Re: Best reference for understanding ELF format

[Peng Yu <pengyu.ut@gmail.com>]

On Tue, Apr 20, 2021 at 11:51 AM Carlos O'Donell
<carlos@systemhalted.org> wrote:
>
> On Tue, Apr 20, 2021 at 12:39 PM Jeffrey Walton via Libc-help
> <libc-help@sourceware.org> wrote:
> >
> > On Tue, Apr 20, 2021 at 12:29 PM Peng Yu via Libc-help
> > <libc-help@sourceware.org> wrote:
> > > ...
> > > People don't usually directly call ld to generate object, executable
> > > or shared library files.
> > >
> > > Instead, gcc/clang are called. So there is nothing wrong to say those
> > > files are produced by gcc/clang, at least at a superficially level.
> > > You can say that gcc/clang don't directly produce the ELF files.
> > > Nevertheless, this doesn't add too much to the topic of this thread.
> >
> > I think Clang can produce object files directly. Clang has an
> > integrated assembler.
>
> Conceptually the point is the same though.
>
> The compiler relies on an assembler and static linker to generate the
> object files.
>
> Whether that assembler is a library and the invocation is a library
> call, or it's a secondary process via a pipe, the concept being
> discussed is the same.
>
> The compiler itself (language front end,

I think the "front end" is actually the cc1 program which is not
exposed to PATH by default for gcc. In my system, it is the following
one, which transforms a .c file to a .s file.

/usr/lib/gcc/x86_64-linux-gnu/10/cc1

Then the .s file is processed by as, to generate a .o file. Then, the
final a.out file is generated from the .o file by ld.

The command gcc is just a wrapper of the three programs (cc1/as/ld).
The following is a minimal work example that demonstrates this.

$ cat main.c
#include <stdio.h>
int main() {
    puts("Hello World!");
    return 0;
}
$ /usr/lib/gcc/x86_64-linux-gnu/10/cc1 -quiet main.c -o main.s
$ as -o main.o main.s
$ ld -dynamic-linker /lib64/ld-linux-x86-64.so.2
/usr/lib/x86_64-linux-gnu/crt1.o /usr/lib/x86_64-linux-gnu/crti.o
/usr/lib/x86_64-linux-gnu/crtn.o main.o -lc

> or driver) doesn't produce
> ELF format object files. Technical clarity around these distinctions
> helps at a later stage when we're trying to explain other concepts
> where the distinction matters or where the distinction is muddled e.g.
> LTO, ThinLTO etc.

As these (LTO, ThinLTO) specific to the llvm pipeline?

-- 
Regards,
Peng