[PATCH] Add -bare option to llvm-objdump

[Steve King]

On Thu, Sep 11, 2014 at 9:51 AM, Rafael Espíndola
<rafael.espindola at gmail.com> wrote:
>>> * What option is used to enable/disable it?
>>
>> Symbolization is the default, -bare turns it off.
>
> On which platform?
>

All platforms.


>>
>> Another point about -bare and diagnostic tests:  scripted checking for
>> expected output is a different circumstance than human eyeballs
>> looking at disassembly.  Scripted checking can break when we make
>> human friendly aesthetic improvements to the output.  Cleaning up
>> these character mismatches each time doesn't improve test coverage,
>> but definitely creates more work.
>
> That is a general statement, I don't think it applies in here. Most checks
> using llvm-objdump don't seem to be looking a at jmp or a call which
> would be changed with symbolization.


Yes, but looking beyond jmp/call symbolizing?  For example, Sean's wish list?


>
>> That said, I'm ready to back off on this idea of per-platform defaults
>> if it's not acceptable to you and Sean.
>
> IMHO each feature is a different case. In the case of symbolization, as you
> mentioned it is *always* enabled on every platform. So in this case I think
> we should follow the system default: always enable symbolization in
> llvm-objdump.


In other words, no -bare option.


>
>>  If you're asking me to comb
>> through and alter a large number of diag tests as a precursor to any
>> symbolizing work, I'll have to sleep on it and let you know.  As we
>> make incremental output improvements, some tests will break
>> repeatedly.  Maybe somebody else could take on this part?  -bare at
>> least solves this problem efficiently, so we can get symbolizing
>> working.
>
> It really seems you are overstating this. Relatively few tests use llvm-objdump,
> fewer use the -d option and very few (any?) depend on it not symbolizing
> the output.


The patch shows 66 tests that use disassembly in some way.  I have no
feel for how many will break or the time investment to fix each
problem.  Should I just try jmp/call symbolization and see how it
goes?