If this were to extend the AutotoolsPackage would you be interested in using the upcoming logic in #6415? To use that, you would set flag_handler = build_system_flags in the ipopt package and declare the class with class Ipopt(AutotoolsPackage):; this would ensure that configure is invoked with CFLAGS=-g -O0, so you wouldn't have to set CFLAGS= to avoid a conflict with the compiler wrapper flag injection.

I think that sounds like a better solution to me. I can take another shot after #6415 is merged.

It looks like hypre might be another candidate for the flag_handler. It looks to me like default flags are conflicting with the compiler wrapper flag injection.

EDIT:
Scratch that - it looks like only the compilation of the configuration test of hypre has the combination of some default flags and spack flags. I'll go ahead and make issues for the packages I find that have conflicting flags (including this one).

@junkudo #6415 has now been merged so you can try out the suggestion at #6714 (comment).

Another possibility is if you don't want to pass the compiler flags as arguments to configure but rather set up environment variables like CPPFLAGS=... before you invoke configure. This can be done by setting flag_handler = env_flags in the Ipopt package.

@scheibelp if you can, could you take a look at my prospective update to this PR at junkudo/spack@2b948b8? I'm running into an issue when I try to actually install ipopt with the same command as before with the new package.py. I'm pretty sure that has nothing to do with flag_handler and everything to do with my conversion to AutotoolsPackage.

For this new package.py, I see in the build output:

==> 'make' 'test'       
...
cd test; make test                                                                                                                                                                                          
make[2]: Entering directory `/g/g14/kudo4/spack/build/spack-stage/spack-stage-j5vFDb/Ipopt-3.12.7/Ipopt/test'                                                                                               
make[2]: stat: hs071_main.cpp: Too many levels of symbolic links                                                                                                                                            
make[2]: *** No rule to make target `hs071_main.cpp', needed by `hs071_main.o'.  Stop. 
...

In the existing package that does not use Autotools:

==> 'make' 'test' 
...

cd test; make test                                                                                                                                                                                          
make[2]: Entering directory `/g/g14/kudo4/spack/build/spack-stage/spack-stage-27zjLT/Ipopt-3.12.7/Ipopt/test'                                                                                               
if /g/g14/kudo4/spack/lib/spack/env/intel/icpc -DHAVE_CONFIG_H -I. -I`echo .` -I../src/Common  -I`echo ./../src/Common` -I`echo ./../src/LinAlg` -I`echo ./../src/LinAlg/TMatrices` -I`echo ./../src/Algori\
thm` -I`echo ./../src/Interfaces`   -O3 -ip -mp1  -DNDEBUG    -DIPOPT_BUILD -MT hs071_main.o -MD -MP -MF ".deps/hs071_main.Tpo" -c -o hs071_main.o hs071_main.cpp; \                                        
then mv -f ".deps/hs071_main.Tpo" ".deps/hs071_main.Po"; else rm -f ".deps/hs071_main.Tpo"; exit 1; fi

This appears to be the same issue as #3387 where a self-referential symbolic link is created by configure.

If you set build_directory = 'spack-build' (as in #3388) is the issue resolved for you?

I should admit I'm not sure why this resolves the issue; I'd be interested in figuring that out if it is confirmed that this resolves the current issue in this thread.

Adding `build_directory = 'spack-build' breaks the build much earlier. It now fails right at the start of make:

configure: Main configuration of Ipopt successful                                                                                                                                                           
==> Executing phase: 'build'                                                                                                                                                                                
==> 'make'                                                                                                                                                                                                  
make: *** No targets specified and no makefile found.  Stop.  

Thoughts?

You also have to update the definitions of build/install if you implement them and set build_directory (those functions start by default in the source base directory, so you would have to change to spack-build). As it happens, the AutotoolsPackage definition handles calling make install and make test automatically, and if you set parallel = False as a class attribute of ipopt the make invocations will also be serial. So I think all you need to do is remove the build/install definitions.

@scheibelp - seems like setting the build_directory to spack-build solved the problem with the self-referential link. I updated the PR.

@scheibelp small bump 😄 Any chance you've looked at this?

To be clear are all issues resolved (i.e. it's doing what you want like not overriding flags) and you are asking if I can merge it?

Yes to both. The issue is resolved and the flags are being passed in through via the build system flags. Like you previously advised, I did have to set build_directory = spack-build.

Thanks!