[PATCH] IR: Add specialized debug info metadata nodes

[Adrian Prantl]

> On Feb 6, 2015, at 3:37 PM, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
> 
> 
>> On 2015-Feb-04, at 17:17, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
>> 
>> 
>>> On 2015-Feb-04, at 16:44, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
>>> 
>>>> 
>>>> On 2015-Feb-04, at 16:17, Adrian Prantl <aprantl at apple.com> wrote:
>>>> 
>>>> 
>>>>> On Feb 4, 2015, at 3:53 PM, Frédéric Riss <friss at apple.com> wrote:
>>>>> 
>>>>> 
>>>>>> On Feb 4, 2015, at 3:46 PM, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
>>>>>> 
>>>>>> 
>>>>>>> On 2015-Feb-04, at 15:37, Frédéric Riss <friss at apple.com> wrote:
>>>>>>> 
>>>>>>>>> 
>>>>>>>>> - The word 'context' is overloaded: `MDNode::getContext()` already
>>>>>>>>> exists, and returns an `LLVMContext&`; `DIDescriptor` uses 'context'
>>>>>>>>> to mean "the node that this one is defined inside".  I chose the
>>>>>>>>> word 'parent' instead of 'context' here.  Is this word okay?  If
>>>>>>>>> not, what about 'scope'?  This will be reflected in the assembly
>>>>>>>>> changes to come (I'd like the C++ names to match the assembly names,
>>>>>>>>> although technically it's not necessary).
>>>>>>>>> 
>>>>>>>>> I'd /probably/ go with scope (we already have scope in the MDLocations, so that seems consistent), but fairly on-the-fence.
>>>>>>>>> 
>>>>>>>> 
>>>>>>>> Weirdly, I didn't even notice that :).  In that case I like 'scope'
>>>>>>>> better too.  I'll update to that before commit.
>>>>>>> 
>>>>>>> Seems most natural. Can the futur getScope() return something that doesn’t derive from MDScope?
>>>>>> 
>>>>>> I don't think so.
>>>>> 
>>>>> That was my impression also, and it makes it even more appropriate IMHO.
>>>>> 
>>>> 
>>>> Just a few general remarks to throw into the discussion:
>>>> 
>>>> - Would it make sense to use something like tablegen to generate the repetitive parts? I’m slightly worried about copy&paste bugs, moderately worried about refactoring it in the future.
>>>> 
>>> 
>>> `tablegen` might make sense here, I'm not sure.  Frankly, I've never used
>>> it before, and it's not clear to me whether it would actually reduce the
>>> amount of code in this case vs. shifting it over to utils/tablegen (given
>>> the macro usage).
>>> 
>>> I think the way the APIs are written, the only real danger is a mismatch
>>> between `MDNodeKeyImpl<>` and the current fields in the class (I believe
>>> other mismatches would be caught by the compiler).  As long as the unit
>>> tests are updated along with any refactoring that happens I think the
>>> danger even there is fairly low.
>>> 
>>> I'm open to it if there's definite value, though.  Maybe as a follow-up?
>>> 
>>> @dblaikie: I've seen you poking around tablegen internals a fair bit; any
>>> thoughts on this one?
>>> 
>>>> <classllvm_1_1DIScope__inherit__graph.png>
>>>> 
>>>> As for scopes, there are several things that bug me about the current class hierarchy that we could fix now:
>>>> - DIFile should not be a scope (the concept of files is IMO orthogonal to scoping and there is always something more appropriate to put a node into: compile unit, module, namespace)
>>>> - DIBasicType should not be scope
>>>> - It’s questionable whether a DICompositetype should be a DIDerivedType
>>>> - A DISubroutineType should be neither a DIDerivedType nor DIScope
>>>> - Using DIDerivedTypes for CV qualifiers is a bit wasteful but it does map nicely to DWARF
>>>> 
>>> 
>>> Definitely want to get to those :).  I think most of those are marked
>>> in the code as `TODO`s (except for the CV-qualifier comment).  But I'm
>>> planning to change the schema itself separately from the infrastructure
>>> for it to simplify the triage of any problems that come up (see comments
>>> in PR22264).
>>> 
>>>> otherwise, thanks for doing this!
>>>> -- adrian
>>> 
>> 
>> Here's an updated patch using "scope" instead of "parent" (and updated
>> to match r228242, from David's review of r228212).
>> 
> 
> *ping*
> 
> (While the discussion about the final structure of the class hierarchy
> and scopes/types/etc. is a good one, it's somewhat independent of the
> patch in question.)
> 
> This is a fast ping, but I'm just about blocked on this, and with all the
> discussion about the future I'm afraid everyone forgot to sign off on
> this step forward.
> 
> 
> <0001-IR-Add-specialized-debug-info-metadata-nodes.patch>

This is fine as far as I am concerned.

-- adrian