[Lldb-commits] [PATCH] D35740: Fix PR33875 by distinguishing between DWO and clang modules

[David Blaikie via lldb-commits]

On Sun, Jul 23, 2017 at 10:54 AM Adrian Prantl <aprantl at apple.com> wrote:

> On Jul 22, 2017, at 2:26 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
>
>
> On Fri, Jul 21, 2017 at 6:14 PM Jim Ingham <jingham at apple.com> wrote:
>
>> Not at present, but you presumably know more about this than I do.  Part
>> of the point of Greg's extracting the DWARF parser from lldb and making it
>> into it's own library in llvm was precisely so that somebody could then
>> write a simple wrapper tool that would poke it with not necessarily
>> complete but interesting canned bits of DWARF and see that it does the
>> right thing.  I thought you were involved with the reviews for that work?
>
>
> Yep yep - though not necessarily clear on the bigger picture goals in
> terms of which components were going where in the long term.
>
>
>>   I was not paying attention to the details of that effort as DWARF
>> parsing's not really my thing.
>>
>> Anyway, the extraction of the DWARF parser was Greg's last act before
>> leaving Apple, and the project stalled at that point.  I don't imagine he
>> could have gotten that code into llvm without some testing, so the kind of
>> test you are thinking of should be done using whatever mechanism you guys
>> devised for the new llvm dwarf parser.
>
>
> Adrian - any chance something like the DwarfGenerator stuff in LLVM could
> be used to test this code?
>
>
>> Of course, it's less interesting to test the llvm version of the DWARF
>> parser if lldb's not using it, so for that to be directly relevant here
>> that piece of work would need to be done.
>>
>
> Perhaps - or reusing the same testing approach without that. Though I
> think this particular failure/fix was in a higher/lower different layer
> than the pure parsing stuff in LLVM, but I could be wrong - there's
> sufficient divergence it's not obvious from a few class names, etc, to tell
> how much overlap (& where) there is.
>
>
> Yes, I would also say that this is one level above the pure parsing. This
> is how LLDB interprets the data. Once the LLVM DWARF parser (which is
> architected more for testability) is complete enough to be used inside
> LLDB, there is no reason to not also implement this level
> (cross-referencing dwarf+dwo) inside LLVM and properly test with a unit
> test or a yaml object description.
>

Any reason this would need to wait until it's sunk into LLVM? It seems like
this kind of testing would be useful to do in LLDB no matter how many
libraries are sunk into LLVM - how far off/much work will it be to have
this sort of testing available in LLDB?


> Inside LLDB an end-to-end test like the existing one is as good as it gets
> now.
>
> -- adrian
>
>
>
>>
>> Jim
>>
>>
>>
>> > On Jul 21, 2017, at 5:51 PM, David Blaikie <dblaikie at gmail.com> wrote:
>> >
>> >
>> >
>> > On Fri, Jul 21, 2017 at 4:05 PM Greg Clayton via Phabricator <
>> reviews at reviews.llvm.org> wrote:
>> > clayborg accepted this revision.
>> > clayborg added a comment.
>> >
>> > Looks like there already is a test case that was failing as Jim
>> mentioned. Accepting based on that.
>> >
>> > Ah, I was thinking more a test that would've failed when LLDB regressed
>> (regardless of whether Clang was still producing this DWARF or not) - does
>> LLDB have tests like that? (either binary, asm, or some other terse way of
>> writing DWARF directly to test "does LLDB do the right thing with this
>> DWARF" sort of tests?)
>> >
>> >
>> >
>> > https://reviews.llvm.org/D35740
>> >
>> >
>> >
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20170724/6b8e1757/attachment.html>