BSD Commit Log Search

sh: Reduce overhead of FUNCNAMESTACK setting.

rename: Add -q flag which can avoid some redirects

pkgqueue: Avoid a few possible path expansions

exit_handler: Remove redundant lock releasing.

Since all locks are released right after this we don't need to
explicitly release any locks.

runtest: truss multiplier of 4 was too small

pkgqueue_list_deps_recurse: Check empty dir before big list of dupes.

builtin rename: Add -v flag

pkgqueue_get_next: Ensure returning empty is an empty ready-to-build queue.

Note the queue may have waiting-for-dependency items but it should
not have anything in the ready-to-build queue.

Similar to the comments in pkgqueue_empty() we need to ensure
no blank jobs are returned. It is possible for a race to happen
where pkgqueue_get_next() returns a blank job, build_queue()
gets it and checks pkgqueue_empty() and sees it is not empty,
loops around and runs the sanity check due to idle builders
and calls crashed_build() on an item that just got balanced
and was temporarily missing in the find(1) call which itself
is not an atomic/locked view of the queue.

This became apparent after making pkgqueue_balance_pool() into a
separate worker process. The first job in the queue would sometimes
get moved while the first pkgqueue_get_next() was called and
immediately get "stuck_in_queue".

lock_acquire: recursion is only expected once for traps.

Move some code that isn't needed for recursions.

No current functional change intended.

tests: Cleanup locks automatically

Also remove blank `${MASTERNAME}-` if not set.

pkgqueue_balance_pool: Run within the pool dir to reduce string data

build_queue: Add some msg_dev

pkgqueue_move_ready_to_pool: Allow running more than once.

This is for test purposes.

This also now captures the error from `mkdir`.

setup_traps: Only push into stack if a handler is given

bulk: Save .p dir listing on error

Revert INJOB changes.

Jobs within jobs work fine. It's just that they do not _inherit_ the
parent process jobs.

Revert "prefixers: Supported nested jobs (need to use pids here)"

This reverts commit 6d6d888f0f796e974517f31984ebd6f89a721308.

Revert "kill_job: Assert not killing a nested job"

This reverts commit 1a1a34c3750688d2df42dff7ab769dbef21fd3c4.

prefix_output test: Make debug easier

sh: Show FUNCNAMESTACK on set -e error and other errors

Merge branch 'image-populate-efi-fstab'

image: Populate the efi mount point in fstab

Merge branch 'fix-pkgclean-respecting-packages-override'

Fix pkgclean respecting PACKAGES being set in a jail-poudriere.conf

pkgqueue_balance_pool: Use lock_acquire().

No functional change intended.

pkgqueue_clean_queue: Clarify comment: It can run from main process too

pkg_notes_get: Avoid setvar problems

make checkquick: Include dry bulk tests

testport: Fix not adding PKGENV if no notes are added.

This was seen on cirrus where there is no git checkout.

build quick tests: Also test PKG_NO_VERSION_FOR_DEPS

set +x in some noisy functions

bulk tests: rework some logclean and error handling.

- Unhide output
- Only consider lock acquisition failure OK
- Don't logclean while acquiring test contexts
- While here avoid bootstrap code with test context lookup
- logclean: Ignore potential harmless race in find_broken_latest_per_pkg_links