[Bug default/27599] New: abidiff's notion of harmless union diff may need refinement

[Bug default/27599] New: abidiff's notion of harmless union diff may need refinement

[gprocida+abigail at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=27599

            Bug ID: 27599
           Summary: abidiff's notion of harmless union diff may need
                    refinement
           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: ---

The fix to PR24731 introduced the notion of harmless union diffs.

Occasionally we check what is being reported as filtered out by abidiff and
peek in more detail with abidiff --harmless. Alas, for kernel code, this
usually generates massive diffs and it's quite hard to spot the real causes.
This bug report is about a frequent red herring in such diffs.

The file tests/data/test-diff-dwarf/PR25058-liblttng-ctl-report-1.txt
illustrates the issue on a smaller scale.

There are two diff stanzas of interest.

type of 'union {char* string; uint64_t constant; double float_constant; char*
identifier; filter_node* child;} u' changed

and

underlying type 'union YYSTYPE' changed

Both have members of type filter_node* (which has changed... somewhere).
Emitting the "harmless union diff" text here is unhelpful as union definition
is unchanged. The current test for a harmless union change is effectively
(sizes are equal and) has_changes() which amounts to an equality comparison of
the two union types.

abidiff --leaf-changes-only doesn't report these changes at all.

[As an aside and to give an idea of what I mean by "massive". I compared two
ABIs I expected to be equal.

abidiff --leaf-changes-only: 0 changes, no diff
abidiff --leaf-changes-only --harmless: 0 changes, 0 no diff
abidiff: 1861 filtered out function and variable changes, no diff
abidiff --harmless: 23848 lines of output

A second trigger appears to be some definition/declaration issues in the XML
which don't get picked up in leaf mode here either.]

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