[llvm] r186035 - Add a comment.

David Blaikie dblaikie at gmail.com
Wed Jul 10 14:55:45 PDT 2013


On Wed, Jul 10, 2013 at 2:47 PM, Adrian Prantl <aprantl at apple.com> wrote:
>
> On Jul 10, 2013, at 2:14 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
>> On Wed, Jul 10, 2013 at 2:08 PM, Adrian Prantl <aprantl at apple.com> wrote:
>>> Author: adrian
>>> Date: Wed Jul 10 16:08:02 2013
>>> New Revision: 186035
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=186035&view=rev
>>> Log:
>>> Add a comment.
>>
>> Is this the minimal source required to reproduce the issue? Is it
>> really dependent on templates?
>
> It’s a backend testcase, so obviously the answer is no,

What I mean is: could we still test this bug with an example that
doesn't have templates in it. (ie: is the source code minimized)

> but I’m not convinced that manually instantiating templates in a comment in a test case is the best use of anyone’s time :-)

Not sure I follow, but maybe: I do think it's a valuable use of
someone's time to provide simple test cases. The way I'd prefer to see
debug info test cases simplified is by simplifying the input source,
not hand-simplifying the IR. This will keep them small, but
maintainable.

>>
>> I'd like to have the source reduced & the IR left untouched if
>> possible - rather than manually reducing the IR from more complex
>> source. This'll make it easier to regenerate the test case if it needs
>> to be tweaked/enhanced in the future.
>
> My take on this is that if someone _really_ wants to tweak this test in the future they will probably be happier with the actual source code that the test was generated from.

Not sure I follow you here. Your sentiment ("happier with the actual
source code that the test was generated from") is one I agree with, so
I'm not sure where we're misunderstanding each other. When I say a
"reduced" I mean actually usable source (can be compiled, does
demonstrate the bug) but just as simple/small as possible - ie: if the
bug can be reproduced with one class instead of two, with classes
without templates, etc, then we should do that.

>
> Let me know what you think -- if you feel strongly about it I can build a reduced-C++/full LLVM version if the thing.
>
> cheers,
> adrian




More information about the llvm-commits mailing list