Re: [PATCH] more adjustments to elf_find_function

["Jan Beulich" <JBeulich at novell dot com>]

>I am not sure that this patch is quite right.  There are two possible 
>problems with it:
>
>1) It does not apply the same adjustment to arm_elf_find_function in 
>elf32-arm.c.

That could be easily addressed; I wonder, however, why the same
(generic) code exists in two places. I didn't even know there's a second
instance of it, and for such arch-specific files I'd view this as a task
the maintainers of the arch would have to carry out (after all it must
have been them to decide the duplicate this and perhaps a lot more
functionality).

>2) I think that the transition from state==nothing_seen to 
>state==symbol_seen might be in the wrong place.  What happens if the 
>symbols encountered are in this order:
>
>   STT_LOPROC
>   STT_FILENAME
>   STT_FUNC
>
>The first symbol will change the state to symbol_seen, but will not
set 
>'func'.  The second symbol will change the state to 
>file_after_symbol_seen so that when the third symbol is encountered 
>'filename' will be set to NULL and no filename will be reported.
>
>I think that the change from state nothing_seen to state symbol_seen 
>ought to be inside the switch() statement, at the end of STT_FUNC. 
What 
>do you think ?

I think it's right (at least I intended it to behave exactly as you
describe it). In that place I can't judge about the meaning of symbols
between STT_LOPROC and STT_HIPROC anyway, so considering them 'normal'
symbols seemed more obvious to me. If an arch indeed has a symbol type
that needs to be ignored here, then a new hook would be needed.
In any case the state change can't be at the end of the STT_FUNC case:
STT_OBJECT and STT_TLS (as well as any future types) ones would then be
mis-treated, and especially wrt. future extensions I used the assumption
that those (see the relatively new STT_TLS) would be 'normal' rather
than 'special' in the sense used here.

Jan