gas: set_symtab()'s setting of BSF_KEEP

gas: set_symtab()'s setting of BSF_KEEP

[Jan Beulich]

Alan,

while looking into how to address an Arm64 regression I ended up
coming across this code, originating from commit 76d129399558. The
first thing that puzzled me was the != for checking BSF_SECTION_SYM.
Shouldn't this be a check of just the single bit? The other thing
looking slightly odd was the mix of || and !( && ). After
transforming the whole thing to

	    if (!(asympp[i]->flags & BSF_SECTION_SYM)
		|| !bfd_is_const_section (asympp[i]->section)
		|| asympp[i]->section->symbol != asympp[i])
	      asympp[i]->flags |= BSF_KEEP;

it then occurred to me that the last part of the condition ought
to be redundant with the combination of the first two. Or is there
a way for *ABS* to have more than a single symbol having
BSF_SECTION_SYM set on it?

Thanks for any further insight,
Jan

Re: gas: set_symtab()'s setting of BSF_KEEP

[Alan Modra]

On Wed, May 04, 2022 at 10:54:25AM +0200, Jan Beulich wrote:
> Alan,
> 
> while looking into how to address an Arm64 regression I ended up
> coming across this code, originating from commit 76d129399558. The
> first thing that puzzled me was the != for checking BSF_SECTION_SYM.

The test is to prevent any attempt to change the const global_syms[].
That's why flags != BSF_SECTION_SYM is quite OK.  It would be cleaner
to write a bfd_is_const_symbol similar to bfd_is_const_section.

-- 
Alan Modra
Australia Development Lab, IBM