On Mon, 2025-09-01 at 14:50 +0200, Tomas Glozar wrote: > pá 29. 8. 2025 v 23:35 odesílatel Crystal Wood <[email protected]> napsal: > > > > I'm not seeing the "tests/hwnoise.t ... 2/6" part. What is printing > > that? > > > > Test::Harness (which is one implementation of what is called a "TAP > harness") is printing that >
It would be nice to not depend on an out-of-tree test harness (does anything else in the kernel use this?), especially without a prominent comment mentioning it. I've just been doing "sudo sh tests/whatever.t". > (the "prove" command which runs the tests), > it runs the tests, consumes the output and formats it. Ugh, "prove" looks like it's some sort of formal rv thing... misleading name for just running a test suite. > > > > Furthermore, it's standard to first print "not ok" and then the > > > comments, see the documentation to Test::More [1], so I think we > > > should keep that. > > > > > > [1] https://metacpan.org/pod/Test::More#ok > > > > https://xkcd.com/927/ :-P > > > > I'm not a Perler, so I didn't recognize it as anything standardized. > > Still seems backwards to me, both in terms of making it easier to see > > which test failed, and in terms of being a pain to implement. And we > > don't even get the benefit of seeing the test name printed before it > > runs, to make it easier to see what's taking a long time. > > > > We can certainly modify the test system so that it works better for > us. The TAP standard that is used by the harness to process the output > of the RTLA test engine [1] does not specify where the additional > output should go, it's just a convention. I just say this might be > confusing to people who are used to the "not ok, then error" format, > which is associated with this type of test. I'm not suggesting that we break the convention; I just didn't realize it was a widespread format. > > I'll have a look whether there is a way to easily print the test names > besides the number. > > [1] https://testanything.org/tap-specification.html It already does print the test name. If it didn't I probably would have made more substantial changes to the test harness :-) > > > Seems like Linux uses a variant of this, though: > > Documentation/dev-tools/ktap.rst > > > > Yes, this is used by kselftests [2], which are written in C. > > [2] https://docs.kernel.org/dev-tools/kselftest.html Would it make sense for us to use that? Though the current docs linking to a wiki that's labelled "obsolete content" isn't encouraging... > But bpftrace is licensed under Apache 2, so it cannot be in the GPLv2 > kernel source easily. Sigh... -Crystal
