Re: role of executable_checksum & LTO?

[Jan Hubicka <hubicka at ucw dot cz>]

> Hello all,
> 
> What is the role of executable_checksum from c-common.h & generated by
> genchecksum.
> 
> In particular, in the MELT runtime plugin (actually in the MELT branch)
> I was supposing it was always defined. However, when referencing it from
> melt-runtime.c I got an undefined symbol when linking the lto1
> executable.
> 
> What is the reason of not having executable_checksum inside lto1 binary?
> I was naively supposing that writing the checksum of the compiler inside
> LTO generated data (e.g. sections in LTO emitted *.o files) is useful,
> for example to detect when a *.o LTO-generated by a gcc-4.5.0 was linked
> with a *.o LTO-generated by a gcc-4.5.1 (when it will exist).

I believe that longer term strategy is to stabilize LTO format and give
every section version (major+minor number) and do checking based on it,
to make this particular trick possible.

Checksums are used for precompiled headers and possibly some other things I am
missing.
> 
> Do we expect to be able to LTO-link a foo.o produced by gcc-4.5.0 -flto
> with a bar.o produced by gcc-4.5.1  -flto? I was believing that we don't

Probably not in 4.5 series given the shape of LTO code, but for 4.6 I consider
this one of goals.

> claim that and I was imagining that the executable_checksum could help
> on that issue.

honza
> 
> The reason I wanted to use executable_checksum is that I was thinking of
> emitting its value in the C code foo.c generated by MELT from foo.melt
> and checking its dlsym-ed value when dlopen-ing the foo.so module
> (dlopen-ed by MELT). This to warn users which uses a foo.so module built
> with a slightly different version of MELT.
> 
> Cheers.
> 
> -- 
> 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} ***
>