On 09/09/2025 08:16, Ojaswin Mujoo wrote:
This requires the user to know the version which corresponds to the feature.
Is that how things are done for other such utilities and their versions vs
features?
I was going to suggest exporting something like
_require_fio_atomic_writes(), and _require_fio_atomic_writes() calls
_require_fio_version() to check the version.
(Sorry, I made a half reply in my last email)
This looks better than only using _require_fio_version. But the nature is still
checking fio version. If we don't have a better idea to check if fio really
support atomic writes, the _require_fio_version is still needed.
Or we rename it to "__require_fio_version" (one more "_"), to mark it's
not recommended using directly. But that looks a bit like a trick 😂
Thanks,
Zorro
Hey Zorro, I agree with your points that version might not be the best
indicator esp for downstream software, but at this point I'm unsure
what's the workaround.
One thing that comes to mind is to let fio do the atomic write and use
the tracepoints to confirm if RWF_ATOMIC was passed, but that adds a lot
of dependency on tracing framework being present (im unsure if something
like this is used somewhere in xfstests before). Further it's messy to
figure out that out of all the IO fio command will do, which one to
check for RWF_ATOMIC.
It can be done I suppose but is this sort of complexity something we
want to add is the question. Or do we just go ahead with the version
check.
I think that just checking the version is fine for this specific
feature. But I still also think that versioning should be hidden from
the end user, i.e. we should provide a helper like
_require_fio_atomic_writes
thanks,
John
