Re: [RFC] [gdb/testsuite] Use dwarf assembly in gdb.dlang/dlang-start.exp

[Tom de Vries <tdevries@suse.de>]

On 2/20/23 15:14, Iain Buclaw wrote:
> Excerpts from Tom de Vries via Gdb-patches's message of Februar 20, 2023 2:48 pm:
>> For test-case gdb.dlang/dlang-start.exp, I run into:
>> ...
>> gdb compile failed, default_target_compile: Can't find gdc.
>> ...
>>
>> AFAICT, my distro has no support for gdc, but I'd like to have the test-case
>> running and passing, so let's rewrite the test-case using dwarf assembly
>> (though arguably, it's not a bad idea to have test-cases excercising
>> actual compilers).
>>
>> My distro does have a package providing dmd, so let's try out simple.d
>> compiled with dmd, and investigate what the start command does.
>>
> 
> I'm pretty sure I'd get testsuite failures for not having Ada or Rust
> installed either. 

Yes, most likely.

> Shouldn't the testsuite just return UNSUPPORTED when
> there's a missing dependency?
> 

That's certainly an option.  OTOH, the approach taken in this patch is 
to instead eliminate the dependency, which ensures that everybody 
running the entire testsuite will exercise this test-case.

That sort of approach is not feasible for ada which has a lot of 
test-cases, and is not needed either (for me) for ada or rust because 
compilers are readily available in packages.

>> AFAICT, gdb uses "D main" because the minimal symbol _Dmain:
>> ...
>> $ nm simple | grep 438448
>> 0000000000438448 t
>> 0000000000438448 W _Dmain
>> ...
>> is demangled into "D main":
>> ...
> 
> Yes. If I try to second guess my thinking from 2016 or so. I believe it
> was because "_Dmain" could come from any language, but having a space in
> the identifier pretty much guarantees only a D compiler generated it.

I see, interesting, thanks for sharing that.

Thanks
- Tom