On 2/16/23 8:32 AM, Michal Privoznik wrote:
This is a v2 of:
https://listman.redhat.com/archives/libvir-list/2023-February/237731.html
diff to v1:
- Merged patches that were ACKed in v1,
- Dropped 4/4 from the original series (the one that sets --foreground),
and implemented a different approach
Michal Prívozník (5):
qemu_passt: Avoid double daemonizing passt
qemu_passt: Report passt's error on failed start
qemu_passt: Make passt report errors to stderr whenever possible
qemu_passt: Deduplicate passt killing code
qemu_passt: Let passt write the PID file
This is everything that was in the patch I sent last week, with the
following additions
1) adding NULLSTR() around the reference to errbuf in patch 2/5
2) adding "--stderr" to the commandline in patch 3/5 (which I found to
be unnecessary in my testing - as Stefano says everything goes to stderr
until passt has completed its init anyway)
3) the other bit of patch 3/5 which adds an extra message telling the
user to look into the designated logfile for the error - this is
unnecessary (and actually now counter-productive, as it forces you to
look elsewhere for the error when you wouldn't have needed to) because
of patches I've sent to passt.
4) patch 4/5 that is a cleanup de-duplicating code
5) patch 5 changes additional code (that I didn't touch in my patch) to
use virPidFileReadPath() instead of virPidFileReadPathIfLocked(), and
virProcessKillPainfully() instead of the higher level
virPidFileForceCleanupPath().
So it all seems fine (except the error reporting stuff), but why revert
a patch only to push back the same changes in a deconstructed fashion
plus some fixups, rather than just posting a followup or two?
src/qemu/qemu_passt.c | 61 +++++++++++++++++++++++++++++--------------
1 file changed, 41 insertions(+), 20 deletions(-)
