bug#25111:

[Phillip Lord]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: phillip.lord@russet.org.uk (Phillip Lord)
>> Cc: 25111@debbugs.gnu.org
>> Date: Wed, 07 Dec 2016 16:40:02 +0000
>> 
>> I think that the first of these is incorrect. There is no need to bind
>> `inhibit-modification-hooks' to `t'. More over, there may be reasons by
>> bind `inhibit-modification-hooks' to `nil' (i.e. "If you do want
>> modification hooks to be run...").
>
> So if we envision that some hook will bind inhibit-modification-hooks
> to nil, then that is the reason to bind it to t in a hokk which
> doesn't want such hooks to be run.

Indeed, this is true, and it's a difficulty with the hook. I mean, it is
*meant* to run when the buffer is modified and yet here is an example of
where we cannot be sure that it will.


>> It possible that the documentation should say "Mostly, you should avoid
>> modifying the buffer on these hooks, any other functionality using these
>> modification-hooks will not be called."
>
> You mean, not mention the variable at all?  That'd be loss of useful
> information, I think.
>
>> The reason I ask all of this as a result of a concrete use
>> case. yasnippet modifies the buffer in these hooks, in turn breaks my
>> own package, lentic, which uses these hooks to respond to changes.
>
> So how would you want the manual to help avert such calamities?


My own feeling is that "inhibit-modification-hooks" should *only* be for
modifications that really should not be detected by anything else. I can
think of examples of this (I used to change the buffer to display a
completion string to the user for instance, although I now use an
"after-string" overlay property).

The simplest advice makes calls to the modification hooks consistent is
to say "You should not modify the buffer on these hooks". The potential
solution, for instance, for yasnippet is to record the changes on
after-change-function, and then change the buffer on
post-command-hook. I think this would work? Is this what the manual
should say?

Phil