r210584 - Fix crash with x86_64-pc-win32-macho target. <rdar://problem/17235840>

[Bob Wilson]

> On Jun 11, 2014, at 9:11 AM, Chandler Carruth <chandlerc at google.com> wrote:
> 
> 
> On Wed, Jun 11, 2014 at 5:04 PM, Bob Wilson <bob.wilson at apple.com> wrote:
> There’s a big difference between “I don’t know how to write good tests for this” and “we need to rip it out because no one is maintaining it”.
> 
> Yes, there is. And I think I called it out in my email? If we can't find anyone on the lists who knows how to write a good test (not just you, but *anyone*) then no one is maintaining it.

Sorry, but if you already called it out, then I missed it. Can you explain? It still seems to me like a huge difference.

>  
> We will definitely maintain it, and we (Apple) will have to deal with the pain caused by the lack of good tests.
> 
> This is a shared project, and we *all* have to deal with the pain caused by the lack of good tests. That's why it's so important to have them, especially for targets which not all developers have access to.

Can you please explain why the lack of tests for this is going to cause pain for anyone outside Apple? If we’re really the only ones using it, then only our stuff is going to break, and clearly the burden for fixing it will be entirely on us.

>  
> It basically means that we are stuck waiting to find problems when people use it, rather than catching them earlier when we run the test suite. We can add tests for those issues as they arise.
> 
> I have seen you and others routinely ask people adding support for platforms include sufficient testing that developers not working on those platforms can at least make reasonable changes to the shared infrastructure with confidence that these platforms aren't being grossly broken. I don't understand why this platform is different. I mean, it may be circumstantially different because no one noticed how bad the testing is until now, but I don't understand why there is any response to that other than "yea, we need to get the maintainers of this platform to add some tests pronto so we can keep supporting it reasonably". =/

Good tests should have been added when the support for this platform was added. It was a mistake that we let it go in without those tests.

> 
> The last time I tried to change header search in Clang I had a truly horrific time of it precisely because of the utter lack of testing on this front. I've worked reasonably hard to ensure that the platforms I maintain have good tests now for both header and library search. I think its reasonable to ask someone to do the same for this platform. If you're supporting this platform (because you're fixing bugs in it) it seems reasonable to ask you to either do this or find someone who can do this.

If there are no tests, then I don’t see why you or anyone else should need to be overly concerned about breaking it.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140611/d0652761/attachment.html>