[PATCH] D126907: Deferred Concept Instantiation Implementation Take 2

Erich Keane via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Mon Aug 22 09:26:33 PDT 2022


erichkeane added a comment.

In D126907#3739750 <https://reviews.llvm.org/D126907#3739750>, @Mordante wrote:

> In D126907#3738383 <https://reviews.llvm.org/D126907#3738383>, @erichkeane wrote:
>
>> There was a test I dealt with previously where a ton of the header files were run with modules enabled (and an auto generated files), so I'm shocked to find there is MORE differences here.  @ldionne : This is another example where trying to test libcxx is particularly difficult to do apparently.
>
> I disagree; testing libcxx is easy. Unfortunately what's missing is good documentation explaining how to test different configurations. A similar libc++ related issue came up in another Clang review recently where the libc++ CI setup was unclear. Afterwards I had a talk with @aaron.ballman and I agreed to improve the documentation. While improving the documentation I noticed some issues with our CI setup. So before publishing documentation that doesn't reflect reality, I first wanted to fix these issues. While testing my fixes I ran into the build failure due to this patch. So now we're at a full circle.
>
> With the revert of this patch it's possible to fix the CI issues for libc++ and I will continue with improving the documentation.
>
> Note that this specific issue isn't directly detected in the libc++ CI. This would require a full clang build and testing with modules. That's not done, full builds are only tested without modules. If wanted I supply a patch how to test this configuration in the libc++ CI.
>
> If you need further assistance feel free to reach out to me or other libc++ developers, the easiest way to reach us is #libcxx on Discord.

This is the 3rd time some non-obvious-how-to-run libcxx test failure has caused a revert request to this patch, which is obviously getting frustrating. Additionally, in none of the 3 cases was I alerted automatically.

I appreciate that there are actually ways to RUN these tests on my machine (Aaron is not as lucky, see above), but the lack of a "test all the libcxx things" flag is shocking.  It seems that I'd need at least 2 separate CMake directories to test these configurations?  That seems shocking to me.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D126907/new/

https://reviews.llvm.org/D126907



More information about the cfe-commits mailing list