The future of __main (was Re: __main call for c++ on sparc-elf target)

[Jim Wilson]

On Thu, 2003-06-26 at 08:54, Chris Lattner wrote:
> I just wanted to say that I hope that __main never goes away:

I think it will go away eventually, for ELF targets at least.

Another problem with __main is that there is no guarantee that there is
a function called main.  The C standard does not require a main function
for freestanding programs.

Some of your criticisms of init/fini are fixed by the follow on
init_array/fini_array ELF sections.  init/fini contain code, but
init_array/fini_array are just lists of function addresses.  They are
similar to the ctor/dtor lists we have now, except that the code that
runs them is now in the loader instead of in the program.  I think that
solves most of your size problem.  It also makes them more transparent
to a whole program optimizer.  See the ELF standard for more info.

The GNU tools have been very slow to move to init_array/fini_array, but
there is support for them in glibc and binutils now.  So it is mostly a
matter of fixing gcc to use them exclusively now instead of init/fini
sections when possible.  There is some gcc support for this already, but
I don't know how complete it is.

Jim