Re: [PATCH V2] testsuite: support mold linker

["Martin Liška" <mliska@suse.cz>]

On 12/5/22 15:08, Jan Beulich wrote:
> On 05.12.2022 14:48, Martin Liška wrote:
>> Mold linker demotes symbols like main to be local and the patch
>> adjusts expected output from nm.
>>
>> Moreover, simplify the patterns and remove accidental type 'D' that
>> is supported value for _?main functions.
> 
> Hmm, in my v1 comment I've said "accidental" to the difference, not
> to the presence of D. In fact I was expecting you to uniformly use
> [TtDd] everywhere. See also adacfc818440 (sadly without any real
> description). (That commit is also where [some of] the odd ([...]+)?
> regexp constructs were introduced.) Cc-ing Jakub in case he recalls
> background.

Oh, I'm also curious why 'D' was added in the revision.

> 
> Furthermore I now even less understand ...
> 
>> --- a/libbacktrace/Makefile.am
>> +++ b/libbacktrace/Makefile.am
>> @@ -498,7 +498,7 @@ TESTS += mtest_minidebug
>>  
>>  %_minidebug: %
>>  	$(NM) -D $< -P --defined-only | $(AWK) '{ print $$1 }' | sort > $<.dsyms
>> -	$(NM) $< -P --defined-only | $(AWK) '{ if ($$2 == "T" || $$2 == "t" || $$2 == "D") print $$1 }' | sort > $<.fsyms
>> +	$(NM) $< -P --defined-only | $(AWK) '{ if ($$2 == "T" || $$2 == "t" || $$2 == "D" || $$2 == "d") print $$1 }' | sort > $<.fsyms
> 
> ... this part of the change, where - as in v1 - you add a check for
> 'd', while a check for 't' (which supposedly is what you're after)
> was already there.

This nm invocation is supposed to save dynamic symbols (.dyns) and normal symbols ('.fsyms'),
where .fsymc contains both data and functions.

The symbol which I need to properly list is:

mtest.c:int global = 1;

which gets 'b' with mold linker.

Martin

> 
> Jan