Re: [RFC][patch] Allow to disassemble line.

[Paul Pluzhnikov <ppluzhnikov at google dot com>]

On Thu, Oct 1, 2009 at 11:52 PM, Joel Brobecker <brobecker@adacore.com> wrote:

>> Attached patch makes it so 'disas/l' will disassemble current line, and
>> 'disas/l foo.c:22' will disassemble line 22 of foo.c
>
> Without looking at the implementation itself for now, I have to say that
> I have been missing this feature very badly.
>
> Questions:
>
>   1. What should we do if there are more than one match for the given SAL?

I only ever needed this feature while single-stepping through the code
(i.e. this line crashes, but why?).

Perhaps it is reasonable to get rid of the parameter, and just say that
'disas/l' always disassembles current line, thus eliminating the ambiguity.

Alternatively, if there are two code segments matching 'foo.c:22', but
one of them is inlined into the current function and the other is inlined
somewhere else, then clearly the user is interested in the "current" one
(same for templates).

>   2. A little trickier: How do we want to handle the case where a line
>       of code is split in more than one block of instructions.   This happens
>       really often when debugging optimized code.
>
>       Right now, the easy solution is to only disassemble the first
>       block.   It'd be nice to have them all, though. Perhaps printing
>       the instructions for each block one after the other, with something
>       like a little sign in between indicating the next instruction is
>       part of another block?
>
>       0x... <fun_name+nn>      bla bla bla
>       0x... <fun_name+mm>      bla bla bla
>       [...]
>       0x... <fun_name+oo>      bla bla bla

Some alternatives:

A) determine min(low), max(high) core address for all SALs, then disassemble
   that entire range, but there are discontinuous source lines turn on
   DISASSEMBLY_SOURCE automatically, so it becomes clear which instructions
   come from which line(s).

   The trouble with this approach is that the two blocks could be quite
   far apart. I haven't see GCC do that, but under MSVC I've seen wildly
   discontinuous blocks of code. I think GCC may soon start doing hot/cold
   code splitting as well, especially with FDO.

B) disassemble only the SAL which "covers" current $pc (as that's likely
   what the user is looking for). This would be consistent with proposed
   elimination of parameter, so 'disas/l' always implies 'current $pc'.

-- 
Paul Pluzhnikov