Something broke nightly tester (no rule to build target)

[llvmbot]

Bugzilla Link	1283
Resolution	FIXED
Resolved on	Feb 22, 2010 12:54
Version	trunk
OS	Linux
Reporter	LLVM Bugzilla Contributor

Extended Description

Last night's run of my nightly test yielded errors like:

*** No rule to make target Output/sumarray-dbl.linked.rbc', needed by Output/sumarray-dbl.linked.bc'

for every single test program. The consequence was no results. The only change
to the makefile system was this:

--- Makefile.programs 21 Mar 2007 00:19:32 -0000 1.263
+++ Makefile.programs 28 Mar 2007 08:31:17 -0000 1.264
@@ -221,12 +221,11 @@ LLCBETAOPTION := -sched=list-td
endif
ifeq ($(ARCH),IA64)
LLCBETAOPTION := -sched=simple
endif
ifeq ($(ARCH),x86)
-LLCBETAOPTION := -enable-rematerialization
-# -enable-tail-merge
+LLCBETAOPTION := -enable-tail-merge
#-regalloc=local -fast
endif
ifeq ($(ARCH),Sparc)
LLCBETAOPTION := -enable-sparc-v9-insts
endif

But, I don't see how that could cause this problem.

[lattner]

This happens when llvm's configure script can't find llvm-gcc.

[llvmbot]

That's not it. The llvm/Makefile.config has llvm-gcc configured in it.
Furthermore, the CFrontend and C++Frontend tests would have failed but they didn't.

[llvmbot]

I checked my build logs. The update/build/install of llvm-gcc went smoothly last
night.

[llvmbot]

This is apparently not a transiet problem as it failed for a second night in a
row. I have no idea what's causing this.

[llvmbot]

This problem is now fixed. The configuration test for llvm-gcc includes a sanity
check that was looking for the "implementation" keyword. That keyword just got
removed from llvm. The fix is to make the sanity check look for "target
datalayout" which is probably more distinct anyway.

Patch here:

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070326/046405.html