Bug 108028 – REGRESSION (r140912): Broke specifying non-trivial ARCHS value to make / build-webkit

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Mark Rowe (bdash)]

Prior to r140912 the following worked and had the expected behavior:

> make debug ARGS="ARCHS='x86_64 i386' ONLY_ACTIVE_ARCH=NO"

> ./Tools/Scripts/build-webkit ARCHS='$(ARCHS_STANDARD_32_64_BIT)' ONLY_ACTIVE_ARCH=NO

The former now results in completely bogus arguments being passed to xcodebuild:

> xcodebuild -configuration Debug -sdk / -arch  -arch  -arch  -arch x86_64 

As does the latter:

> xcodebuild -project WTF.xcodeproj -configuration Debug -sdk / -arch  -arch $( -arch ) ONLY_ACTIVE_ARCH=NO

This is breaking the nightly builds, since they pass explicit ARCHS flags to the build scripts, and it also breaks our established method of building for non-default / multiple architectures.

The fact we're now passing an explicit "-sdk" argument for OS X builds is also not good, but isn't a cause of this breakage.

[David Farler]

I think I might like to just revert the previous patch entirely and rethink this. Using ARGS= in the makefile and SDKROOT= ARCHS= with build-webkit / xcodebuild, almost everything will work with the same usage as now. Mixing make variables and unparsed Xcode options in $(ARGS) or un-spliced through webkitdirs.pm doesn’t seem worth it. The only problem leftover is reducing the modules lists in Tools/Makefile and Source/Makefile, which I can shore up in a later patch.

[David Farler]

Created attachment 184909 [details]
Patch

[WebKit Review Bot]

Comment on attachment 184909 [details]
Patch

Clearing flags on attachment: 184909

Committed r140939: <http://trac.webkit.org/changeset/140939>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.