[Bug libdw/22546] New: dwarf_aggregate_size() doesn't work for multi-dimensional arrays

["sourceware at dima dot secretsauce.net" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=22546

            Bug ID: 22546
           Summary: dwarf_aggregate_size() doesn't work for
                    multi-dimensional arrays
           Product: elfutils
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libdw
          Assignee: unassigned at sourceware dot org
          Reporter: sourceware at dima dot secretsauce.net
                CC: elfutils-devel at sourceware dot org
  Target Milestone: ---

Hi. I'm observing that dwarf_aggregate_size() returns bogus results when
looking at double-dimensional arrays. For instance, looking at

  double dd[3][5];

It says the aggregate is 64-bytes long instead of 120. The bug is that it ends
up computing (3+5)*8 instead of 3*5*8.

I'm attaching a simple test case. It loads the current process's debug
information, and prints out the size of dd, defined as above. It shows the
failure:

  dima@fatty:/tmp$ gcc -g -ldw -lelf dwarf_aggregate_size_bug.c -o
dwarf_aggregate_size_bug

  dima@fatty:/tmp$ ./dwarf_aggregate_size_bug                                   
  Found DIE for 'dd'. size: 64

I'm using libelf1 0.168-1 on Debian/sid, although at least version 0.170 has
the bug too.

I'm attaching a proof-of-concept patch to solve this. Note that this patch
isn't final: previously we would compute a separate stride for each dimension,
while the patch only computes the stride once. I don't know what case
specifically that is meant to handle. Tests pass before and after.

If for some reason dwarf_aggregate_size() isn't meant to be used this way, then
it should still be changed to fail in a more obvious way. Answering 64 here is
NEVER the right answer.

-- 
You are receiving this mail because:
You are on the CC list for the bug.