[lazerg]

BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH isn't in any brotli release yet (1.2.0 is the latest), so building with --shared-brotli fails since 7ba22c2. It's an enum value and the bundled brotli still reports 1.2.0, so there's nothing to #ifdef on. This only exposes the constant when brotli is bundled and skips the matching test otherwise.

I built with --shared-brotli against Homebrew brotli 1.2.0 and ran parallel/test-zlib* (90 tests, all pass). The GN change is untested, I had no gn locally.

Fixes: #66894

[nodejs-github-bot]

Review requested:

• @nodejs/gyp

[github-actions]

Caution

AgentScan found account activity patterns that may be consistent with automation. This is a heuristic, not proof that this pull request was opened by an agent or violates policy. AI-assisted contributions are permitted, but automated tooling must not open pull requests without advance approval, and contributors must personally understand, test, verify, and take responsibility for every submitted change. See the AgentScan analysis, AI use policy, and automation policy for additional context.

[aduh95]

Suggested change

	}, {
	'defines': [ 'NODE_SHARED_BROTLI' ],

[lazerg]

✅ Done

[aduh95]

Can you remove the cherry-pick to validate it does fix this build and still pass the tests

node/tools/nix/sharedLibDeps.nix

Lines 25 to 32 in 697ecce

	brotli = pkgs.brotli.overrideAttrs (old: {
	patches = (old.patches or [ ]) ++ [
	(pkgs.fetchpatch2 {
	url = "https://github.com/google/brotli/commit/0ed4268a4401abcb4472096340477cf4af7d1a2b.patch?full_index=1";
	hash = "sha256-dVpnYI2JLDEt2VQkbI6RnbpdcQSMB5bTQ6TJp5CLnyw=";
	})
	];
	});

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.58%. Comparing base (7ba22c2) to head (c47b864).
⚠️ Report is 10 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #66895   +/-   ##
=======================================
  Coverage   90.58%   90.58%           
=======================================
  Files         797      797           
  Lines      278434   278465   +31     
  Branches    53531    53531           
=======================================
+ Hits       252216   252251   +35     
+ Misses      16571    16570    -1     
+ Partials     9647     9644    -3     

Files with missing lines	Coverage Δ
src/node_zlib.cc	79.86% <100.00%> (-0.44%)	⬇️

... and 37 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[ziyao233]

Since commit introducing this regression is related to a CVE, I wonder would such workaround breaks the fix?

[0x9fff00]

Brotli 1.2.1 has been released with a backport of the fix for CVE-2026-111497 which added BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH.

Since commit introducing this regression is related to a CVE, I wonder would such workaround breaks the fix?

Yes. If --shared-brotli is used, Brotli 1.2.1 is required in order to fix CVE-2026-111497 because in that case the vendored copy of Brotli in Node.js is not used.

[github-actions]

Platform	Number of requisites	Proportion of new derivations
x86_64-linux	544 -> 541 => +-3	0 / 541 = 0.0%
aarch64-linux	545 -> 542 => +-3	0 / 542 = 0.0%
x86_64-darwin	457 -> 454 => +-3	0 / 454 = 0.0%
aarch64-darwin	470 -> 467 => +-3	0 / 467 = 0.0%
Total	2016 -> 2004 => +-12	0 / 2004 = 0.0%

Changelog

@@ -120 +119,0 @@
-/nix/store/kminssh6dm17s3f6cqszaxwpyyhm6i88-brotli-1.2.0 (aarch64-darwin)
@@ -122,2 +120,0 @@
-/nix/store/6aaijr2v6rlgailsir2l1q9j3k72flxx-brotli-1.2.0 (aarch64-linux)
-/nix/store/895yl6jdl10mkffq8qrgimy2i7snla3q-brotli-1.2.0 (x86_64-darwin)
@@ -125 +121,0 @@
-/nix/store/2nyhgrln2i4rw0rdijyqh0nn333ws3nx-brotli-1.2.0 (x86_64-linux)
@@ -128,2 +123,0 @@
-/nix/store/c4y22jmyh7wjhbdidbnhz661jpg9kaim-brotli-1.2.0-dev (aarch64-darwin)
-/nix/store/44lj6r7v54yfzcd7p6l6n76g7ma7y633-brotli-1.2.0-dev (aarch64-linux)
@@ -132 +125,0 @@
-/nix/store/af449dhklypcj2kild1fv4q9dy2bmjdv-brotli-1.2.0-dev (x86_64-darwin)
@@ -134 +126,0 @@
-/nix/store/ainnzs18blgfd5f3n21wnq0m4j5ryf6j-brotli-1.2.0-dev (x86_64-linux)
@@ -136 +127,0 @@
-/nix/store/kl3m09q74kf4g4mfk9whw65w8dw0jg3a-brotli-1.2.0-lib (aarch64-darwin)
@@ -138 +128,0 @@
-/nix/store/r5vy3djdp4qpddj4bsw8hqcl9asyp1h2-brotli-1.2.0-lib (aarch64-linux)
@@ -140 +129,0 @@
-/nix/store/yvkvvnyvnaw08bl9gvnkbk33m339v4ff-brotli-1.2.0-lib (x86_64-darwin)
@@ -142 +130,0 @@
-/nix/store/f667bn2vz2vv1hf4ayjc49r3w967ivhr-brotli-1.2.0-lib (x86_64-linux)

[lazerg]

@aduh95 ✅ Done

I removed the Nix cherry-pick of the brotli patch (c47b864). The shared-libraries CI now builds against unpatched brotli, so it checks the fix by itself.

I did not use your #ifdef suggestion for src/node_zlib.cc. BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH is an enum value, and #ifdef cannot see an enum value. The constant would be missing from every build. The constants now come from BROTLI_DECODER_ERROR_CODES_LIST in the brotli header (67111ab).

Local result: all 90 parallel/test-zlib* tests pass with system brotli 1.2.1. With brotli 1.2.0, the block-switch test is skipped.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/78562/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/78563/