778035 - OSX64 builds failing due to missing CC/CXX exports

Status: RESOLVED FIXED

(SeaMonkey :: Release Engineering, defect)

Description

[Edmund Wong (:ewong)]

Bug 775539 is to setup OSX boxes to use clang instead of gcc; but,
clang has not been set up on the OSX machines.  In the meantime,
the OSX64 boxes are not building due to missing CC/CXX exports.

m-c's /build/macosx/common and c-c's build/macosx/common currently
looks like this:

 if [ -d "$topsrcdir/clang" ]; then
     # mozilla-central based build
     export CC=$topsrcdir/clang/bin/clang
     export CXX=$topsrcdir/clang/bin/clang++
 elif [ -d "$topsrcdir/../clang" ]; then
     # comm-central based build
     export CC=$topsrcdir/../clang/bin/clang
     export CXX=$topsrcdir/../clang/bin/clang++
 fi

Since SeaMonkey doesn't have clang installed, the if and elif blocks
don't apply and thusly CC and CXX are not set.

Comment 1

[Edmund Wong (:ewong)]

Attachment: Add else block to set CC/CXX for systems that don't have clang. (m-c)

The rationale for this patch is to ensure that the c-c patch to this bug is 
the same as the one in m-c's build/macosx/common so that during the check
of macosx/common, it would not fail.  In theory, this patch should not affect
FF's build.

Comment 2

[Edmund Wong (:ewong)]

Attachment: Add else block to set CC/CXX for systems that don't have clang. (c-c) (v1)

Comment 3

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Comment on attachment 646462 [details] [diff] [review]
Add else block to set CC/CXX for systems that don't have clang. (m-c)

Review of attachment 646462 [details] [diff] [review]:
-----------------------------------------------------------------

I don't know much about Mac, sorry.

Comment 4

[(not currently active) Ted Mielczarek]

I agree that we should be doing something to make this suck less. I'm not sure if falling back to GCC is really what we want. I'd prefer to make the mozconfig error if CC/CXX are not set. You could just set CC/CXX in your build environment, then.

My reasoning is that we don't really want to support GCC 4.2 on Mac much longer, so I'd rather not encode it into the mozconfigs.

Comment 5

[Justin Wood (:Callek)]

(In reply to Ted Mielczarek [:ted] from comment #4)
> I agree that we should be doing something to make this suck less. I'm not
> sure if falling back to GCC is really what we want. I'd prefer to make the
> mozconfig error if CC/CXX are not set. You could just set CC/CXX in your
> build environment, then.
> 
> My reasoning is that we don't really want to support GCC 4.2 on Mac much
> longer, so I'd rather not encode it into the mozconfigs.

The problem is that until SeaMonkey gets tooltool up, we need to have matching files from comm-* to mozilla-* otherwise the check-sync-dirs fails.

Would you be ok with this as a TEMPORARY solution until I can get the clang stuff working on our systems?

Comment 6

[(not currently active) Ted Mielczarek]

Can't you just export CC/CXX to the paths to GCC in the meantime? We're not planning on unsupporting GCC 4.2 immediately, so that should work fine.

Comment 7

[(not currently active) Ted Mielczarek]

Comment on attachment 646462 [details] [diff] [review]
Add else block to set CC/CXX for systems that don't have clang. (m-c)

Review of attachment 646462 [details] [diff] [review]:
-----------------------------------------------------------------

I'll let Rafael make the final call here.

Comment 8

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Comment on attachment 646462 [details] [diff] [review]
Add else block to set CC/CXX for systems that don't have clang. (m-c)

Using /usr/bin/gcc is definitely *NOT* OK. That is llvm-gcc, which is not supported and is know to be broken.

Given that it took releng just a few hours in the previous bugs to setup tooltool for a given branch, I think it would be better to just ask them to set it up for seamonkey.

If there is a good reason why that is not possible, you should use /usr/bin/clang or as a very last resort /usr/bin/gcc-4.2.

Comment 9

[Justin Wood (:Callek)]

(In reply to Rafael Ávila de Espíndola (:espindola) from comment #8)
> Comment on attachment 646462 [details] [diff] [review]
> Add else block to set CC/CXX for systems that don't have clang. (m-c)
> 
> Using /usr/bin/gcc is definitely *NOT* OK. That is llvm-gcc, which is not
> supported and is know to be broken.

