On Thu, Sep 10, 2026 at 11:18:21PM +0200, Michael Haener wrote: > v2: > - reduced the scope to the timestamp fix
The error handling changed too. sstate_package(), sstate_checkhashes() and sstate_eventhandler() reached os.utime() through a filter that let PermissionError and EROFS pass -- "Handle read-only file systems gracefully" -- and raised the rest. sstate_touch_atime() catches OSError unconditionally, so ENOSPC, EIO and ESTALE are silent at those three now. Deliberate? > so its age can no longer be told from the cache itself scripts/sstate-cache-management.py is the consumer, and the patch helps it. remove_duplicated() keeps the entry with the larger st_mtime, and --remove-duplicated promises "only the newest one will be kept". Today use bumps the modification time, so building an older configuration last keeps its object and deletes the newer one. st_mtime is a float; writing it back moves the modification time by -112 ns on ext4 here, once. os.utime(path, ns=(time.time_ns(), stat_info.st_mtime_ns)) round trips exactly. Thanks, Zhan Xusheng
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#245606): https://lists.openembedded.org/g/openembedded-core/message/245606 Mute This Topic: https://lists.openembedded.org/mt/121188981/21656 Group Owner: [email protected] Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
