I posted a patch with a stupid mistake last night. Running the right
tests would've saved me the embarrassment. Trouble is to run them you
first have to know they exist, and then you have to know the right test
script incantation. "make check-block" can't do that for you.
I believe we can reasonably expect people patching block stuff to know
about "make check-block" in addition to plain "make check", but asking
them to know everything tests/qemu-iotests/check can do is a bit much.
Here's what it can test, as far as I can tell, judging from
tests/qemu-iotests/{common,group}:
* Image formats raw, bochs, cow, cloop, parallels, qcow, qcow2, qed,
vdi, vmdk, vpc, vhdx
* Protocols file, rbd, sheepdog, nbd, ssh, nfs
* Test groups aio, auto, backing, blkdbg, img, io, quick, rw, snapshot
* Arbitrary cache modes (usual caveats re cache=none on funny file
systems apply)
* A few miscellaneous test options
Current "make check-block" runs tests/qemu-iotests-quick.sh, which tests
format qcow2, protocol file, group quick with cache=none. Runs 55 tests
and takes about 2 minutes for me on a not particularly fast test box
(spinning rust).
Should "make check-block" run more tests? Which ones?
Should we have a variant of "make check-block" for testing other
(format, protocol) combinations?
Should we have a variant of "make check-block" that tests everything
that's testable without special setup?
How can we make advanced iotest features more discoverable for
developers not already intimately familiar with it?
