It doesn't seem like anyone is especially interested in this, and I'm
reasonably confident that the patch is correct. I'm going to go ahead
and commit it.
On Mar 10, 2009, at 11:20 AM, Bob Wilson wrote:
> The utils/NewNightlyTest.pl script seems to use LLVMGCCDIR in a way
> that is inconsistent with the top-level configure script. I would
> like to fix this but I don't want to break everyone's nightly
> tests. Can someone review the attached patch to make sure I'm not
> missing something?
>
> The NewNightlyTest.pl script checks the LLVMGCCDIR environment
> variable and uses it for two things: 1) it passes it to the top-
> level configure script's --with-llvmgccdir option, and 2) it sets
> the LLVMGCCPATH variable which is used later in the script as the
> directory containing the llvm-gcc binary. The inconsistency is that
> the configure script seems to expect --with-llvmgccdir to refer to
> the "prefix" directory, e.g., "usr/local", without
"bin". But, if
> you set the environment variable to the prefix directory, then the
> script's reference to "$LLVMGCCPATH/llvm-gcc" will fail.
>
> Obviously there are a lot of nightly testers running successfully
> out there. I can only guess that they are not setting LLVMGCCDIR
> and instead relying on finding llvm-gcc on the PATH. Or, perhaps
> they are setting LLVMGCCDIR to the prefix directory and the script's
> attempt to invoke "$LLVMGCCDIR/llvm-gcc" fails in a non-fatal way
> (it appears to be used only to set "target_triple" in the
results).
>
> Anyway, here's my proposed patch.
> <nightlytest.patch>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev