On 25/09/2019 23:16, Luc Van Oostenryck wrote:
> On Fri, Apr 26, 2019 at 05:46:22PM +0100, Ramsay Jones wrote:
>> On 26/04/2019 16:20, Randy Dunlap wrote:
[snip]
>> However, I should have made it clear that I meant to suggest
>> that we should _replace_ the definition of LLVM_CFLAGS with
>> that single call to llvm-config. So, I also tested the
>> following on cygwin:
>>
>> $ git diff
>> diff --git a/Makefile b/Makefile
>> index f816a50..7e8c2ab 100644
>> --- a/Makefile
>> +++ b/Makefile
>> @@ -165,8 +165,7 @@ ifeq ($(shell expr "$(LLVM_VERSION)" : '[3-9]\.'),2)
>> LLVM_PROGS := sparse-llvm
>> $(LLVM_PROGS): LD := g++
>> LLVM_LDFLAGS := $(shell $(LLVM_CONFIG) --ldflags)
>> -LLVM_CFLAGS := -I$(shell $(LLVM_CONFIG) --includedir) \
>> - $(shell $(LLVM_CONFIG) --cppflags)
>> +LLVM_CFLAGS := $(shell $(LLVM_CONFIG) --cppflags)
>> LLVM_LIBS := $(shell $(LLVM_CONFIG) --libs)
>> LLVM_LIBS += $(shell $(LLVM_CONFIG) --system-libs 2>/dev/null)
>> LLVM_LIBS += $(shell $(LLVM_CONFIG) --cxxflags | grep -F -q -e '-stdlib=libc++' && echo -lc++)
>> $
>>
>> ... which also works! The only difference is that '-I/usr/include'
>> is not passed to gcc twice.
>>
>> Looking at commit 65840c61dc ("build: only need includedir from
>> llvm-config", 2018-12-18), Luc only wanted the 'pre-processor'
>> flags not all of the '--cflags'. The '--cppflags' argument to
>> llvm-config is used for just that. ;-)
>>
>> If you could confirm that the above works for you also, that
>> would be great. Thanks!
>
> Yes, I agree and applied.
> Thank both and really sorry for the delay.
Welcome back Luc. ;-)
[BTW, we are still in 'RC' period! I have been using sparse
version v0.6.1-rc1-7-g2b96cd8 in 'production' all this time ...]
I just fetched sparse and tested master@3c74809, and it failed
on cygwin!
At first I thought it was this patch (ie it was all sparse-llvm
tests which failed), but I was equally sure I had tested back
when Randy sent the patch (_not_ just compile tested).
However, if I go back to v0.6.1-rc1-7-g2b96cd8, it also fails, so
this is not the smoking gun. Then I remembered that the cygwin
package for clang/llvm was updated recently ...
Back when Randy sent the patch I was on clang/llvm v3.x.y (I _think_
it was 3.0.1, but don't quote me), but now I am on v8.0.1.
I haven't looked into the errors/warnings yet, but they seem to
relate to 'unknown/misused' pseudo-ops used for debug info.
Just a heads up! (If Randy hasn't also updated, he may well still
be fine).
ATB,
Ramsay Jones
