On 10/4/26 11:03 PM, Naveen SiddeGowda wrote:
Hi Jeff,
>> So I know this has already been integrated, but how was this tested?
Thank you for pointing this out and I sincerely apologize for the
regressions caused by the patch.
Before posting it, I ran the complete GCC and G++ testsuite using make
check-gcc check-g++.
However, the unpatched trunk already produced approximately 39,015
unexpected RISC-V failures in
my environment.
That's off the charts bad. How are you configuring, building and
testing? You could take an order of magnitude off that set of
failures and I'd still consider it way too high.
Because of these baseline failures—likely related to problems with my
QEMU setup, I was
unable to reliably identify any additional regressions introduced by
the patch.
There's scripts which can help. Look in contrib/
I will first fix the testing environment, establish a clean and
reproducible baseline and then rerun
the tests to properly assess the patch’s impact. I agree that
code-quality regressions must be detected
before submission and will ensure that RISC-V testing is reliable for
future patches.
Happy to help with figuring out what's going wrong for you. Is this in
an MDCP VM?
jeff
