788971 - Pymake chokes when native commands raise exceptions without a "code" attribute

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Mike Hommey [:glandium]]

Bug 784262 caused bug 788954, which is a race condition happening because export-idl and export are trying to do the same thing in the same directory at the same time. It seems to be causing deadlocks on windows builders in bug 788917.

Comment 1

[Rain [:rain1] (inactive)]

So this is a bug in rules.mk that also manifested as a deadlock in Pymake?

Comment 2

[Mike Hommey [:glandium]]

Apparently, yes.

Comment 3

[Mike Hommey [:glandium]]

random dead-lock.

Comment 4

[Rain [:rain1] (inactive)]

Attachment: oops

Comment 5

[(not currently active) Ted Mielczarek]

Comment on attachment 658881 [details] [diff] [review]
oops

Review of attachment 658881 [details] [diff] [review]:
-----------------------------------------------------------------

::: pymake/process.py
@@ +264,5 @@
>                  pass # sys.exit(0) is not a failure
>              else:
>                  print >>sys.stderr, e
> +                traceback.print_exc()
> +                return -127

Can we be smarter about this and use getattr(e.code, -127)?

Comment 6

[Rain [:rain1] (inactive)]

I'd rather not -- e.code isn't defined in general. It's also defined for a few other exceptions, e.g. HTTP errors in urllib2, and I don't think we should turn those into something equivalent to process exit codes!

Comment 7

[Rain [:rain1] (inactive)]

https://hg.mozilla.org/users/bsmedberg_mozilla.com/pymake/rev/5a286a07783f
https://hg.mozilla.org/integration/mozilla-inbound/rev/a111a94df092

Comment 8

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/a111a94df092