-
Notifications
You must be signed in to change notification settings - Fork 372
Overlay: additional feature flags #2967
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
base: main
Are you sure you want to change the base?
Conversation
The loadConfig() function is mostly the same as getDefaultConfig(), except that it calls loadUserConfig() and stores the results in originalUserInput. This refactoring commit replaces the loadConfig() call with getDefaultConfig() and loadUserConfig(), which allows deleting a large amount of duplicated code.
In an upcoming change, getOverlayDatabaseMode() will depend on the contents of Config. As a result, getOverlayDatabaseMode() needs to be called after the rest of Config has already been populated. This commit performs the refactoring to move the getOverlayDatabaseMode() into initConfig(), after the rest of Config has already been populated.
Before we introduce additional features for controlling overlay analysis enablement, change the unit tests to specify features directly instead of through a isFeatureEnabled boolean field.
This change eliminates the need for setup-codeql.test to import from feature-flags.test, which makes the former run all tests defined in the latter.
1ab4d58
to
bb8638d
Compare
bb8638d
to
f3fa160
Compare
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.
Pull Request Overview
This PR extends the overlay analysis feature to support more granular per-language enablement through additional feature flags. Instead of relying on a single overlay_analysis
flag, the action now checks individual language-specific flags and code-scanning-specific flags to determine whether overlay analysis should be enabled.
- Adds 22 new overlay analysis feature flags for language-specific and code-scanning-specific configurations
- Implements chunked API requests to handle the increased number of feature flags (25 features per request)
- Refactors configuration generation to better determine when overlay analysis is appropriate based on the selected languages and configuration
Reviewed Changes
Copilot reviewed 15 out of 22 changed files in this pull request and generated 3 comments.
Show a summary per file
File | Description |
---|---|
src/feature-flags.ts | Adds new overlay analysis feature flags and implements chunked API requests |
src/config-utils.ts | Adds language-specific overlay analysis logic and refactors config generation |
src/testing-utils.ts | Moves utility functions and adds new test helpers for feature flag testing |
src/setup-codeql.test.ts | Updates import to use testing-utils for initializeFeatures |
src/feature-flags.test.ts | Adds test for chunked API requests and moves initializeFeatures |
src/config-utils.test.ts | Adds comprehensive tests for new overlay analysis logic |
src/codeql.ts | Refactors code scanning config generation to use shared utility |
src/trap-caching.ts | Minor import style changes |
while (featuresToRequest.length > 0) { | ||
featureChunks.push(featuresToRequest.splice(0, FEATURES_PER_REQUEST)); |
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.
The splice operation modifies the original array, which could lead to unexpected behavior if featuresToRequest is used elsewhere. Consider using a copy of the array or a different iteration approach.
while (featuresToRequest.length > 0) { | |
featureChunks.push(featuresToRequest.splice(0, FEATURES_PER_REQUEST)); | |
for (let i = 0; i < featuresToRequest.length; i += FEATURES_PER_REQUEST) { | |
featureChunks.push(featuresToRequest.slice(i, i + FEATURES_PER_REQUEST)); |
Copilot uses AI. Check for mistakes.
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.
featuresToRequest
is used only for populating featureChunks
, so it is safe to modify.
@@ -195,7 +196,7 @@ export interface AugmentationProperties { | |||
/** | |||
* Extra query exclusions to append to the config. | |||
*/ | |||
extraQueryExclusions?: ExcludeQueryFilter[]; | |||
extraQueryExclusions: ExcludeQueryFilter[]; |
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.
Changing extraQueryExclusions from optional to required is a breaking change to the interface. This could cause compilation errors for existing code that doesn't provide this property.
extraQueryExclusions: ExcludeQueryFilter[]; | |
extraQueryExclusions?: ExcludeQueryFilter[]; |
Copilot uses AI. Check for mistakes.
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.
All uses of extraQueryExclusions
are in this repository, and they have all been checked and updated. Furthermore, AugmentationProperties
is never deserialized from files, so there is no concern for compatibility with earlier versions.
const remoteFlags = response.data as GitHubFeatureFlagsApiResponse; | ||
.map(([f]) => f); | ||
|
||
const FEATURES_PER_REQUEST = 25; |
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.
