llvm dev - Mar 2019 - DWARF debug info unit tests: Tests for wrong triple?

The DWARF debug info unit tests in llvm/unittests/DebugInfo/DWARF/ forms a
generator based on the host triple. This is problematic for cross-compile
configurations where LLVM does not fully support code generation for the
host platform.

I have found no rationale for why the host triple was chosen (ref:
https://reviews.llvm.org/D27326?id=80002#inline-236026).

I would imagine that it is more useful to test for the default target
rather than the host.

-- HT
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190320/5b8abd12/attachment.html>

Adding in aprantl here to make sure he's aware.

Making sure I understand - you're building a compiler on machine a
that llvm doesn't support code generation to and running the llvm
tests on that platform?

That's probably a bit out of what any of us expected here as the
typical one is to use build llvm on a platform that it already
supports. I do agree that default target would also work here as a
suitable test.

-eric

On Wed, Mar 20, 2019 at 3:47 PM Hubert Tong via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
>
> The DWARF debug info unit tests in llvm/unittests/DebugInfo/DWARF/ forms a
generator based on the host triple. This is problematic for cross-compile
configurations where LLVM does not fully support code generation for the host
platform.
>
> I have found no rationale for why the host triple was chosen (ref:
https://reviews.llvm.org/D27326?id=80002#inline-236026).
>
> I would imagine that it is more useful to test for the default target
rather than the host.
>
> -- HT
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

Am I right in assuming that the default target is the host triple for a typical
non-cross compiler that most of us (or rather our bots) typically build? If yes
then it sounds like changing the default triple to the default target would be a
noop for most of us and thus a reasonable change?

-- adrian
> On Mar 20, 2019, at 4:28 PM, Eric Christopher <echristo at gmail.com>
wrote:
> 
> Adding in aprantl here to make sure he's aware.
> 
> Making sure I understand - you're building a compiler on machine a
> that llvm doesn't support code generation to and running the llvm
> tests on that platform?
> 
> That's probably a bit out of what any of us expected here as the
> typical one is to use build llvm on a platform that it already
> supports. I do agree that default target would also work here as a
> suitable test.
> 
> -eric
> 
> On Wed, Mar 20, 2019 at 3:47 PM Hubert Tong via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
>> 
>> The DWARF debug info unit tests in llvm/unittests/DebugInfo/DWARF/
forms a generator based on the host triple. This is problematic for
cross-compile configurations where LLVM does not fully support code generation
for the host platform.
>> 
>> I have found no rationale for why the host triple was chosen (ref:
https://reviews.llvm.org/D27326?id=80002#inline-236026).
>> 
>> I would imagine that it is more useful to test for the default target
rather than the host.
>> 
>> -- HT
>> _______________________________________________
>> LLVM Developers mailing list
>> llvm-dev at lists.llvm.org
>> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

On Wed, Mar 20, 2019 at 7:29 PM Eric Christopher <echristo at gmail.com>
wrote:
> Adding in aprantl here to make sure he's aware.
>
> Making sure I understand - you're building a compiler on machine a
> that llvm doesn't support code generation to and running the llvm
> tests on that platform?
>
Yes, or rather the support is in-progress, but the host platform should be
good for hosting a cross-compiling LLVM toolchain.

>
> That's probably a bit out of what any of us expected here as the
> typical one is to use build llvm on a platform that it already
> supports. I do agree that default target would also work here as a
> suitable test.
>
> -eric
>
> On Wed, Mar 20, 2019 at 3:47 PM Hubert Tong via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
> >
> > The DWARF debug info unit tests in llvm/unittests/DebugInfo/DWARF/
forms
> a generator based on the host triple. This is problematic for cross-compile
> configurations where LLVM does not fully support code generation for the
> host platform.
> >
> > I have found no rationale for why the host triple was chosen (ref:
> https://reviews.llvm.org/D27326?id=80002#inline-236026).
> >
> > I would imagine that it is more useful to test for the default target
> rather than the host.
> >
> > -- HT
> > _______________________________________________
> > LLVM Developers mailing list
> > llvm-dev at lists.llvm.org
> > https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190321/3fe30da4/attachment.html>