This allows `niri-visual-tests` to still be built and run in the dev
shell where it's necessary, as well as brings back the nightly `rustfmt`
used by the project
We can't use `fenix` again though as it doesn't wrap `ld` like nixpkgs
and rust-overlay do; without it, the way we link `dlopen()`'d libraries
breaks
Previously, inputs like Crane and Fenix were used to only build the
`niri` package. This isn't really required, and can easily be replaced
by nixpkgs' `rustPlatform` -- which will also lead to less dependencies
being pulled into user's lockfiles
this boils down to adding some extra dependencies to the shell
environment. they're also inherited from craneArgs because the ones from
the package are actually transformed into the WRONG outputs of the
packages. also refactors to use craneLib.devShell because it's somewhat
cleaner.
* Implement wlr-screencopy
* Finish the implementation
Lots of changes, mainly to fix transform handling. Turns out, grim
expects transformed buffers and untransforms them by itself using info
from wl_output. This means that render helpers needed to learn how to
actually render transformed buffers.
Also, it meant that y_invert is no longer needed.
Next, moved the rendering to the Screencopy frame handler. Turns out,
copy() is more or less expected to return immediately, whereas
copy_with_damage() is expected to wait until the next VBlank. At least
that's the intent I parse reading the protocol.
Finally, brought the version from 3 down to 1, because
copy_with_damage() will need bigger changes. Grim still works, others
not really, mainly because they bind v3 unnecessarily, even if they
don't use the damage request.
---------
Co-authored-by: Ivan Molodetskikh <yalterz@gmail.com>
