bug#30204: 27.0.50; Changing major mode resets non-default invisibility spec

[Philipp <p.stephani2@gmail.com>]

> Am 13.08.2021 um 14:50 schrieb Lars Ingebrigtsen <larsi@gnus.org>:
> 
> Philipp Stephani <p.stephani2@gmail.com> writes:
> 
>> C-x b *foo* RET
>> M-: (progn (add-to-invisibility-spec 'bar) (insert (propertize "foo" 'invisible 'foo)))
>> 
>> Now "foo" is visible, as it should be.
>> 
>> M-x fundamental-mode => "foo" becomes invisible, because
>> `kill-all-local-variables' has reset `buffer-invisibility-spec' to t.
>> This can be confusing for modes that use conditional invisibility
>> properties such as markdown-mode.
> 
> I'm not quite sure I understand the problem here -- if you have a buffer
> in markdown-mode (that's set these properties), why would you change to
> fundamental-mode?

Why not?  Changing the major mode of a buffer is a supported operation.  For example, maybe I decide mid-way while writing a Markdown document that I'd rather want a plain text file without special syntax.

> 
> And I think this is correct behaviour -- if you've switched to
> fundamental-mode, then you don't want these remnants from markdown-mode
> to survive, I think?

Yes, but they do survive, i.e. once markdown-mode has made some text conditionally-visible (which it does for URLs etc.), I'd expect it to become unconditionally-visible instead of unconditionally-invisible when changing the major mode.

> 
>> Why can't you make buffer-invisibility-spec be a permanent-local
>> variable in the cases where that is TRT?
>> 
>> Hmm, yes, that sounds like a good idea. Let me try that. 
>> 
>> I don't think this works: The permanent-local property is a global symbol
>> property, but I'd need it to be per-buffer. 
> 
> Yes, I don't think we have any mechanism that says "this buffer-local
> value should survive if you change major mode, but only this one".  And
> I think that's probably a good idea, because it sounds like a confusing
> interface.
> 
> 

Yes, and probably it's difficult to define a reasonably generic API for this.
Maybe we should aim for a more specific solution: For example, Emacs could make all conditionally-visible text unconditionally-visible when changing major modes?  Or have a way for major modes to declare which invisibility specifiers are mode-specific and should be undone when leaving the mode?