added a diagnostic for source elements which trigger more than 1 million comparisons on their own. That number isn't wholly arbitrary - 1 million only takes a few seconds for simple types, which is fast enough that you may not notice the possible perf issue without the error, and is just below how many comparisons two unions of 2^10 elements (1024) would take to compare naively. (Our union element cap is 100,000, so you can go way higher in terms of union size (2^16 is fine, 2^17 is not) - but these large unions are fine to use if and only if they consistently hit our optimized codepaths or aren't compared against one another in the naive quadradic way!) |
weswigham
commented
Aug 19, 2022
| // get better perf, or the constructs in use may be a good candidate for specialized optimizations | ||
| // in the compiler itself to reduce the amount of work required. | ||
| if (assignabilityCacheContribution > 4_000_000) { | ||
| error(node, Diagnostics.This_source_element_is_ultimately_responsible_for_0_million_type_comparisons_It_is_likely_very_slow_and_may_impact_editor_performance_Simplify_the_types_in_use, Math.floor(assignabilityCacheContribution / 1_000_000)); |
There was a problem hiding this comment.
As an aside, since the error is about cache sizes, it's inherently unstable - the first location to do these comparisons will get this error, others that do the same comparisons will not. Historically, location-unstable errors like this haven't been good for incremental build and the like, since the errors shifting mucks with up-to-dateness checks. This could maybe be worked around by indicating this error status in the relationship cache itself (eg, this top-level comparison triggered this overflow, so report the complexity error at every site where this comparison is performed), but given the likely rarity of this error, I'm not sure how worth it that'd be to do, given the complexity that'd entail compared to how simple this is.
There was a problem hiding this comment.
@weswigham I like the detailed explanation, thank you 👍 . I think, acknowledging the instability in cache sizes and its potential impact on incremental builds is crucial. Regarding the error status, I wonder if there's a simpler way to provide consistent feedback for similar comparisons without compromising complexity. Perhaps exploring a middle ground could help maintain simplicity while addressing occasional instability?
weswigham
commented
Aug 19, 2022
| // By manufacturing the fallback in here, we guarantee it has a higher typeid than the bit strings, | ||
| // and thus is sorted to the end of the union, guaranteeing relationship checking passes with a maximal | ||
| // number of comparisons when a naive comparison is done (guaranteeing this test is slow) | ||
| return null as any as Box<ElevenBits | (FallbackPrefix extends never ? never : `${FallbackPrefix}${string}`)>; // return type is a union |
There was a problem hiding this comment.
As a double-aside, while writing this speedrun-to-slow-the-compiler-down test, I've also come up with some heuristics for union comparisons that could probably make this test specifically complete in linear time/comparisons.
Specifically, the "union of specific things with a generalized fallback member" pattern is very common. If we store the first target member that passes and check it first for subsequent source members, it can shortcut iteration in these scenarios and prevent a quadratic search. Similar element-index-carryover could also be used to handle corresponding-unions-but-missing-a-element better, like comparing A | B | D with A | B | C | D, or, more generally, any filtered union with the original. I'll probably put up a PR for that at some point.
There was a problem hiding this comment.
Is that the same / similar to #47511?
There was a problem hiding this comment.
I don't think so, since that's talking about conditional type distribution and I'm just talking about relating two unions. Specifically, in that case, it's about how once you have selected both branch types (assuming they don't use the check type in them), you can stop distributing, since further results won't change the outcome. Similar naive complexity, though.
weswigham
added
the
Experiment
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.
Technically this fixes the crash in #50290, though not the underlying change causing us to make 16M more comparisons (and accompanying worse perf).