[LLVMdev] teaching FileCheck to handle variations in order
kparzysz at codeaurora.org
Mon Sep 10 09:20:18 PDT 2012
On 9/7/2012 4:08 PM, Chandler Carruth wrote:
> Fundamentally, I agree with you that these ordering dependencies are
> unfortunate in tests. However, I would make a few observations:
I disagree. A testcase should test what it was meant to test, and
nothing else. Instruction ordering, for example, is not something that
we can expect to remain the same. Various optimizations can affect it
(changes in the scheduler being the most obvious case), causing spurious
failures that would need to be investigated and addressed.
As a matter of fact, we should make an effort to make our tests
independent of any unrelated (but valid) optimizations that may be added
at some point in the future. I realize that this is not always possible
to achieve completely, but the more precise we get with the tests, the
better off we will be in the long run.
I don't think that FileCheck is modified as frequently as the compiler
itself, so the added complexity (which apparently is not large in this
case) is not likely to add a significant recurring cost.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
More information about the llvm-dev