Cache accessibe symbol chains and serialized type parameter name generation #43973
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
typescript-bot
added
Author: Team
For Milestone Bug
weswigham
commented
May 5, 2021
src/compiler/checker.ts
Outdated
| (context.typeParameterNamesByText || (context.typeParameterNamesByText = new Set())).add(result.escapedText as string); | ||
| // avoiding iterations of the above loop turns out to be worth it when `i` starts to get large, so we cache the max | ||
| // `i` we've used thus far, to save work later | ||
| (context.typeParameterNamesByTextNextNameCount ||= new Map()).set(rawtext, i); |
There was a problem hiding this comment.
This is the "cache serialized type parameter name generation" part - this additional tiny bit of caching brought the runtime of the example down from 7s to 5.5s on my machine.
| const firstRelevantLocation = forEachSymbolTableInScope(enclosingDeclaration, (_, __, node) => node); | ||
| const key = `${useOnlyExternalAliasing ? 0 : 1}|${firstRelevantLocation && getNodeId(firstRelevantLocation)}|${meaning}`; | ||
| if (cache.has(key)) { | ||
| return cache.get(key); |
There was a problem hiding this comment.
This cache could probably be better - intelligently sharing results for commonish-meanings and having a hierarchical cache both sound nice in theory, but are a pain to implement performantly (the last time I tried I ruined perf in the general case). This relatively simple cache was sufficient to bring the example runtime down from 20s to 7s on my machine, though is likely to only really effect really degenerate cases, like the one in the linked issue.
| const preProgram = ts.length(rootFiles) < 100 ? ts.createProgram(rootFiles || [], { ...compilerOptions, configFile: compilerOptions.configFile, traceResolution: false }, host) : undefined; | ||
| // pre-emit/post-emit error comparison requires declaration emit twice, which can be slow. If it's unlikely to flag any error consistency issues | ||
| // and if the test is running `skipLibCheck` - an indicator that we want the tets to run quickly - skip the before/after error comparison, too | ||
| const skipErrorComparison = ts.length(rootFiles) >= 100 || (!!compilerOptions.skipLibCheck && !!compilerOptions.declaration); |
There was a problem hiding this comment.
Declaration emit for this test still takes 5s on my machine. getPreEmitDiagnostics invokes declaration emit to get declaration emit errors (for historical reasons they've been considered "preEmit" errors), meaning declaration emit had to run twice, so 10s. Single-threaded, this was fine, but in high resource contention scenarios (ie, runtests-parallel) this was long enough to stretch to 40s and timeout. So I added an opt-out case to the preEmit/preEmit-after-emit error consistency check to keep this test runtime low (so it could pass without a timeout when running in parallel, and probably on CI).
| @@ -0,0 +1,34 @@ | |||
| // @declaration: true | |||
| // @skipLibCheck: true | |||
| // @noTypesAndSymbols: true | |||
There was a problem hiding this comment.
And I've skipped the type/symbol baselines for this test, since they double again the test runtime (and also thus cause a timeout) by essentially invoking the same logic as declaration emit for no real additional gain (compared to the declaration emit we're already running).
|
@typescript-bot pack this |
|
Heya @weswigham, I've started to run the tarball bundle task on this PR at 0f4e922. You can monitor the build here. |
|
Heya @weswigham, I've started to run the perf test suite on this PR at 0f4e922. You can monitor the build here. Update: The results are in! |
|
Hey @weswigham, I've packed this into an installable tgz. You can install it for testing by referencing it in your and then running |
sheetalkamat
reviewed
May 5, 2021
| @@ -3935,12 +3935,12 @@ namespace ts { | |||
| return typeCopy; | |||
| } | |||
|
|
|||
| function forEachSymbolTableInScope<T>(enclosingDeclaration: Node | undefined, callback: (symbolTable: SymbolTable) => T): T { | |||
| function forEachSymbolTableInScope<T>(enclosingDeclaration: Node | undefined, callback: (symbolTable: SymbolTable, ignoreQualification?: boolean, scopeNode?: Node) => T): T { | |||
There was a problem hiding this comment.
I dont see anywhere we are passing ignoreQualification true or false ?
Also can we make these parameter nonOptional but type | undefined instead to ensure every call back considers what needs to be passed?
There was a problem hiding this comment.
Inside this function, it's not; however the function we often pass as a callback uses it as a second parameter (and it's optional there) for its' other callsites, so this optional parameter is here so the callback parameter is compatible with both that usage and the node-finding usage.
|
@weswigham Here they are:Comparison Report - master..43973
System
Hosts
Scenarios
Developer Information: |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
-3-5% check time in |
dmichon-msft
reviewed
May 6, 2021
dmichon-msft
reviewed
May 6, 2021
|
@typescript-bot pack this 52499b1 added another layer of caching, this one within |
|
Heya @weswigham, I've started to run the perf test suite on this PR at 909d519. You can monitor the build here. Update: The results are in! |
|
Heya @weswigham, I've started to run the tarball bundle task on this PR at 909d519. You can monitor the build here. |
|
Hey @weswigham, I've packed this into an installable tgz. You can install it for testing by referencing it in your and then running |
|
@weswigham Here they are:Comparison Report - master..43973
System
Hosts
Scenarios
Developer Information: |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
andrewbranch
approved these changes
May 11, 2021
| @@ -4479,7 +4489,7 @@ namespace ts { | |||
| enclosingDeclaration, | |||
| flags: flags || NodeBuilderFlags.None, | |||
| // If no full tracker is provided, fake up a dummy one with a basic limited-functionality moduleResolverHost | |||
| tracker: tracker && tracker.trackSymbol ? tracker : { trackSymbol: noop, moduleResolverHost: flags! & NodeBuilderFlags.DoNotIncludeSymbolChain ? { | |||
| tracker: tracker && tracker.trackSymbol ? tracker : { trackSymbol: () => false, moduleResolverHost: flags! & NodeBuilderFlags.DoNotIncludeSymbolChain ? { | |||
There was a problem hiding this comment.
There exists a ts.returnFalse 🙃
There was a problem hiding this comment.
Eh, while I understand we use noop to reduce memory load, this fallback tracker (along with the other host objects I adjusted) is constructed infrequently enough that it doesn't really matter, I think.
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.
Which can improve declaration emit performance significantly in some edge cases. Together, these caches bring the example in the linked issue down from 20s to 5.5s on my machine, which is a pretty marked improvement.
Fixes #42937