You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As a repo snapshot, the v8.0.1 contains an outdated yarn.lock, which browserslist is unhappy with. This is never an issue for our downstream npm consumers since the lockfile will not be published.
I recommend we implement the following two changes to the cigtm script:
Fetch the latest stable release version from github.com
Because Babel only publishes changed packages, there are chances that the latest @babel/core version, with the lockfile in the repo snapshot, does not work with latest Node.js. So we should fetch the latest version from github.com rather than npm registry. For example, the following command should print v8.0.4 as of September 9, 2026:
As for head: true, I am afraid of noises from the HEAD, things might break especially during the time when Babel merges quite a few feature PRs before we do a minor release. If we set head: true chances are that an unreleased @babel/core might break the citgm. Most of time we will fix any conflict/errors after merging, but I don't feel it necessary to pop up the noise to here.
Activity
liuxingbaoyu commentedon Sep 9, 2026
Thank you for the reminder, I will look into it.
liuxingbaoyu commentedon Sep 9, 2026
There appear to be multiple issues, but for some reason, citgm is testing
v8.0.1, whereas the latest version is https://github.com/babel/babel/tree/v8.0.4.MikeMcC399 commentedon Sep 9, 2026
https://www.npmjs.com/package/@babel/core?activeTab=versions shows that the npm registry has
v8.0.1as the latest published.liuxingbaoyu commentedon Sep 9, 2026
I see, Babel only publishes packages that have changed.
Can we have citgm always fetch the
mainbranch?targos commentedon Sep 9, 2026
Yes, that can be done with
"head": truein the lookup config.JLHwung commentedon Sep 9, 2026
The root cause of the failure is that browserslist is unhappy with the
caniuse-liteprovided in the repo snapshot of Babelv8.0.1.I think this issue can be addressed in the CIGTM script.
Currently cigtm downloads
As a repo snapshot, the v8.0.1 contains an outdated yarn.lock, which
browserslistis unhappy with. This is never an issue for our downstream npm consumers since the lockfile will not be published.I recommend we implement the following two changes to the cigtm script:
Because Babel only publishes changed packages, there are chances that the latest
@babel/coreversion, with the lockfile in the repo snapshot, does not work with latest Node.js. So we should fetch the latest version from github.com rather than npm registry. For example, the following command should print v8.0.4 as of September 9, 2026:Then we can supply the version to the downloader.
BROWSERSLIST_IGNORE_OLD_DATAenvironment variable to1in the github workflow:Per https://github.com/browserslist/browserslist/blob/f46a5b74e75fec2bda62c6520fc2296bda3c0fe3/node.js#L476, it disables the browserslist check.
As for
head: true, I am afraid of noises from the HEAD, things might break especially during the time when Babel merges quite a few feature PRs before we do a minor release. If we sethead: truechances are that an unreleased@babel/coremight break the citgm. Most of time we will fix any conflict/errors after merging, but I don't feel it necessary to pop up the noise to here.MikeMcC399 commentedon Oct 8, 2026
If there anybody planning to make changes to the settings for
@babel/corein citgm following on from the previous issue analysis?The alternative would be to add a temporary full skip entry until a fix is available.
MikeMcC399 commentedon Oct 12, 2026
citgm continues to fail even with the latest release @babel/core@8.0.7
To reproduce locally on Ubuntu 26.04.1 LTS, Node.js 24.21.0 LTS
at-babel-core-8.0.7.log
MikeMcC399 commentedon Oct 13, 2026
@babel/core@8.0.7does not fail in Jenkins for v24.x & v26.xClosing this issue in favor of separate issue #1194 regarding tests that are not skipped outside of Jenkins.