Hi Bruno, Your example is a good argument, I haven't considered third party fixtures over which the developer of the test suite has less control. I agree that requiring users to jumps through extra hoops to give third-party fixtures the desired order is too high of a price to pay for the nudge towards good style (explicitly declaring dependencies) that we get from undefined order.
Once we decide to give some guarantees about the order of fixture execution your new approach certainly makes sense. Cheers, Vasily On Fri, Mar 16, 2018 at 12:34 PM Bruno Oliveira <nicodde...@gmail.com> wrote: > Hi Vasily! > > On Fri, Mar 16, 2018 at 7:17 AM Vasily Kuznetsov <kvas...@gmail.com> > wrote: > >> I very much agree with Floris that if you need fixture A to run before >> fixture B and otherwise things break, this is called "dependency" and it's >> better if it's explicitly declared. >> > > Definitely, if a fixture requires something that is done by another > fixture, then that dependency should be explicitly defined; but the issue > is more that people expect higher level scoped fixtures to be executed > first, and when you mix autouse fixtures and usefixtures markers, the order > is non-intuitive. > > > >> I can't easily imagine a situation where declaring dependencies would be >> too much work or not desirable for some other reason but maybe it's just my >> imagination not being good enough :). >> > > I wrote an example which demonstrates this in a separate reply to Floris. > > Outer scopes running before inner scopes does sound kind of logical but >> everything running lazily on demand (as it does now) also makes sense. >> > > Just to be clear, in my PR fixtures are still created lazily, it is just > that we sort them by scope (preserving order) first. > > Cheers, > Bruno. >
_______________________________________________ pytest-dev mailing list pytest-dev@python.org https://mail.python.org/mailman/listinfo/pytest-dev
