This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFA] unexpected multiple location for breakpoint


> So, if I understand you well, you suggest that we rework a bit the
> "contained-in" elimination loop to avoid lowest-address assumptions.
> If 2 different PCs are inside the same block, then keep the lowest
> address.  If PC1 is in BLOCK1 and PC2 in BLOCK2, and BLOCK2 is contained
> in BLOCK1, then PC2 should be discarded, even if PC2 > PC1. (and
> if the blocks are distinct, then keep both PCs).

Answering myself, I have another situation, which is a bit unrelated,
but could still be interesting in this discussion:

The source itself is a while loop

  52.   while CONDITION loop
  53.      Do_Something (...);
           [some other code]
  71.      delay Random_Time;
  72.   end loop;

We are trying to break on line 53.

This source was compiled at -O1, and part of line 53 got hoisted outside
of the actual loop.  The rest of line 53 is scattered around the rest
of the code in the actual loop.  Something like this:

        // line 53
        [...]
        // line 56, 62
        [...]
        //line 53:
        cmp ...
    .LL40:
        [...]
        // line 53:
        [...]
       
        (snip)

        // end of loop
        cmp ...
        jmpne .LL40

As a user, we're expecting the debugger to stop on line 53 at every
iteration.  But because we only break on the first instance of line 53,
and because that first instance is outside the actual loop, we end up
stopping only once.

(there are some lexical blocks in the assembly code I'm looking at,
but nothing significant to line 53, I believe).

According to you, bug or feature?

>From a user's perspective, I think it is hard to accept this as normal
or expected, since a few "next" operations instead of a "continue" shows
that the debugger gets back to line 53 several times during the loop
execution.  If this is a bug, then the only solution I can think of is
inserting a breakpoint at *every* instances of line 53, regardless of
lexical-block relationships.

Thoughts?

-- 
Joel


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]