The magic number 25 should be defined as a named constant at the module level to improve maintainability and make it easier to adjust if API limits change.
const FEATURES_PER_REQUEST = 25; | |
const FEATURES_PER_REQUEST = FEATURES_PER_REQUEST_LIMIT; |
Copilot uses AI. Check for mistakes.
const feature = OVERLAY_ANALYSIS_FEATURES[language]; | ||
if (feature && (await features.getValue(feature, codeql))) { | ||
continue; | ||
} | ||
const codeScanningFeature = | ||
OVERLAY_ANALYSIS_CODE_SCANNING_FEATURES[language]; | ||
if ( | ||
codeScanningFeature && | ||
(await features.getValue(codeScanningFeature, codeql)) | ||
) { | ||
enableForCodeScanningOnly = true; | ||
continue; | ||
} | ||
return false; |
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.
If I'm reading this correctly, then we won't check the code-scanning variant for a language if the main feature flag for that language is enabled. If, for example, the overlay_analysis_java
flag is already enabled, then toggling overlay_analysis_code_scanning_java
will have no effect. That seems reasonable, but we should take care that anyone enabling feature flags is aware of this.
for (const chunk of featureChunks) { | ||
const response = await getApiClient().request( | ||
"GET /repos/:owner/:repo/code-scanning/codeql-action/features", | ||
{ | ||
owner: this.repositoryNwo.owner, | ||
repo: this.repositoryNwo.repo, | ||
features: chunk.join(","), | ||
}, | ||
); | ||
const chunkFlags = response.data as GitHubFeatureFlagsApiResponse; | ||
remoteFlags = { ...remoteFlags, ...chunkFlags }; | ||
} |
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 chunking seems like the riskiest part of the PR, since it could affect all uses of the Action. I don't see anything wrong with it, and I see you've added a test (with some API-mocking) for it, but do we have a way to see it working in practice before merging/deploying?
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.
All the PR checks run with the changes in this PR, so any outright failure would show up on CI checks.
For reference, the commit without request chunking that I first pushed to this PR had 244 check failures. It was hard to miss.
In addition to passing CI checks, I reran one of the workflows with debug logs. In the "Initialize CodeQL" step, we can see in the debug logs that the action was able to obtain status for all declared features (under "Loaded the following default values for the feature flags from the Code Scanning API"), which lends further support that the request chunking is working as intended.
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 am also wondering what the performance impact of this is (both for users and the backend). If I follow our implementation correctly, we always fetch the values for all FFs that the Action knows of. So that means, once merged, the Action will fetch the status of all of these FFs for every run for everyone. I imagine it's fine if it's one request with a few FFs, but if this now spans multiple requests for every CodeQL analysis, that could quickly add up?
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.
So that means, once merged, the Action will fetch the status of all of these FFs for every run for everyone.
The Action fetches the status of all declared Feature
s for every run for everyone. This PR does not change that behavior—that will continue to be the case even if we do not merge this PR.
What this PR changes is to support fetching feature status in batches of 25 in accordance with the API limit. In practical terms, where "fetching the status of all declared features" used to require 1 API call, now it requires 2 API calls. Each call takes less than 1 second on my laptop, so adding one more call should not have any perceivable impact on user experience. With the volume of Code Scanning runs, I don't think it will have perceivable impact on backend load either.
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.
The Action fetches the status of all declared Features for every run for everyone. This PR does not change that behavior—that will continue to be the case even if we do not merge this PR.
Sorry for the ambiguity in what I wrote. What I meant with "all these FFs" was "all the new FFs introduced by this PR". I am happy for you to keep this as is as long as you're happy that doubling the number of requests to this endpoint isn't a problem.
@mbg I'd also appreciate your review and approval on this PR before I merge. |
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.
LGTM. Just one minor suggestion for the test.
// We only need to call getValue once, and it does not matter which feature | ||
// we ask for. Under the hood, the features library will request all features | ||
// from the API. | ||
const feature = Feature.OverlayAnalysis; |
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.
Given that Feature
elements are intended to be removed when the feature has fully shipped, we would need to update this test when OverlayAnalysis
is removed. How about something like this instead:
const feature = Feature.OverlayAnalysis; | |
const feature = Object.values(Feature)[0]; |
However, I see that we similarly pick specific elements elsewhere in the tests (and that they have to be updated when they are removed). It's your choice if you want to fix this in this case or let us deal with it more generally.
This PR updates the action to check additional per-language features for enabling overlay analysis.
See back-linked issue for additional context.
Merge / deployment checklist