[Lldb-commits] [PATCH] D47708: PDB support of function-level linking and splitted functions

[Leonard Mosescu via lldb-commits]

>
> For this particular case, I don't think having a split function is
> particularly interesting.


I agree that conceptually all disjoint ranges are the same. Until they are
not: I haven't looked closely at the code changes but it's easy to
introduce assumptions about the layout hierarchy
(range->function->segment->section->module)

Even if the code doesn't do this today it seems a good idea to have check
to catch attempts to introduce such assumptions in the future.
If these are intended to be regression tests it seems even more important
to do so.

Also, for either regression on integration tests I think it's a good idea
to include the common real world cases in addition to synthetic test cases.

That being said, these are just some drive-by comments and I'm not opposed
to getting this change in.




On Fri, Jun 8, 2018 at 12:23 PM, Zachary Turner <zturner at google.com> wrote:

> Ahh sorry, I jumped in kinda late and the thread was already quite long so
> I didn’t read everything. It would probably be some overhead to learn how
> to write the asm files but you can probably have clang-cl generate one for
> you and just move the directives around
> On Fri, Jun 8, 2018 at 12:18 PM Pavel Labath <labath at google.com> wrote:
>
>> On Fri, 8 Jun 2018 at 19:58, Zachary Turner <zturner at google.com> wrote:
>> >
>> >
>> >
>> > On Fri, Jun 8, 2018 at 11:52 AM Pavel Labath <labath at google.com> wrote:
>> >>
>> >> On Fri, 8 Jun 2018 at 18:48, Leonard Mosescu <mosescu at google.com>
>> wrote:
>> >> >
>> >> > To me, a linker order file (of any linker) sounds like a good
>> >> abstraction level for generating that kind of input (on linux, I might
>> >> have preferred a .s file with hardcoded .loc directives, but that
>> >> doesn't seem to be a thing on windows).
>> >
>> >
>> > Why not?  I think that would work fine (it’s called .cv_loc in codeview
>> though)
>>
>> Because I didn't know that's possible? :D
>>
>> I dropped a remark about .s files several pages back but noone picked
>> it up, so I assumed it's infeasible for some reason. If that is the
>> case, then this really sounds like a good idea. This way we should be
>> even able to generate the split function without relying on PGO or
>> checked-in binaries.
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20180608/96b97e95/attachment.html>