feat(linter): bulk suppression by Afsoon · Pull Request #19328 · oxc-project/oxc

Afsoon · 2026-02-12T13:12:34Z

AI Disclosure

I have used AI to:

Fix all my grammar mistakes in this PR description, as English isn't my first language.
Iterate on some ideas that I had during development, for example whether it is better to modify the concurrent hash map or send diffs through a channel.

Apart from these situations, the code has been written without AI assistance. As mentioned in my first PR in Oxlint, these contributions help improve Oxlint while at the same time helping me improve my Rust skills.

Description

This PR implements bulk-suppressions for Oxlint. I had in mind the goals mentioned in #10549 (comment)

Implementation details

It is split into sections with all the decisions that I have taken to implement this PR.

Suppression Manager

The code is located in crates/oxc_linter/src/suppression/mod.rs and it is split into three files:

tracking.rs: Contains all the code that handles the loading, saving, and updating of oxlint-suppressions.json.
mod.rs: Implements the SuppressionManager. It also acts as a middleman between oxlint and tsgolint and tracking to update the information, and diff to execute the diff operations.
diff.rs: Hides all the complexity of how the diff works from the rest of the project.

Filtering Diagnostics

The Suppression Manager only filters diagnostics, that have a valid OxcCode, otherwise the message is ignored as we can't scope it to a rule or plugin.

The messages are only filtered if the new count is lower or equal than the registered one; otherwise, they are displayed. Same behavior as ESLint.

Sharing and updating the data

Oxlint uses rayon::scope to create as many workers as there are available threads. Those workers do the heavy linting work, and this is the data that I need for filtering or creating the file.

The semantics of Send + Sync in Rust make it hard, for good reason, to share information between threads. So the quick approach is for each worker to read the whole file just to pick a subset of the data. On top of that, the feature is probably more oriented toward medium/large projects, so this approach doesn't scale.

So the implemented approach is the following:

SuppressionManager ~~duplicates the data into a concurrent hash map. This map is safe to be shared between threads and is only used in read mode.~~ Update: The data is wrapped in a Arc instead of duplicating it. It reduces a level of indirection. Thanks @wagenet for the suggestion.
Each worker holds a reference to a DiffManager, this struct contains all data necessary for his operations.
The file is linted. The errors emitted by the lint are used to track the number of detected errors.
DiffManager calculate a diff based on the args provided and if bulk-suppressions file exists:
- If we have diffs and no CLI args are provided, then the diffs are converted to OxcDiagnostic to be reported.
- If we have diffs and at least one of the CLI args is provided, then the diffs are sent to the parent through a channel.

Once the linting is completed and we are back in the main thread, we consume the channel with the diffs sent by the workers. If we are in update or create mode and no diffs have been emitted, then we do nothing. Otherwise, we mutate the map in the SuppressionManager with the operations sent by the workers and save it.

In a few words, I'm trading memory usage for what is, in my opinion, a simpler process. The option that I discarded was relying on interior mutability, but that would mean ensuring that all operations are safe and that no worker panics.

TSGo lint errors are the only recorded errors.

Rather than completely changing how fixable and non-fixable errors are emitted, I have attached to the existing flow. Type errors and related errors are ignored and not recorded.

CLI Args

Currently, both commands update the file, no matter the diff found. The only situation that doesn't work the same is when the file oxlint-suppressions.json does not exist. The only way to create it is by running oxlint --suppress-all.

Comments

~130 files from this PR are file in the fixtures folders.
The current state of the PR handle lint errors emited by Oxlint, JSPlugin and TSGO.
All suppression test fixtures are self contained in the suppression folder.
Only the Default Output is rendered if the file has been created or updated.
~~--fix or related commands do not update the file.~~ UPDATE: The fix commands are handle.
I have manually tested the feature in PostHog and in https://github.com/TheAlexLichter/oxlint-react-compiler-rules
Added a SuppressionTest that implements Drop to handle the cleanup of this tests.
Thanks @Sysix for you manually test in Windows, and mentioned how to run the test in Windows using the CI.
Windows CI are green https://github.com/oxc-project/oxc/actions/runs/22840798852.

