Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
The current state of Win PGO on mozilla-inbound
4 views
Skip to first unread message
Philip Ringnalda
Oct 22, 2011, 2:02:07 AM10/22/11
to
https://hg.mozilla.org/integration/mozilla-inbound/rev/73512caf4647 got
PGO builds, and tests were fine.
https://hg.mozilla.org/integration/mozilla-inbound/rev/29edecdbbea7 was
a backout, which I certainly hope wouldn't have affected PGO
https://hg.mozilla.org/integration/mozilla-inbound/rev/924091e5589e was
dbus, which I certainly hope wouldn't have affected PGO on Windows
https://hg.mozilla.org/integration/mozilla-inbound/rev/c09b6f933f76 and
https://hg.mozilla.org/integration/mozilla-inbound/rev/107c941e27dc are
on the hook
https://hg.mozilla.org/integration/mozilla-inbound/rev/41f8fb72be22 only
touched comments
https://hg.mozilla.org/integration/mozilla-inbound/rev/cb50e96dbee0 got
PGO builds, and the Windows M1 and M4 runs had "tagNameGetter.call is
not a function" and "getter.call is not a function" and "property
descriptor's getter field is neither undefined nor a function" errors
testing the PGO builds which they did not have testing the non-PGO builds.
I retriggered the M1 and M4 tests, and got the same errors on the same
build.
I retriggered the Win PGO build on that rev (difficult, but possible if
you can figure out how).
Since there's no way to trigger a PGO build on an arbitrary rev, I
triggered nightly builds on both 107c941e27dc and c09b6f933f76 and
cancelled the non-Windows ones.
If we're lucky, when those finish and trigger tests, they'll show both
that it's reproducible across builds, not just across test runs on a
build that exhibits it, and will show which rev started it. If we're not
lucky, and it's intermittent, then the field is pretty wide open, choose
your multiple of four hours into the past.
If we're at all sane, we'll take this as a clear sign that we should
stop doing on-change non-PGO, and go back to building and testing what
we ship, what we care about, on-change.
PGO builds, and tests were fine.
https://hg.mozilla.org/integration/mozilla-inbound/rev/29edecdbbea7 was
a backout, which I certainly hope wouldn't have affected PGO
https://hg.mozilla.org/integration/mozilla-inbound/rev/924091e5589e was
dbus, which I certainly hope wouldn't have affected PGO on Windows
https://hg.mozilla.org/integration/mozilla-inbound/rev/c09b6f933f76 and
https://hg.mozilla.org/integration/mozilla-inbound/rev/107c941e27dc are
on the hook
https://hg.mozilla.org/integration/mozilla-inbound/rev/41f8fb72be22 only
touched comments
https://hg.mozilla.org/integration/mozilla-inbound/rev/cb50e96dbee0 got
PGO builds, and the Windows M1 and M4 runs had "tagNameGetter.call is
not a function" and "getter.call is not a function" and "property
descriptor's getter field is neither undefined nor a function" errors
testing the PGO builds which they did not have testing the non-PGO builds.
I retriggered the M1 and M4 tests, and got the same errors on the same
build.
I retriggered the Win PGO build on that rev (difficult, but possible if
you can figure out how).
Since there's no way to trigger a PGO build on an arbitrary rev, I
triggered nightly builds on both 107c941e27dc and c09b6f933f76 and
cancelled the non-Windows ones.
If we're lucky, when those finish and trigger tests, they'll show both
that it's reproducible across builds, not just across test runs on a
build that exhibits it, and will show which rev started it. If we're not
lucky, and it's intermittent, then the field is pretty wide open, choose
your multiple of four hours into the past.
If we're at all sane, we'll take this as a clear sign that we should
stop doing on-change non-PGO, and go back to building and testing what
we ship, what we care about, on-change.
Marco Bonardo
Oct 22, 2011, 3:48:29 AM10/22/11
to
On 22/10/2011 08:02, Philip Ringnalda wrote:
> Since there's no way to trigger a PGO build on an arbitrary rev, I
> triggered nightly builds on both 107c941e27dc and c09b6f933f76 and
> cancelled the non-Windows ones.
Indeed that's most likely the bug to fix, we should be able to trigger
> Since there's no way to trigger a PGO build on an arbitrary rev, I
> triggered nightly builds on both 107c941e27dc and c09b6f933f76 and
> cancelled the non-Windows ones.
pgo builds and tests on demand. If we can't then we can't either fix issues.
-m
Marco Bonardo
Oct 22, 2011, 5:09:01 AM10/22/11
to
On 22/10/2011 08:02, Philip Ringnalda wrote:
> https://hg.mozilla.org/integration/mozilla-inbound/rev/73512caf4647 got
> PGO builds, and tests were fine.
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/29edecdbbea7 was
> a backout, which I certainly hope wouldn't have affected PGO
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/924091e5589e was
> dbus, which I certainly hope wouldn't have affected PGO on Windows
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/c09b6f933f76 and
> https://hg.mozilla.org/integration/mozilla-inbound/rev/107c941e27dc are
> on the hook
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/41f8fb72be22 only
> touched comments
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/cb50e96dbee0
Philor's strategy to use Nightlies payed off, I backed out cb50e96dbee0
> PGO builds, and tests were fine.
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/29edecdbbea7 was
> a backout, which I certainly hope wouldn't have affected PGO
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/924091e5589e was
> dbus, which I certainly hope wouldn't have affected PGO on Windows
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/c09b6f933f76 and
> https://hg.mozilla.org/integration/mozilla-inbound/rev/107c941e27dc are
> on the hook
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/41f8fb72be22 only
> touched comments
>
> https://hg.mozilla.org/integration/mozilla-inbound/rev/cb50e96dbee0
as the first changeset showing orange PGO tests.
Now waiting results to check it was effective.
-m
Marco Bonardo
Oct 22, 2011, 8:24:35 AM10/22/11
to
On 22/10/2011 11:09, Marco Bonardo wrote:
> Philor's strategy to use Nightlies payed off, I backed out cb50e96dbee0
> as the first changeset showing orange PGO tests.
> Now waiting results to check it was effective.
> -m
Got confirmation, win PGO permaorange is now fixed. Thank you philor for
> Philor's strategy to use Nightlies payed off, I backed out cb50e96dbee0
> as the first changeset showing orange PGO tests.
> Now waiting results to check it was effective.
> -m
those retriggers.
Edmorley is looking into a merge now, since I'm running out of time for
today.
-m
Mike Hommey
Oct 26, 2011, 2:08:36 AM10/26/11
to Marco Bonardo, dev-tree-...@lists.mozilla.org
On Sat, Oct 22, 2011 at 09:48:29AM +0200, Marco Bonardo wrote:
> On 22/10/2011 08:02, Philip Ringnalda wrote:
> >Since there's no way to trigger a PGO build on an arbitrary rev, I
> >triggered nightly builds on both 107c941e27dc and c09b6f933f76 and
> >cancelled the non-Windows ones.
>
> Indeed that's most likely the bug to fix, we should be able to
> trigger pgo builds and tests on demand.
More than that, we should also be able to trigger pgo try builds
> On 22/10/2011 08:02, Philip Ringnalda wrote:
> >Since there's no way to trigger a PGO build on an arbitrary rev, I
> >triggered nightly builds on both 107c941e27dc and c09b6f933f76 and
> >cancelled the non-Windows ones.
>
> Indeed that's most likely the bug to fix, we should be able to
> trigger pgo builds and tests on demand.
with the try chooser.
Mike
0 new messages