[llvm] r175961 - Make some fixes for LiveInterval repair with debug info. Debug value

[David Blaikie]

On Sat, Feb 23, 2013 at 2:09 PM, Cameron Zwarich <zwarich at apple.com> wrote:

> On Feb 23, 2013, at 12:26 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
> On Sat, Feb 23, 2013 at 12:23 PM, Cameron Zwarich <zwarich at apple.com>wrote:
>
> Running it on test-suite with -g caused assertions to fail, so I assumed
>> that was good enough.
>>
>
> Generally we prefer to have regression tests checked in with code changes
> - the test suite is used to catch things we failed to test but we take the
> opportunity to fix/improve the regression test coverage when the test-suite
> uncovers something.
>
>
> That is generally the case for LLVM, but for experimental codegen changes
> in the past we have generally used the test-suite without adding a LIT test
> for every issue.
>

That's not been my experience - I realize some experimental out-of-tree
code sometimes requires in tree changes that aren't testable without a unit
test (custom code to call into the APIs in a way that would emulate the
out-of-tree code), but for anything that can be tested with the existing
tools (even using custom flags, specific passes, etc) I usually expect &
have seen test cases.


> Maybe that should change (and it would be easier to do if we had codegen
> tests that could just run a single pass), but it's nothing unusual.
>

Certainly the LLVM code generation layer has some issues in terms of
fine-grained testability, but it's still generally tested so far as I know.

- David
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130223/a7a4362c/attachment.html>