[pytest-dev] sorting out broken config initialization - by doing major releases with api breakages

[RonnyPfannschmidt]

Am 21.08.2018 um 16:03 schrieb Bruno Oliveira:
> 
> 
> On Tue, Aug 21, 2018 at 9:14 AM RonnyPfannschmidt
> <opensource at ronnypfannschmidt.de
> <mailto:opensource at ronnypfannschmidt.de>> wrote:
> 
>     Hi Bruno,
> 
>     i created https://github.com/pytest-dev/pytest/projects/2 as a starting
>     point a while back,
>     but its still far from the botton of things,
> 
>     My current impression is, that a detailed analysis will take a major
>     block of time.
> 
>     My intentionally inflated estimate is about 1 work week,
>     since everything about config initialization gives me the impression its
>     intertwined.
> 
>     which is why i'd prefer a a iterative approach that's allowed to break
>     things. (i simply cant afford even a 2 day block to get into things, and
>     i certainly don't want to spread it over time as it adds quite some
>     extra overhead for getting back into things)
> 
> 
> I agree an interactive approach is the way to go. I suspect just by
> investigating why -p doesn't work with xdist works is enough to bring
> some action points to light.

that happens because we have a mismatch in the startup sequences of the
config initialization
on xdist workers we always load all plugins first, while on normal
execution we block the unwanted ones first
> 
> Personally my next goal is the task to replace the internal warnings
> system to use the standard warnings module, which I should be able to
> tackle soon as I have some vacation time coming up, but after that I'm
> happy to help with the configuration refactoring.
> 
> Cheers,
> Bruno.
> 
>