On Tue, 22 Oct 2024 21:30:40 GMT, Erik Joelsson <er...@openjdk.org> wrote:
> The target mac-jdk-bundle can fail randomly. MacBundles.gmk defines a large > number of individual copy rules, which can execute in any order. The > "install-file" (our copy) macro on macos includes a check for weird > attributes using `xattr` so that we can remove them. We use the switch `-s` > to make sure that xattr operates on symlinks instead of their targets. > However, if we find something, we also run `chmod` to make sure we have the > permissions to remove attributes in the first place, but the chmod command > does not have the `-h` switch to operate on the symlink instead of the > target. So if the symlink gets copied before its target, there is a chance > that we try to run chmod before the target exists, which will result in a > failure. My proposed fix is to add `-h` to the chmod command so that > everything operates on the symlink instead of the target. Marked as reviewed by ihse (Reviewer). This finally explains the odd problem you reported in https://github.com/openjdk/jdk/pull/20837#issuecomment-2327132241. That has been worrying me a bit, if there were some issue with that PR after all, but it turrned out to just be an unlikely race. *phew* ------------- PR Review: https://git.openjdk.org/jdk/pull/21649#pullrequestreview-2389221730 PR Comment: https://git.openjdk.org/jdk/pull/21649#issuecomment-2432657995
