Re: gnus shouldn't be making general-purpose variables buffer-local

[David Engster <deng@randomsample.de>]

Katsumi Yamaoka <yamaoka@jpl.org> writes:
>>>>>> David Engster wrote:
>> So maybe, we should change `gnus-summary-set-local-parameters' to only
>> set those variables buffer-local that begin with a proper prefix, like
>> "gnus-" and  "mm-" ?
>
> But it will disable the existing features concerning the parameters
> of which the names don't begin with "gnus-" and  "mm-". 

That shouldn't happen, since those parameters are usually retrieved
through `gnus-group-get-parameter', `gnus-group-fast-parameter', or
`gnus-group-find-parameter' (see e.g. how nnimap uses the uidvalidity
parameter). If an existing feature breaks because of that change, I
would argue that it is broken, since it shouldn't depend on group
parameters being buffer-local variables.

In short: Making a group parameter buffer-local should only be done for
overriding a global value.

That gives me an idea: in `gnus-summary-set-local-parameters', instead
of defining fixed prefixes like "gnus-" or "mm-", we could simply make
only those parameters buffer-local which are globally bound?

> The best way seems to be to add "gnus-" or "mm-" to all the
> parameters.  However, it will take time for coding and verifying.

The main problem I see is that this is an incompatible change,
i.e. parameters from existing groups would have to be updated.

> Maybe, it should do after the Emacs 23.1 release.

I agree in any case.

Regards,
David