[llvm-commits] [llvm] r61598 - /llvm/trunk/Makefile.rules

[Bill Wendling]

On Jan 3, 2009, at 10:08 AM, Dan Gohman wrote:

> On Jan 2, 2009, at 5:11 PM, Bill Wendling wrote:
>
>> =
>> =====================================================================
>> --- llvm/trunk/Makefile.rules (original)
>> +++ llvm/trunk/Makefile.rules Fri Jan  2 19:11:39 2009
>> @@ -998,8 +998,7 @@
>> $(LibName.A): $(ObjectsO) $(LibDir)/.dir
>> 	$(Echo) Building $(BuildMode) Archive Library $(notdir $@)
>> 	-$(Verb) $(RM) -f $@
>> -	$(Verb) $(Archive) $@ $(ObjectsO)
>> -	-$(Verb) $(Ranlib) $@
>> +	$(Verb) $(Archive) $@ $(ObjectsO) && $(Ranlib) $@
>
> Not that I know what's going on here, but isn't this equivalent to
> what was there before r61551?
>
I've been trying to address a strange error in the buildbots where the  
"ranlib" command seems to be applied to a .a file before that file is  
completed writing to disk. This affects the libLTO.a file the most.  
However, the libLTO.a is run through ltmain.sh instead of here, so  
this patch doesn't fix anything. My current hypothesis is that it's  
some strangeness (bug) in how Mac OS X writes out files when the ar  
command is run, because if I manually run ranlib on the same file that  
failed before, it succeeds without any problems. So the .a file isn't  
corrupted after the ar command.

I left the above patch because it doesn't do any harm, but it can be  
reverted if people object to it.

-bw