* I think that another mass test is due. That should give me a
stream of stuff to fix.
Agreed. If the reloc support is ready for testing, this will be our first
mass test using any ET_REL files at all. Also, the archive now (when it
finishes updating) has for the first time many binaries produced by gcc
4.4. That will probably be the oldest gcc version in which we try to get
any DWARF anomalies fixed, so for producer diagnosis that part of the data
will be our main focus in the future.
We know at least that we'll see lots of aranges anomalies in the older
binaries (if not still 4.4 ones). That guarantees that in the simple mass
test, many thousands of cases will have output and nonzero status.
The test script leaves its state around for each failure. I think I have
enough disk space on the test box that it won't be a problem to have
thousands of failures, and it shouldn't too bad for me to upload just the
logs. But sorting them all out might be a pain.
If you don't already, perhaps it is worthwhile to have a switch suppress
the semantic checks we already know are common failures. aranges!=ranges
is the only one I recall off hand, is it the only one?
I'll leave the choices up to you, and then I'll leave the sorting through
the logs all up to you too. ;-) Just say what set of switches to use for
* Or go through the spec and re-tag attribute expectedness.
Let's wait on this one. I have two reasons in mind. First, I think what
we'll really want is a more sophisticated set of checks/tables than just
individually required/expected/optional/unexpected. (We can discuss those
details.) Second, this is off the critical path for writer/compression
work and checking its correctness. This is very valuable stuff for testing
and fixing the compiler et al, but not the highest priority for us.
I suspect your elfutils time allocation this week will be taken up just
with existing loose ends and testing fallout. If not, or to plan for next
week, let's look to that critical path. I think perhaps dwarflint now
covers everything for correctness of the writer per se (vs producer
semantics issues). Are there any holes I've forgotten?
If dwarflint is "complete" as a checker for the writer, then it's time to
shift more dwarflint features/polish into background tasks and look at some
of tasks in the writer itself. I will try to soon do some brain dumps
about all the pieces whose design I've contemplated, so we can talk about
what to work on next.