Re: Dynamic array bounds problem

[Jonas Maebe <jonas dot maebe at elis dot ugent dot be>]

On 24 Dec 2008, at 16:22, Jan Kratochvil wrote:

> On Wed, 24 Dec 2008 11:11:12 +0100, Jonas Maebe wrote:
> > struct {
> >  intptr_t refcount;
> >  intptr_t high;
> >  char data[1];
> > } dynarr_t;
> >
> > A dynamic array variable is always a pointer which is either NULL  
> > (not
> > allocated) or which points to the "data" field of the above struct  
> > (so
> > the length and refcount are at negative offsets). The lower bound is
> > always 0, the current upper bound is stored in the "high" field.
>
> Just be sure DW_TAG_variable's DW_AT_location gives the address of  
> the pointer
> to dynaddr_t, not the pointer value itself.

This is correct correct.

> This GIT tag should work (although there is an update now).  You  
> should check
> that TYPE_DYNAMIC is present in the sources.

That's a very good point to make. I tried to follow the instructions  
at http://people.redhat.com/jkratoch/vla/MOVED, but my git (the  
default one from Scientific Linux 5.2, not sure which version that it  
is) didn't recognise the "-tb" option for the checkout command (it  
gave an error). I then left off the "t" and used "checkout -b", and it  
'worked' (in the sense that there was no error message anymore). But  
there was no output as far as I remember, so this may just as well  
have done something completely different than integrating the changes  
from your branch.

The kicker is that I cannot check whether or not TYPE_DYNAMIC is  
present in the sources I compiled, because the machine on which that  
checkout resides has apparently been turned off for the holidays.

However...

> > Any hints on what I might be doing wrong?
>
> Some basic hints can be given by `set debug target 1' to at least  
> see which
> addresses are being read while evaluating the array boundaries.

I've installed git on my MacBook, checked out everything there (and  
the git from Fink apparently does know -tb, so now I also know that it  
means "track branch") and compiled it there (it's great that Mac OS X  
support is being added to mainline gdb!). I did notice that  
TYPE_DYNAMIC was not in the sources before I explicitly merged in the  
archer-jankratochvil-vla branch though, so it seems I misread that it  
was already merged (looking again, I guess that it was merged into the  
archer-rmoseley-demo-merge branch rather than in the archer branch).

Anyway, now instead of a [0..0] array I got error messages from gdb  
stating that it was trying to read invalid memory locations (so my  
previous checkout probably did not contain TYPE_DYNAMIC). With the  
"set debug target 1", I discovered that my upper bound expression was  
wrong: it was missing an initial deref.

So I changed it into the following (no readelf on Mac OS X though, and  
its copy of dwarfdump doesn't fully decode dwarf opcodes unfortunately):

0x000000ce:         TAG_subrange_type [7]
                     AT_lower_bound( 0x00000000 )
                     AT_upper_bound( <0x5> 97 06 34 1c 06  )
                     AT_type( {0x000000f7} ( LONGINT ) )

That upper bound expression is the following:

DW_OP_push_object_address; DW_OP_deref; DW_OP_lit4; DW_OP_minus;  
DW_OP_deref

And now everything works perfectly!

Thanks a lot for your help,


Jonas