On 5/26/2026 1:14 PM, Daniel Henrique Barboza wrote: > Hi, > > I happen to use 'ninja -C build test to do a (usually) quick sanity test > on patches and changes and whatnot. An usual run of this target tests > around 327 individual tests with the RISC-V stuff I usually build. > > After pulling today's master 'ninja -C build test' is now running a > whooping > 1366 tests. Here's an example: > > > $ ninja -C build -j 9 test > ninja: Entering directory `build' > mkvenv: checking for pygdbmi>=0.11.0.0 > mkvenv: installing pygdbmi==0.11.0.0 > [2/3] Running all tests > [1-12/1366] (....) > (....) > [250-260/1366] 🌕 block-qcow2-optional+block-slow - qemu:io-qcow2-014 > [250-261/1366] 🌖 block-qcow2-optional+block-slow - qemu:io-qcow2-014 > [252-262/1366] 🌒 block-qcow2-optional+block-slow - qemu:io-qcow2-026 > (...) > > I noticed a lot of these "optional+block-slow" tests being ran now, > something that > wasn't happening before. > > If I use 'make check' I have the results I was having with 'ninja -C > build test' > before the recent changes. > > I tried to bisect and got to this commit: > > commit daf0d008822be3cb14cdb533f7237c3c789bc373 (HEAD) > Author: Daniel P. Berrangé <[email protected]> > Date: Mon Jan 12 13:01:49 2026 +0000 > > iotests: remove redundant meson suite for iotests > > > From what I can tell this commit triggered (unintended?) stuff that was > already in the > code. In other words I am not sure if this commit is actually the > culprit, so I am > reluctant to even call this a bug. >
Recently Daniel sent a series to enable more iotests, which is great! https://lore.kernel.org/qemu-devel/[email protected]/ It seems like a side effect is that they are default now when using ninja -C test. They should be correctly filtered if you use make check-*. > My question is: should I be using 'ninja test'? It seems to me that the > common practice > is to use 'make check' and maybe I've been using an "unsupported" > testing scheme. > Officially, it's not how we're supposed to run things. I asked the same question a few months ago and was invited to not rely on meson testing, and thus ninja test by extension. That's sad IMHO, meson interface is much better than make interface. Looking at build/build.ninja, (grep "build test:"), test target does: ./build/pyvenv/bin/meson test --no-rebuild --no-stdsplit --print-errorlogs I prefer to use meson test, since you can easily list and run specific tests. ./build/pyvenv/bin/meson test --print-errorlogs \ --setup thorough --list I personally always run (well, I don't, GitHub does for me) tests with thorough setup, so all tests are ran anyway. > > Cheers, > Daniel > > > Regards, Pierrick