PostHog

Oxlint React Compiler Rules

Fixes

Support ESLint-style bulk suppression via suppressions file #10549

codspeed-hq · 2026-02-12T13:18:21Z

Merging this PR will not alter performance

✅ 4 untouched benchmarks
⏩ 47 skipped benchmarks¹

_{Comparing Afsoon:feat(oxlint)-bulk-suppression (a58c1f9) with main (59e1091)}

47 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

camc314 · 2026-02-14T10:34:31Z

Amazing work! Going to look at this on monday (or tomorow if i have time)

camc314

We should also add tests for the following scenarios:

supressions with cutom plugins
supression with an invalid (malformed) supression file
combining --fix with --suppress--all to make sure that we end up with the correct supression count
Not sure how we can test this, but have you considered/thought about how we're going to handle paths on different machines? e.g. windows/Mac path separator differences

camc314 · 2026-02-20T15:58:33Z

crates/oxc_linter/src/service/runtime.rs

+            self.linter.options.suppress_all,
+            self.linter.options.prune_suppressions,
+        )
+        .unwrap();


I don't think it's safe to unwrap here in the case that the file is invalid - we probably want to pass this as an option to runtime::new or linter::new, and handle the failure gracefully with an error message

What do you think of this approach:

apps/oxlint: Try to load the file using the suppression_manager. If nothing goes wrong, a Some or None is returned. If something goes wrong, then an error is generated, but the lint works as expected. This logic will replace https://github.com/oxc-project/oxc/pull/19328/changes#diff-fd61a8cebacde1fbb072652808cc144fb381a435c98434693779485de0bdf8a2R418-R434.

The data needed by the lint is drilled through the argumens from here. apps/oxlint -> LintRunner::lint_files ->LintService::run -> Runtime::run

The OutputFormatter based on the state the suppression_manager has, will output the current messages, added OxlintSuppressionFileAction::MalformedFile status.

The other approach that comes to my mind after send the previous comments is the following. In LinterRunner::lint_files try to read the file, if it's malformed then lint the files, and return an Err("oxlint-suppression.json is malformed.") instead of Ok(self)

camc314 · 2026-02-24T09:03:52Z

crates/oxc_linter/src/lib.rs

        allocator: &'a Allocator,
    ) -> Vec<Message> {
-        self.run_with_disable_directives(path, context_sub_hosts, allocator, None).0
+        //TODO Inside LintRunner::run, we should check self.options.suppression_options, to handle each setting correctly


Suggested change

//TODO Inside LintRunner::run, we should check self.options.suppression_options, to handle each setting correctly

// TODO Inside LintRunner::run, we should check self.options.suppression_options, to handle each setting correctly

Do we need to action this before this PR is merged?

What is the correct handling here?

This is a legacy comment that I forget to remove. This was an early assumption that was partially incorrect.

camc314 · 2026-02-24T09:10:10Z

crates/oxc_linter/src/fixer/mod.rs


 #[derive(Debug, Clone, Eq, PartialEq)]
 pub struct Message {
+    pub suppression_id: Option<SuppressionId>,


isn't this information available inside error (oxc_diagnostic) ? We should have the error scope/error code to use?

This shuold help us to avoid some excess allocations cloning rule name/rule id.

The main regressions in performance appears to be as a result of these extra allocations

I have updated the code to use OxcCode from the message, and none of the test are broken.

Confirmed. No more performance regression detected

camc314 · 2026-02-24T09:16:43Z

crates/oxc_linter/src/service/runtime.rs

                                context_sub_hosts,
                                allocator_guard,
                                me.js_allocator_pool(),
+                                &suppression_file,


it might be worth making this an Option so that we don't need to create the supression file structure in the commen case where it isn't being used?

