Move from display-buffer-overriding-action to display-buffer-base-action

[bmag]

Currently Purpose uses display-buffer-overriding-action to plug into display-buffer, but that makes it harder to use display-buffer-alist for customizing display rules. We might get better results by using display-buffer-base-action.

Pros:

• display rules are customizable by display-buffer-alist. We can get rid of purpose-special-action-sequences and purpose-action-function-ignore-buffer-names.
• simplified purpose--action-function, as it won't need to re-implement parts of display-buffer.

Cons:

• integration with other packages needs to be re-evaluated (especially Popwin can be problematic).
• need to find reliable way to run purpose-display-buffer-functions after display-buffer displays a buffer. To be fair, current situation isn't ideal either.
• drop on users a major change of display configuration
• switch-to-buffer advice falls back to correct switch-to-buffer behavior when purpose-mode is disabled?

[bmag]

Another pro: the docstring of display-buffer-overriding-action was changed in 26.2 to say specifically that it should not be changed permanently by elisp programs. Meaning, Purpose is doing something that contradicts the intended usage of display-buffer-overriding-action. Looks like some packages and modes have started utilizing this variable as intended by Emacs: #167, #168, #169.

[wyuenho]

I just had a first pass of implementing the ideas in this ticket. As a minimal change, simply swapping display-buffer-overriding-action with display-buffer-base-action is enough to obviate the need for #167 #168 and #169. I haven't run into any issues thus far. Perhaps we should break this ticket down, and try doing just that.

display rules are customizable by display-buffer-alist. We can get rid of purpose-special-action-sequences and purpose-action-function-ignore-buffer-names. simplified purpose--action-function, as it won't need to re-implement parts of display-buffer.

Yes, but clean up can be done separately.

need to find reliable way to run purpose-display-buffer-functions after display-buffer displays a buffer. To be fair, current situation isn't ideal either.

Good old advice-add to display-buffer to run the hooks after invocation? I'm not sure what the problems are currently. This needs some explaination. purpose-display-buffer-functions is left as nil by default and I've never found a need to use it for the many years I've been using this package.

drop on users a major change of display configuration

All minor details as a consequence of the clean up effort. It can be done separately and I'm sure there are ways to minimize the impact.

integration with other packages needs to be re-evaluated (especially Popwin can be problematic).

TBH, we should just remove this integration. The popwin extension works perfectly fine. Anything else that doesn't work with it can easily be done by modifying purpose-x-popwin-buffer-names and friends.

switch-to-buffer advice falls back to correct switch-to-buffer behavior when purpose-mode is disabled?

While I'm not sure what this is referring to, I'm sure switch-to-buffer-obey-display-actions introduced in 27.1 will actually have an effect instead of ignored as now.

[bmag]

I actually started implementing some of it in base-action branch, that is based on develop branch. The develop branch was meant to become version 2.0 of Purpose, but that work has stalled about 4-5 years ago and it didn't keep up with changes from master. I'll break the changes from develop and base-action into separate detailed issues/PRs against master soon and create a "project" to track them.

There will also be a PR for moving to display-buffer-base-action, and I'll also break down this current ticket, then it will be easier to work on each part separately.

I think we can get rid of purpose-display-buffer-functions, yeah, I'd like to deviate a bit less from stock Emacs behavior.

TBH, we should just remove this integration. The popwin extension works perfectly fine.

IIRC the integration prevents popwin from deleting the purpose-dedicate window parameter. It can hit users that have popwin as a dependency of some other package, and not under their direct control.

switch-to-buffer-obey-display-actions

I'd like to rely on it and remove the advice, but I'd also like to support Emacs 26. The issue with advising switch-to-buffer and also moving away from display-buffer-overriding-action, is that the buffer might be displayed by some function in display-buffer-alist which most likely won't display the buffer in the currently selected window. For now this is something to keep in mind and test once there is a PR.