Fair enough, but it *might* be different on SeaMonkey machines, we'll have to verify that.

> Given that it took releng just a few hours in the previous bugs to setup
> tooltool for a given branch, I think it would be better to just ask them to
> set it up for seamonkey.

Well SeaMonkey is managed by a completely different system and not MoCo releng, and doesn't yet have tooltool support setup. *AND* we don't have a tooltool repository available for us yet (since the current tooltool repo for MoCo has private bits in it, the public mirror of public bits needs coding work to even be useable or to have stuff on it)

> If there is a good reason why that is not possible, you should use
> /usr/bin/clang or as a very last resort /usr/bin/gcc-4.2.

Until we can setup tooltool I think its safer to use gcc-*, unless it breaks at some point. The main point here is to keep stuff working (right) for SeaMonkey, since the in-tree mozconfigs are designed for official machines, not end-users, hardcoding expectations has been ok so far :-)

Comment 10

[Edmund Wong (:ewong)]

Attachment: Add else block to set CC/CXX for systems that don't have clang. (m-c) (v2)

Comment 11

[Edmund Wong (:ewong)]

Attachment: Add else block to set CC/CXX for systems that don't have clang. (c-c) (v2)

Comment 12

[Edmund Wong (:ewong)]

Attachment: Add else block to set CC/CXX for systems that don't have clang. (m-c) (v2)

Comment 13

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

> Well SeaMonkey is managed by a completely different system and not MoCo
> releng, and doesn't yet have tooltool support setup. *AND* we don't have a
> tooltool repository available for us yet (since the current tooltool repo
> for MoCo has private bits in it, the public mirror of public bits needs
> coding work to even be useable or to have stuff on it)

Oh, I was under the impression that (like Thunderbird), SeaMonkey was using releng's infrastructure.

> > If there is a good reason why that is not possible, you should use
> > /usr/bin/clang or as a very last resort /usr/bin/gcc-4.2.
> 
> Until we can setup tooltool I think its safer to use gcc-*, unless it breaks
> at some point. The main point here is to keep stuff working (right) for
> SeaMonkey, since the in-tree mozconfigs are designed for official machines,
> not end-users, hardcoding expectations has been ok so far :-)

Exactly, they are designed to work with releng's infrastructure. If you are not using that, you should not be using these files.

Comment 14

[(not currently active) Ted Mielczarek]

Comment on attachment 649923 [details] [diff] [review]
Add else block to set CC/CXX for systems that don't have clang. (m-c) (v2)

Review of attachment 649923 [details] [diff] [review]:
-----------------------------------------------------------------

Okay. In the interest of getting SeaMonkey builds unblocked I'm going to r+ this patch. The caveat is that we are going to unsupport gcc-4.2 after Firefox 17 branches (the next cycle), and this change is going to be backed out when that happens. That does give you a few weeks to get your build environment in order, so hopefully that's enough to get you where you need to be.

Comment 15

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Attachment: revert it

To be committed 2012-08-27 after the merge.

Comment 16

[(not currently active) Ted Mielczarek]

Comment on attachment 650235 [details] [diff] [review]
revert it

Review of attachment 650235 [details] [diff] [review]:
-----------------------------------------------------------------

This is fine, but can you file a new bug for this? (If we already have a "stop supporting GCC-4.2" bug, you could make it block that one.)

Comment 17

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Done. Do you want me to move the patch to 781302 or leave it here?

Comment 18

[(not currently active) Ted Mielczarek]

For sanity you should probably move it there. You can just propagate my r+ there.

Comment 19

[Edmund Wong (:ewong)]

Pushed to m-c:
https://hg.mozilla.org/mozilla-central/rev/08e0e3b95f03

Comment 20

[Edmund Wong (:ewong)]

Pushed to comm-central:
http://hg.mozilla.org/comm-central/rev/2019e8845fbf

Comment 21

[Mozilla RelEng Bot]

Try run for 513eaff31bfd is complete.
Detailed breakdown of the results available here:
    https://tbpl.mozilla.org/?tree=Try&rev=513eaff31bfd
Results (out of 16 total builds):
    exception: 16
Builds (or logs if builds failed) available at:
http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-513eaff31bfd