Uh, I had technical issues and did not get all mails as I expected.
Cobbling things together in one big reply now, with references and
quotes hopelessly broken ... hope you can sort it out.
Anyway, thanks to Eric for chiming in.
> Ideally, it would be nice to have tests, though I have no clue how to
> approach writing them.
I have created a somewhat minimal Gnus setup to develop and test this
patch on my development laptop, where I normally do not use Gnus. It
consists of a bunch of files and directories and a bit of configuration.
I can follow up on this if you like, but preferably in a separate
thread.
>> If we're currently in article-mode. The call to
>> `gnus-article-show-summary' would protect against the case where the
>> summary buffer has been killed in the meantime [...].
Not really. The following executed in an article buffer:
(progn
(kill-buffer gnus-summary-buffer)
(gnus-article-show-summary))
results in
Debugger entered--Lisp error:
(error "There is no summary buffer for this article buffer")
signal(error ("There is no summary buffer for this article buffer"))
error("There is no summary buffer for this article buffer")
gnus-article-show-summary()
[...]
Which, OTOH, shows that I was wrong in one aspect: Gnus at least in some
cases *does* give a reasonable error message when the summary buffer for
an article buffer is gone.
>> Probably it would be enough to wrap the whole containing `let*' in a
>> (with-current-buffer gnus-summary-buffer ...). If we're already in the
>> summary buffer, no harm done.
>
> I am not sure if it is safe.
> There is
> (save-window-excursion (gnus-summary-select-article))
> which calls (set-buffer gnus-summary-buffer)
I agree with Ihor here and would rather go for individual wraps into
`with-current-buffer'. As I have done in my patch already,
incidentially.
>> Ugh, this whole thing is a mess. I think the first question is: should
>> this function "fix" the state of Gnus before it makes a link? Should it
>> attempt to re-open the Summary buffer if it's been closed? Should it
>> switch current articles if the open article buffer is not the one that
>> point is on in the Summary buffer?
>>
>> If we make a decision about that, then it should be easier to decide how
>> to handle the code changes themselves.
>
> ol-gnus should store link for thing at point in current buffer. Ideally,
> without side effects. Everything else should be implementation detail.
Could we agree on: ol-gnus should store Gnus links with as little effort
and side-effect as possible while providing reasonable functionality for
the common use cases. I think, again incidentially, that my patch
matches this criterion.
What optionally could be improved, though, is error handling in these
pathological cases. But that would probably require some macro
(ol-gnus-with-current-summary-buffer BODY)
to have the error handling available in the separate places. Not sure
whether this is worth the effort.
> Or, at least, it was not sufficient at the time when ol-gnus has been
> written (quite a while ago).
I don't think this has changed, really.
