Closed btracey closed 9 years ago
Wait. Not sufficient.
PTAL
At least on my machine, OpenBLAS and gonum disagree on the correct behavior with Idamax where there are NaNs present. If you'd like I can log the failure instead of causing a t.Error.
LGTM with one minor optional comment and yes, I'd like a logged failure rather than an error - with a TODO to remove it when the OpenBLAS bug (yes) is fixed. Can we do that without weakening the other Idamax test cases?
The fuzzer I have ( https://github.com/btracey/blasfuzz/blob/master/onevec/idamax.go ) tests differently if the slice has a NaN value or not. My plan was to do a similar test (if the slice has a NaN, log the failure).
That approach sounds fair.
Fixes #135