Re: [review] Add gdb_compile_openmp to lib/gdb.exp

[Pedro Alves <palves@redhat.com>]

On 11/6/19 11:05 PM, Kevin Buettner (Code Review) wrote:

> For "pthreads", I think that we want to call gdb_compile when 
> making the .o files and gdb_compile_pthreads when making the executable. 
> If I make the change that you propose, then gdb_compile_pthreads is used
> for both executables and objects.

I suppose we could have different $func-like variables for compiling
and linking.

> I can still make this change if you want; but if I do this, I think that
> I should also add some code to gdb_compile_pthreads so that it will
> call gdb_compile (and also avoid printing the PASS message) when
> $type is not "executable".

Hmm, yes, pedantically we're supposed to compile .o files with -pthread also,
not just when linking, as it may enable predefined #defines that are necessary
for correct use of threads, like _REENTRANT, which used to be important for
glibc, though it isn't nowadays, I believe.

On a side note, it seems to me that better than calling gdb_compile_pthreads
repeatedly, it'd be better if we called a gdb_caching_proc that probed which
variant of -pthread, -lpthread, etc. is necessary and returned it, so that
build_executable_from_specs would just append the option to the set of options.
That would make gdb_compile_shlib_pthreads unnecessary, for example.  In general
these gdb_compile_foo functions been like a worse idea than options to me, since
they don't allow combinations.

Thanks,
Pedro Alves