[PATCH] [Polly] Update ISL to isl-0.15-3-g532568a

Michael Kruse llvm at meinersbur.de
Thu Jun 18 07:49:01 PDT 2015

In http://reviews.llvm.org/D10505#190234, @grosser wrote:

> In http://reviews.llvm.org/D10505#190225, @Meinersbur wrote:
> > In http://reviews.llvm.org/D10505#190085, @grosser wrote:
> >
> > > It seems you somehow changed the format of isl_config.h. When importing isl I just copied this file from an isl_config.h.in. It seems you now use a slightly different approach to obtain the contents of this file which causes changes all over this file.
> >
> >
> > It is the output of ./configure on my machine.
> >
> > Using isl_config.h.in straight away is about the least favorable thing you can do. It just defines nothing. I am surprised it worked.
> The intention was to define as little as possible. My feeling was that isl would not use (or need) most of these macros anyhow, so I wanted to leave them undefined such that it clearly breaks if there arises a need at some point.

According to that logic we should use an empty isl_config.h

> Using the configure of your machine just means it will work on your machine, but it seems there is a risk that your machine has certain features that are not available on other machines and consequently would break the build there if enabled.

It at least ensures that the macros are compatible with each other and work for gcc and clang. Using an unconfigured isl_config.h to work on any machine is pure coincidence.

> As said before, I am happy to have this improved (and with your cmake-ificaton of isl you may have some good ideas), but let's not change our approach within a massive update of isl. Those two things are orthogonal.

Not really. isl_config.h is version-specific. Upgrading ISL logically also required upgrading isl_config.h




More information about the llvm-commits mailing list