feat(capture): Finalize on :w and show the destination

[seflue]

Today :w in a capture buffer writes a temporary file that nothing reads. Only the finalize mapping refiles the capture, and it also closes the window.

With this change, :w refiles the capture and opens the destination in the same window, with the cursor where it was. :wq refiles and closes the window. The finalize mapping works as before. For org-roam this means :w creates the new node and leaves you in it, without a prompt since #1196.

refile() takes a new disposition option: 'close' (default, used by the mapping) or 'show' (used by :w).

Three details that are not obvious from the diff:

• The BufWriteCmd is nested. The refile adds the destination with bufadd, and the buffer cache drops its cached miss only on BufAdd, which a non-nested autocmd suppresses. Without it the write fails with an invalid buffer id.
• show() detaches the capture window's autocmds before swapping the buffer. The capture buffer is bufhidden=wipe, and its BufWipeout handler would start the refile again.
• show() runs filetype detect, because bufload skips detection and :edit reuses the loaded buffer.

[kristijanhusak]

How does "kill" behave in this scenario? Even after writing the file, you should be able to kill (cancel) the capture.

[seflue]

That is actually the question. My intuition is: :w creates the file from a capture. That I can lose my work after a write by mistakenly pressing a "cancel" key combination is a real danger to me and I lost some work multiple times by that.

[kristijanhusak]

Isn't org-roam creating a capture buffer with a path to a proper, non-temporary file? nvim-orgmode capture creates a temp file, so when the capture is finished, headline is refiled from the temp file to a destination.

[seflue]

I'm working on a solution that is closer to what Emacs does. The biggest difference here is that Neovim cannot maintain (narrowed) views of the same buffer (where the write actually goes to the buffer).
My current belief state is now: for full-file capture we should write directly into the target buffer (that is what Emacs does) and on kill we should clear the buffer and ask if we actually should remove the file. Then the user can say "no" and undo if this was a mistake - Emacs + Roam works the same. There the confirmation question for deleting the file is implemented in Roam but I don't see a reason why it shouldn't always be the case.

[kristijanhusak]

Sounds good to me, as long as it is only used for org-roam stuff. Plain nvim-orgmode capture should stay as it is, because it uses temp files.

[seflue]

I reworked the branch into two commits. The second one implements what I described above, but only for whole_file captures into a file that doesn't exist yet, so in practice new roam nodes. Everything else still goes through the temp file. One thing changes for hooks: on_pre_refile now runs when such a capture opens, because roam resolves the node path in it and the path is needed to open the buffer.

The first commit only changes Escape on the :q! prompt, for all captures. Escape counted as No there and threw the capture away, now it reopens the capture instead.