Afsoon · 2026-02-24T10:36:24Z

supressions with cutom plugins

Correct me If I'm wrong, but I think this test case is covered here https://github.com/oxc-project/oxc/pull/19328/changes#diff-d6f62b74c4fcbe284a4c8b21f90cdf6200f2d41ecde269fcfeefd0a4d3c71174

supression with an invalid (malformed) supression file

combining --fix with --suppress--all to make sure that we end up with the correct supression count

Noted it.

Not sure how we can test this, but have you considered/thought about how we're going to handle paths on different machines? e.g. windows/Mac path separator differences

My idea was to normalize the path as POSIX format, and reading the code I think I missed it this part. We don't need a full solution to handle the differences between Windows and Linux and Mac, our routes are relatives to CWD. I'm thinking conditional test that runs only in Windows and does:

Reading an existing file.
Creating a file with paths in POSIX format.
Update a file that keeps the paths in POSIX format.

This afternoon I will have a computer with Windows to be able to manually test it.

If you see it's good enough @camc314, I can convert the PR in draft. Once I have all the changes and new tests, I will mark it as ready and ping you? Also this week I'm a bit busy at work, and in the weekend I'm out.

Afsoon · 2026-02-25T07:49:00Z

I will set the PR to draft.

combining --fix with --suppress--all to make sure that we end up with the correct supression count

This test is failing because I'm filtering the messages with fix, no matter if the fix option is turned on. Probably I will add test for fix-suggestion and fix-dangerously

Not sure how we can test this, but have you considered/thought about how we're going to handle paths on different machines? e.g. windows/Mac path separator differences

I will install a VM, because the manual test is taking forever in the computer they have lend me.

PR comments

I have applied all PR comments. Once the previous points are solved, I will refactor the code again, mostly to improve readibility.

Two targeted changes to the bulk-suppression machinery introduced in oxc-project#19328 (oxlint-suppressions.json support). **Change 1 — Eliminate the startup clone in `concurrent_map()`** `SuppressionManager::concurrent_map()` previously constructed a `papaya::HashMap` by iterating the suppression file and cloning every key and value into it. Workers only ever *read* one entry per file, so the copy was pure overhead: ~10–15 ms sequential startup cost for 10k files / 20k suppressions. Fix: wrap `AllSuppressionsMap` in `Arc` at load time. `concurrent_map()` now calls `Arc::clone()` (a reference-count bump), making the hand-off to rayon workers zero-copy. Side effect: the `papaya` dependency can be removed from the worker path in `service/runtime.rs`; `papaya::HashMap::pin()` is replaced by a plain `FxHashMap::get`. The `&SuppressionFile::new(…)` dangling- reference pattern is also fixed while touching that code. **Change 2 — Reduce per-message `RuleName` allocations** `build_suppression_map` called `message.into()` twice per message (once for `get_mut`, once for `insert`). Replaced with `.entry().or_insert()` so the key is computed once. The filter closure in `suppress_lint_diagnostics` also called `message.into()` twice per message (two separate `.get()` calls). Replaced with `let key = RuleName::from(message)` so both lookups share one allocation. Net result: 1–2 heap allocations per message instead of 3–4. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

- All JSPlugins errors are being recorded in the file. Left Typescript as it required more changes. - Add JSPlugin test - Manual tested the binary in Posthog and oxlint-react-compiler-rules - Improved errors messages, and added a label to suggest the user what command run to update the file. - Some basic message in the cli output if the file is created or updated.

