Re: [Qemu-devel] [PATCH 00/20] More configure/Makefile cleanups

[Blue Swirl]

On Tue, Jul 21, 2009 at 1:13 AM, <address@hidden> wrote:
> From: Juan Quintela <address@hidden>
>
> Hi
>
> More configuration cleanups on top of my previous series.
> This one:
> - when we test for a library, if we need it, we add the
>  library to LIBS after checking. Found that we were checking for 2 xen 
> libraries
>  and then adding 3 to LIBS.
> - starting moving variables to CPPCFLAGS.
> - Big SDL cleanup.  SDL configuration have very old cruft there, from
>  testing for target_sofmmu before it was set, to testing for "$sdl" = "yes"
>  on the else part of tesing "$sdl" = yes.
> - I added a configuration define: CONFIG_NOWIN32, I just needed a way to
>  compile files for WIN32 or the rest.  If anyone has a better suggestion
>  for th ename, it is welcome.

The obvious name would be CONFIG_POSIX, but are there any signs of
life in !Win32 && !Posix?

> After this series Makefile is basically clean from logic:
>
> (simple-config-4)$ grep ifdef Makefile
> ifdef BUILD_DOCS
> ifdef INSTALL_BLOBS
> (simple-config-4)$ grep ifndef Makefile
> ifndef CONFIG_WIN32
>
> ToDo:
> - do the same for Makefile.target
> - what to do with CFLAGS/CPPFLAGS.  At this point we have things like:
>    sdl.o audio/sdlaudio.o sdl_zoom.o baum.o: CFLAGS += $(SDL_CFLAGS)
>  Problem is that SDL_CFLAGS ends on fedora (I guess other unix got similar):
>    -I/usr/include/SDL -D_GNU_SOURCE=1 -D_REENTRANT
>  I felt nervous compling some files with -REENTRANT and other without it.
>  Should we use a single CFLAGS for all files?  I would prefer to have a single
>  CFLAGS, but it is not my call to make that decision.  What everybody else 
> thinks?

There is still HELPER_CFLAGS which needs to be separate from CFLAGS.

> Coments, suggestions?

These are actually wishes for new Makefile features:

It would be nice to have one flag to control gcc optimization level,
so that we could do something like
rm foo.o; make OPT=""
to recompile foo.o without gcc -O2 flag so that the execution logic
etc. is handy for debugging with GDB.

Currently all files are recompiled for each -user target. At least
path.c could be compiled only once, other files maybe could use
similar approach as libhw32/64 (poisoning etc.).