1036645 - The config.cache for subconfigures is blown away every time configure runs

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[(no longer active)]

glandium tells me that the heuristic in bug 906403 is broken...

Comment 1

[Mike Hommey [:glandium]]

There are several issues in the script that checks config.cache:
- it doesn't unquote values, so it compares e.g. "'foo'" with "foo", seens the mismatch, and throws away the cache.
- it doesn't consider the combination of ac_cv_env_*_set and ac_cv_env_*_value vs. os.environ entirely properly
- it tries to match the *_alias variables with the environment, when what config.cache stores comes from the configure command line arguments.

Comment 2

[Mike Hommey [:glandium]]

Attachment: part 1 - USe AC_OUTPUT_SUBDIRS to call ICU configure

As ICU needs an objdir different from its source directory, add support for
that in AC_OUTPUT_SUBDIRS.

Comment 3

[Mike Hommey [:glandium]]

Attachment: part 1 - USe AC_OUTPUT_SUBDIRS to call ICU configure

I had forgotten to remove ICU_SRCDIR, which is now useless. Note it's safe to remove because AC_OUTPUT_SUBDIRS does the right thing already.

Comment 4

[Mike Hommey [:glandium]]

Attachment: part 2 - Make using a separate config.cache the default for AC_OUTPUT_SUBDIRS

All subconfigures are essentially doing it already, so just inverse the process. That would also limit problems with additional subconfigures (all the recent ones had to come with their own config.cache)

Comment 5

[Mike Hommey [:glandium]]

Attachment: part 3 - Fix the subconfigure wrapper heuristics to decide when to kill config.cache

Comment 6

[Mike Hommey [:glandium]]

https://tbpl.mozilla.org/?tree=Try&rev=3db73649263d

Comment 7

[Mike Hommey [:glandium]]

Note that by not using AC_OUTPUT_SUBDIRS, ICU's configure was *not* using a config.cache, so this patch set changes that as well.

Comment 8

[Mike Hommey [:glandium]]

Attachment: part 2 - Make using a separate config.cache the default for AC_OUTPUT_SUBDIRS

Turns out the freetype subconfigure was relying on using the main config.cache.

Comment 9

[Mike Hommey [:glandium]]

https://tbpl.mozilla.org/?tree=Try&rev=a1c98df7cb24 With the new part 2, only for android.

Comment 10

[Gregory Szorc [:gps]]

Comment on attachment 8453557 [details] [diff] [review]
part 3 - Fix the subconfigure wrapper heuristics to decide when to kill config.cache

Review of attachment 8453557 [details] [diff] [review]:
-----------------------------------------------------------------

::: build/subconfigure.py
@@ +65,5 @@
> +    parser = argparse.ArgumentParser()
> +    parser.add_argument('--target', type=str)
> +    parser.add_argument('--host', type=str)
> +    parser.add_argument('--build', type=str)
> +    args, others = parser.parse_known_args(args)

Wait, there is a parse_known_args()?! We could be using this for so many mach commands. Dammit.

Comment 11

[Mike Hommey [:glandium]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/c4a56f872991
https://hg.mozilla.org/integration/mozilla-inbound/rev/93f1af4bb4b2
https://hg.mozilla.org/integration/mozilla-inbound/rev/a0ab7fa5dd58

Comment 12

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/c4a56f872991
https://hg.mozilla.org/mozilla-central/rev/93f1af4bb4b2
https://hg.mozilla.org/mozilla-central/rev/a0ab7fa5dd58