[RFC] Fortran-intern "tree browser"

[Daniel Kraft]

Tobias Schlüter wrote:
> 
> Hi Daniel,
> 
> Daniel Kraft wrote:
>> I'm just in the middle of a debugging-session for PR 37099, and find 
>> it somewhat tedious to analyse the gfortran-internal trees (mostly 
>> gfc_expr) via gdb's print dump.  Especially wide-character strings and 
>> things like gmp-constants are quite hard to figure out by hand...
> 
> It should be possible to call gfc_show_expr() and the like from the 
> debugger.  At least I put a patch into the tree for exactly this 
> purpose.  This should address most of your ideas.

Hm, I tried this and am getting an internal error ("don't know how to 
show expr") from the call and the program being debugged aborts...  I'm 
not sure but to me the expression looks rather ok for showing.

Well, maybe I'll figure out what's wrong there...

Thanks,
Daniel

> Cheers,
> - Tobi
> 
>> What do you think about implementing something like GIMPLE's 
>> "browse_tree" functionality?  A global function, that can be called 
>> right away from gdb's prompt and handed in a gfc_code/gfc_expr (or 
>> however we want it), and that helps in analysing it?  This could be 
>> interactive as browse_tree is (although I'm not sure this is 
>> necessary) or at the very simplest case just use the routines from 
>> dump-parse-tree.c to dump the expression readable to stdout.
>>
>> But maybe some other format, or an extended mode for dump-parse-tree, 
>> could be useful, too.  For my current session, I've spent a lot of 
>> time looking at string-lengths in typespecs.  I think for 
>> debugging-output, the type-spec in some form should be available, 
>> too...  But that's just an idea.
>>
>> What do you think about this idea in general?  Or is there already 
>> something one can use for this purpose?  I'd like to hear your 
>> opinions on how this should be done and what features you would 
>> expect.  If you like the idea, I'm volunteering to work something out 
>> for proposing ;) BTW, would such a change have problems getting into 
>> stage3?  But if it is something not visible to ordinary users but only 
>> from the debugger, it should not cause any risks, of course.
>>
>> Yours,
>> Daniel
>>
> 
> 


-- 
Done:     Arc-Bar-Cav-Sam-Val-Wiz, Dwa-Elf-Gno-Hum-Orc, Law-Neu-Cha, Fem-Mal
To go:    Hea-Kni-Mon-Pri-Ran-Rog-Tou