spacemacs-layouts-restrict-spc-tab default value (nil) not respected sometimes due to advice

[bcc32]

Some users at my site reported a recent regression to spacemacs/alternate-buffer (SPC TAB) which seems to be due to the refactor in #17161. With spacemacs-layouts enabled and its layer variables all set to the defaults, you can reproduce with the following steps:

1. emacs
2. SPC f f foo to create a new buffer for a file that doesn't exist yet, and save it
3. SPC l 2 to create a new layout, give it an arbitrary name
4. SPC f f foo, visiting the same buffer as before but from the new layout
5. SPC f f bar to visit another file, creating a new buffer in the new layout
6. SPC TAB goes to *spacemacs* instead of foo's buffer

If I understand the bug correctly, it is because spacemacs/alternate-buffer now relies on previous-buffer, which calls switch-to-prev-buffer, and the latter is advised by default to cause buffer-list to omit the foo buffer. This violates the expected behavior of spacemacs-layouts-restrict-spc-tab = nil.

[bcc32]

This could be fixed by temporarily disabling the advice during spacemacs/alternate-buffer (or just let-binding some variable that inhibits the advice), but I'd be open to other fixes as well.

[fnussbaum]

Thanks for reporting this. There are indeed more cases where the behaviour differs than I expected, perhaps we should partially revert the refactoring. I think this particular case is due to switch-to-prev-buffer respecting the frame's buffer-predicate parameter (which persp-mode and Spacemacs both set by default). This also means that setting the option spacemacs-useful-buffers-restrict-spc-tab to nil does not have an effect anymore.

(IIRC the advice on switch-to-prev-buffer should usually only matter in cases where the buffer switched to was never displayed in the selected window before, because switch-to-prev-buffer only consults the buffer-list as a fallback.)