[Bug default/27769] New: XML reader shares a corpus object between corpora

[Bug default/27769] New: XML reader shares a corpus object between corpora

[gprocida+abigail at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=27769

            Bug ID: 27769
           Summary: XML reader shares a corpus object between corpora
           Product: libabigail
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: default
          Assignee: dodji at redhat dot com
          Reporter: gprocida+abigail at google dot com
                CC: libabigail at sourceware dot org
  Target Milestone: ---

Created attachment 13396
  --> https://sourceware.org/bugzilla/attachment.cgi?id=13396&action=edit
fresh corpus object patch

Commit ebaf3305bc5adc5eb45cca2dbb9b7d94b99b6800 added a more detailed error
message around an assertion about architecture consistency.

I revisited the trigger for this today as I've been testing using whole
histories of XML versions, including a few ABIs which contained stray objects
from a different architecture. Note that the corpora in the XML were
consistently of one architecture each (as abi-instr has no architecture
attribute).

I also separately noticed yesterday that parts of the ABI get dropped if the
same abi-instr path is used in multiple corpora.

Both the assertion failure and the path collision issue go away if the XML
reader allocates a fresh corpus object for each corpus element. I have a
trivial patch to do this (and the test suite passes). However, I haven't yet
looked into whether there was an underlying reason for re-using the corpus
object.

I've attached the patch here for completeness.

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

[Bug default/27769] XML reader shares a corpus object between corpora

[gprocida+abigail at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=27769

--- Comment #1 from Giuliano Procida <gprocida+abigail at google dot com> ---
I wrote:

> the assertion failure [...] go[es] away

The assertion failure actually moves to a more sensible place,
corpus_group::add_corpus.

However, we could even consider supporting multiple architectures in a corpus
group. I'm not sure where that might make sense though.

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

[Bug default/27769] XML reader shares a corpus object between corpora

[gprocida+abigail at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=27769

--- Comment #2 from Giuliano Procida <gprocida+abigail at google dot com> ---
I did a little more digging and now have a short series of patches that address
the problem here as well as a follow-up issue, plus a tiny bit of DWARF reader
cleanup. The patches are together as they all relate to corpus object creation.

See https://sourceware.org/pipermail/libabigail/2021q2/003396.html.

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

[Bug default/27769] XML reader shares a corpus object between corpora

[gprocida at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=27769

gprocida at google dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gprocida at google dot com

--- Comment #3 from gprocida at google dot com ---
The patches, or a later version of them, were applied.

This can be closed as fixed.

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