codegen: rework ABI tests to use a single C program for each test

[pbor]

This is a first step in the plan outlined in #1047

Instead of compiling a C program for each layout and one for each
constant to check, generate a single layout.c and a single
constant.c which output all the values, one per line.
On the rust side, it is easier to compile and run the C program
just once and then parse its output one line at a time to compare
with the corresponding rust record.

Note also that C programs are now taken by the system's header
files (eg atk.h) rather than from -DATK_FOO=1 passed by the
test itself. I think this is actually the right thing to check
since that is the real value we want to compare with.
However this also means that the test will fail when run on
a system older than the one used to generate the gir files, since
some constants may not present in the headers.

[pbor]

A successful run after a regen can be seen here: https://github.com/pbor/gtk-rs/runs/1845572278?check_suite_focus=true

[sdroege]

Can you put a WIP PR of gtk-rs up with the changes this would do?

[sdroege]

Note also that C programs are now taken by the system's header
files (eg atk.h) rather than from -DATK_FOO=1 passed by the
test itself. I think this is actually the right thing to check
since that is the real value we want to compare with.
However this also means that the test will fail when run on
a system older than the one used to generate the gir files, since
some constants may not present in the headers.

This was also the case before as mentioned on IRC, but AFAIU you unconfused yourself since then? :)

[sdroege]

Looks OK otherwise, even simpler and cleaner than before in gir and the generated code will be so much faster.

[pbor]

Can you put a WIP PR of gtk-rs up with the changes this would do?

gtk-rs/gtk3-rs#293

[sdroege]

Looks good to me apart from gtk-rs/gtk3-rs#293 (comment)

@GuillaumeGomez ?