Re: [hatari-devel] Symbol loading for upcoming MINT+ELF program format

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On 29.8.2023 21.36, Eero Tamminen wrote:
> I've added some heuristics to profiler post-processor to select most 
> suitable name for given address:
> - drop 100% duplicates
> - avoid lib/object names
> - prefer shorter name for C/C++ symbols
> - prefer longer names for ASM symbols
>
> (Symbols not starting with '_' or including '::' or ' ' in their names, 
> are assumed to be ASM symbols.)
....> If above profiler post-processor heuristics produce good enough 
results,
> I'll eventually look later into adding pass with identical rules to to 
> Hatari debugger.

I've now pushed support for that.

"gst2ascii" tool will now by default do same symbol filtering as 
debugger "symbols" command.

(One can disable each filter option seprately in the tool, but that's 
mostly intended for debugging whether filtering works OK.)


> In profiler post-processor, symbol renaming could be avoided by using 
> addresses as IDs everywhere, and I'll do that eventually.

Will look into that next.

I changed Hatari to output counts of duplicate names, and there are 
surprising amounts of them. E.g. in ScummVM, there are 53 symbols with 
name "metaengine.o", of which _only one_ had better symbol name with 
same address...


> However, in Hatari debugger, users rather prefer using names for 
> things... So, either duplicate names need to be renamed, or debugger 
> needs to pick one of them at random (e.g. one with lowest address) for 
> breakpoints.


	- Eero