On 2020/07/31 14:54, Sitsofe Wheeler wrote: > On Fri, 31 Jul 2020 at 00:09, Dmitry Fomichev <Dmitry.Fomichev@xxxxxxx> wrote: >> >> Right now, there is no default for libzbc ioengine support. If libzbc is found by >> configure, then libzbc ioengine will always be added. The new flag, --disable-libzbc, >> prevents adding support for libzbc ioengine in this case. If libzbc is not present, >> fio configure is not failed with "feature not found". >> >> Currently, no other feature in fio configure has the default option of "yes" and all >> features that can be explicitly set to yes are set so by script command line options. >> If a feature is set to yes and is not available, configure fails with "feature not found" >> error. Do we really want configure to fail on any system that doesn't have libzbc >> unless --disable-libzbc is added to the command line? >> >> There is one feature option, "disable_lex", with the default value, "". I guess we >> could define libzbc option the same way to improve code clarity. > > That's right: > - "yes": hard fail if probe fails > - "": probe but don't hard fail > - "no": don't probe and disable feature > > So your suggestion of a default of "" is right because I can't believe > we want to hard fail when libzbc is unavailable. > > I learned of this pattern while reading > https://github.com/qemu/qemu/blob/1448629751871c4924c234c2faaa968fc26890e1/configure#L348 > which I think was the basis of fio's configure script. The desire to > show this pattern led to some of the refactoring in > https://github.com/axboe/fio/commit/d490af7f8c0ee7a992f332bb6bcffb30313d6488 > even though it added some unncessary lines for those options... > Thanks. That indeed clarifies everything ! -- Damien Le Moal Western Digital Research