- Remove SuppressionId, this information already exist. Probably the reason of the performance regression. - SuppressionManager inited `apps/oxlint`. This replace the logic if the file has been created or updated. Pending the state if a diff haven’t been found, this will avoid render the file has been updated when it isn’t. - Integration test for malformed json. The error don’t panic, the files are linted but a message is rendered with the error. - `OxcDiagnositc` is transformed to `RuleName`. Pending to use TryFrom instead of From, as it can fail. - Removed misleading TODO comment

- `—fix —suppress-all` test without file not generate a `oxlint-suppressions.json` file and fixes are apply. - `—fix —-suppress-all`test having a initial suppression file, is not behaving correctly as the fix aren’t apply. Are filtered due the suppresion file. Pending to be fixed. - Using Option instead on initiating a default SuppressionFile struct. This solve an edge case of emitting errors when the file was malformed. Pending add more test with a combination of fix commands, a test that prune if only `fix` is provided.

…ify file update if the file has been really updated - Only mentioine the file has been update once the file is write instead of detecting if it exists. - Filtering the messages with the suppression only after of this fix has been apply, not before. - Pending more integration tests handling different of fix

… avoid noise and duplicate code in multiples tests

SuppressionTester struct to handle all the logic of files in the integration tests. - Implemented Drop trait that once the test go out of scope, all the files modified are deleted or replace by their backups. - Test logic hide behind the method `test`, this will do all the need checks to ensure everything works as expected. - Builder patter. to configure if the fixture have an specific configuration of files.

@wagenet

Picked manually the commit cf31c84 oxc-project@cf31c84 by @wagenet. As the papaya wrapper is not necessary at all. Removed the warning import errors and printlns

If an Message isn’t scoped to a plugin and or rule name then is discarded to be tracked.

- Colocated the logic of handle the diff channel in the struct. - A MSCP in the load method and drill proped the sender. This will allow to integrate typescript suppression rules. - No unwrap trying to write the oxlint-suppression file, pending add a message of error.

- Snapshot don’t render anymore the full route only the path relative to CWD. - Asserting parsing the JSON insterad of relying on string to avoid whitespace mismatch.

Now oxlint is filtering, updating or creating a file based on the tsgo diagnostics. An fixture is available to test different scenarios. At this point: - Pruned tsgo diagnostic after it’s fixed witn `—fix`� or `—fix-suggestions` - The diff is a bit more smarter and don’t prune TSGO rules. The other way is pending to be implement. - Diagnostics aren’t send if we are on suppression mod. Otherwise, the errors are immediately reported. Next step is update the fixtures to have TS, and add more specific TS tests. Once it’s done, refactor and rearchitecture a bit the code to reduce code duplicity, hide implementations and normalize to work always with `Message`

All fixtures are executed with type aware and type check to ensure the feature works without issues. Fixed a bug that only occurrs when an user executed oxlint as following `oxlint —fix —suppress-all` AND a suppression file doesn’t exists. Oxlint was unable to create a file because it thought we were in prune mode, instead of suppression after fixing all possible linting errors.

- Added a new SuprresionDiff Struct that abstract all the diff logic, to avoid duplication code and centralized the code. - Manager contains the list of rules linted by TSGo to avoid incorrect pruning rules logic between oxlint and TSGO - The diff handles Message as an input and output. - New test that checks type errors are never recorded. - Remove an unwrap logic inside the update, and hardening the check to avoid remove the last rule in a file in case the diff message doesn’t match with the last rule.

…e semantic name to the test and folders

Afsoon requested a review from camc314 as a code owner February 12, 2026 13:12

github-actions bot added A-linter Area - Linter A-cli Area - CLI A-linter-plugins Area - Linter JS plugins labels Feb 12, 2026

Afsoon changed the title ~~Feat(oxlint): bulk suppression~~ feat(oxlint): bulk suppression Feb 12, 2026

github-actions bot added the C-enhancement Category - New feature or request label Feb 12, 2026

Afsoon force-pushed the feat(oxlint)-bulk-suppression branch from 17607b3 to 506bd73 Compare February 12, 2026 13:21

Afsoon changed the title ~~feat(oxlint): bulk suppression~~ feat(linter): bulk suppression Feb 12, 2026

