Re: Patch: automatic dependencies for gcc

[Basile STARYNKEVITCH <basile@starynkevitch.net>]

Hello All,

Tom Tromey wrote:
>>>>>> "Basile" == Basile STARYNKEVITCH <basile@starynkevitch.net> writes:
> 
> Basile> MELT is able to generate C code and to tldl_dlopenext (almost like
> Basile> dlopen-ing a dynamic library) the binary dynamically loadable stuff (a
> Basile> *.so ELF shared object on Linux & Solaris...) obtained from the
> Basile> generated code. Notice that this code generation and dynamic loading
> Basile> may happen within cc1 (for some peculiar long-lasting passes and
> Basile> compile flags).
> 
> Ok.
> 
> Are you plugging passes into a cross compiler?

Yes, MELT is basically a [possibly generated] plugin stuff. The passes 
are at the GIMPLE level, so it could be a cross compiler.

> 
> It seems to me that if you are only working with native compilers,
> then it would be easy to just build these objects with the installed
> gcc, and that would let you simplify the build stuff a bit.
> 
> Basile> run-basilys.d: run-basilys.h \
> [...]
> 
> Cute trick :-)
> 
> Basile>  Maybe using depcomp could help, but I don't know how.
> 
> For depcomp, you must first discover "dependency mode" of the compiler
> you plan to use.  There's some configure code to do this.  Then you
> run it as a wrapper around the compiler, with some variable settings
> to tell it what to do.
> 
> If you're doing this stuff in the context of gcc/Makefile, you can
> just use the infrastructure there.  It will already know properties of
> the compiler.
> 
> If you're doing this at gcc-runtime, well... simplest would be to make
> the user pick a compiler, preferably gcc; then just always use that
> one and code its properties into whatever Makefile you use at runtime.

The gcc/Makefile.in of the MELT branch generates a script 
melt_compile_script=$(libexecsubdir)/melt-cc-script
(the $(libexecsubdir)/melt-cc-script path is its installed place)
and this script compiles the generated code with the appropriate flags 
(mostly -fPIC -shared and -I....meltincludedir....

This script may be invoked by cc1 when a C file has been generated. This 
script is supposed to produce a dynamically loadable stuff (e.g. *.so 
shared object on ELF/Linux) which is then loaded by ltld_dlopenext. All 
this (the *.c generation, the script invocation to compile it into a 
*.so, the dlopen-ing of this *.so) happens in the same cc1 process.
> 
> But, I'm not sure I fully understand what you're trying to do.
I tried to explain it here, and with a little bit more details on the 
Wiki. Please tell me what is not clear enough.

A big thanks for your comments.

PS: I don't know if this discussion belongs better to the gcc@ list or 
to the gcc-patches@ list!
-- 
Basile STARYNKEVITCH         http://starynkevitch.net/Basile/
email: basile<at>starynkevitch<dot>net mobile: +33 6 8501 2359
8, rue de la Faiencerie, 92340 Bourg La Reine, France
*** opinions {are only mines, sont seulement les miennes} ***