[PATCH] D33804: bugpoint: Prototype disabling symbolication of bugpoint-executed programs

[David Blaikie via llvm-commits]

On Thu, Jun 8, 2017 at 9:15 PM Daniel Berlin <dberlin at dberlin.org> wrote:

> On Thu, Jun 8, 2017 at 8:57 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
>>
>>
>> On Thu, Jun 8, 2017 at 8:11 PM Daniel Berlin via Phabricator <
>> reviews at reviews.llvm.org> wrote:
>>
>>> dberlin accepted this revision.
>>> dberlin added a comment.
>>> This revision is now accepted and ready to land.
>>>
>>> I think this is the right thing to do given what we've seen.
>>
>>
>> Dandy - thanks. I'll see if I can figure out some test coverage for this
>> (find a way to crash things, ensure the test works in all configurations,
>> etc) check it in and then add in the support for the other commands
>> (llc/lli I think) that bugpoint runs directly.
>>
>>
>>> Nobody has come up with a use case where they current crawl backtraces
>>> in their scripts, etc.
>>>
>>
>> In their scripts it'll be up to them to pass or not pass this flag (does
>> mean if they want speed they'll have to know about this flag to pass it to
>> avoid the symbolizing overhead).
>>
>
> It looked like you spawn bugpoint passes with it?
> (IE when do they have to pass it)
>

Ah, right - for opt (which can be overridden with a user provided command -
and yeah, in that case the author of that custom command would have to
opt-out or otherwise handle this new flag). I haven't delved into all the
commands bugpoint can run, but some are a bit more vague/open-ended
"compile command" (well, that defaults to llc, so perhaps could be handled
similarly to opt - pass the flag by default & let the user figure it out).

I had thought maybe some of these entry points might be vague enough that
it wouldn't make sense to be passing this new flag to user-specified
commands. But I'm not 100% sure yet, haven't delved into all of them.

To be honest just changing opt addresses my test slowdown issue - but I
intend to change the others for consistency rather than leaving a landmine
for someone else.


>
> I'm referring to this:
> bool BugDriver::runPasses(Module *Program,
>                           const std::vector<std::string> &Passes,
>                           std::string &OutputFilename, bool DeleteOutput,
>                           bool Quiet, unsigned NumExtraArgs,
>                           const char *const *ExtraArgs) const {
> ...
>
>   Args.push_back("-disable-symbolication");
> ...
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170609/3df346b9/attachment.html>