Afsoon force-pushed the feat(oxlint)-bulk-suppression branch 2 times, most recently from e3d0507 to e9640be Compare February 14, 2026 06:48

Boshen assigned camc314 Feb 14, 2026

Afsoon force-pushed the feat(oxlint)-bulk-suppression branch 2 times, most recently from f989ce1 to 110e14f Compare February 21, 2026 08:56

camc314 reviewed Feb 24, 2026

View reviewed changes

Afsoon marked this pull request as draft February 25, 2026 07:49

Afsoon force-pushed the feat(oxlint)-bulk-suppression branch 2 times, most recently from d155b8b to 4dfc17f Compare February 25, 2026 09:42

wagenet mentioned this pull request Feb 26, 2026

feat(lsp): show bulk-suppressed violations as faded diagnostics #19791

Draft

5 tasks

wagenet mentioned this pull request Feb 26, 2026

perf(linter): reduce bulk-suppression overhead in concurrent path #19792

Draft

2 tasks

Afsoon force-pushed the feat(oxlint)-bulk-suppression branch 6 times, most recently from d3730ac to 1488445 Compare March 9, 2026 05:52

Afsoon and others added 28 commits April 2, 2026 18:17

Phase 0.8.9.5: All test green and remove unused code

5d0d64b

Phase 0.8.9.6: update snapshots test

3308a94

Phase 0.9: Linting warnings fix

a013c49

Suppression test functions that wrap the setup and AAA test logic. To…

8005c59

… avoid noise and duplicate code in multiples tests

More suppression test to handle the different fix situations.

990344f

Perf improvements and warning errors removed

3fd4c09

Picked manually the commit cf31c84 oxc-project@cf31c84 by @wagenet. As the papaya wrapper is not necessary at all. Removed the warning import errors and printlns

[autofix.ci] apply automated fixes

6737da3

Refactor: Using TryFrom instead of From

ab9e326

If an Message isn’t scoped to a plugin and or rule name then is discarded to be tracked.

Clippy warnings

284e596

fix napi snapshot test

569a78c

Normalize path to be always / for windows compatibility

19957da

Fixing test in Windows:

45c4cfc

- Snapshot don’t render anymore the full route only the path relative to CWD. - Asserting parsing the JSON insterad of relying on string to avoid whitespace mismatch.

Uncomment test in windows

462ccf1

Fixing Clippy warnings

0a0f7d9

duplicate fixture folders to avoid false positives

06682e2

Test napi in windows

63e2f84

Revert run Windows CI jobs to run in main

fec8ab4

Moving all integration test to suppression folder and giving it a mor…

4fccda8

…e semantic name to the test and folders

Delete stale snapshopt test

a58c1f9

Afsoon force-pushed the feat(oxlint)-bulk-suppression branch from 188de40 to a58c1f9 Compare April 2, 2026 16:17

camchenry self-requested a review April 3, 2026 03:08

	//TODO Inside LintRunner::run, we should check self.options.suppression_options, to handle each setting correctly
	// TODO Inside LintRunner::run, we should check self.options.suppression_options, to handle each setting correctly

Uh oh!

Conversation

Afsoon commented Feb 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

AI Disclosure

Description

Implementation details

Suppression Manager

Filtering Diagnostics

Sharing and updating the data

TSGo lint errors are the only recorded errors.

CLI Args

Comments

Fixes

Uh oh!

codspeed-hq bot commented Feb 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Footnotes

Uh oh!

camc314 commented Feb 14, 2026

Uh oh!

camc314 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Afsoon Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Afsoon commented Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Afsoon commented Feb 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Afsoon commented Feb 12, 2026 •

edited

Loading

codspeed-hq bot commented Feb 12, 2026 •

edited

Loading

Afsoon Feb 24, 2026 •

edited

Loading

Afsoon commented Feb 24, 2026 •

edited

Loading

Afsoon commented Feb 25, 2026 •

edited

Loading