bug#74905: [PATCH] Implement search for nnvirtual Gnus groups

["Björn Bidar via Bug reports for GNU Emacs, the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>]

James Thomas <jimjoe@gmx.net> writes:

> Björn Bidar wrote:
>
>> James Thomas <jimjoe@gmx.net> writes:
>>
>>> Björn Bidar wrote:
>>>
>>>> James Thomas <jimjoe@gmx.net> writes:
>>>>
>>>>> James Thomas wrote:
>>>>>
>>>>>> Björn Bidar wrote:
>>>>>>
>>>>>>> James Thomas <jimjoe@gmx.net> writes:
>>>>>>>
>>>>>>>> Björn Bidar wrote:
>>>>>>>>
>>>>>>>>> This patch implements search for nnvirtual. I'm using
>>>>>>>>> publi-inbox's
>>>>>>>>> with nnvirtual to group each group into one.
>>>>>>>>> However searching wasn't possible in these nnvirtual groups.
>>>>>>>>> I implemented gnus-search-run-search based on the existing
>>>>>>>>> nnselect gnus-search-run-search function.
>>>>>>>>
>>>>>>>> Thanks! I haven't looked into it, but here are some quick comments:
>>>>>>>>
>>>>>>>>> I'm looking for feedback on the patch. I don't exactly know how
>>>>>>>>> the search function is called when multiple groups of the same
>>>>>>>>> type
>>>>>>>>> are
>>>>>>>>> involved. For nnvirtual each group is its on server, does that
>>>>>>>>> mean
>>>>>>>>> the
>>>>>>>>> function will be always called only for each group? In that case
>>>>>>>>> everything should be good.
>>>>>>>>
>>>>>>>> That seems to be the case: see
>>>>>>>> gnus-group-read-ephemeral-search-group
>>>>>>>> and gnus-group-make-search-group.
>>>>>>>
>>>>>>> OK good than my understanding from my tests matched with the rest of
>>>>>>> the
>>>>>>> code.
>>>>>>> Thanks for these examples I haven't looked at the create group
>>>>>>> functions
>>>>>>> as the searched methods don't have to create groups even when they
>>>>>>> start
>>>>>>> a new search by another backend just like e.g. if the user would
>>>>>>> call
>>>>>>> a
>>>>>>> search on another imap group.
>>>>>>>
>>>>>>>>> +(deffoo nnvirtual-request-list (&optional server)
>>>>>>>>> +  (when (nnvirtual-possibly-change-server server)
>>>>>>>>> +    (with-current-buffer nntp-server-buffer
>>>>>>>>> +      (erase-buffer)
>>>>>>>>> +      (dolist (group nnvirtual-component-groups)
>>>>>>>>> +        (insert (format "%S 0  1 y\n" group))))
>>>>>>>>> +    t))
>>>>>>>>
>>>>>>>> Did you check if gnus-start.el#L1801 withstands this? It seems to
>>>>>>>> me
>>>>>>>> to
>>>>>>>> assume that nnvirtual doesn't have -request-list.
>>>>>>>
>>>>>>> It does. If the user has falsely add nnvirtual to one of the select
>>>>>>> methods than it will call it try to call the function which doesn't
>>>>>>> fail
>>>>>>> or do anything. The only thing that happens from that is it will
>>>>>>> show
>>>>>>> the false results as groups contained in the nnvirtual method
>>>>>>> without
>>>>>>> a
>>>>>>> parameter.
>>>>>>
>>>>>> No, I mean, the point of that code seems to be that nnvirtual is
>>>>>> activated _last_, i.e. after any component groups of other backends.
>>>>>
>>>>> No, sorry, I was a little confused about this. But it seems to me that
>>>>> -request-group removes each group from the component list (unlike your
>>>>> -request-list). Is that relevant? IDK. The latter seems to do info
>>>>> updation as well, based on -always-rescan.
>>>>>
>>>>
>>>> Request list supposed to list the available groups on the server it is
>>>> also for example used in gnus-search-imap class gnus-search-run-search
>>>> function. Since for the case of nnvirtual all groups on the server
>>>> are all groups contained in the group it is the right function.
>>>
>>> True, but since it returns the group data like -request-group and often
>>> acts as a _substitute_, shouldn't it mimic its operation? If not, is
>>> there a good reason?
>>
>> It does mimic it's operation or doesn't it?
>
> I mean, I was asking if it _should_ or not. I think it would be a good
> idea to get some 'grep'ed evidence of code coverage that -request-list
> would not supplant the intentions of -request-group.

To verify my earlier thoughts and have an example I tested this.

-request-group functions are for retrieving the metedata of the group
and possibly update the existing metata against the new one.

-request-list function list all groups on the server with possibly
the status of group e.g. killed, zombie or none if subscribed, the
highest amount of articles and the group name.

The latter is the one we need.

>>>> I took all the explenation for each functions purpose from the manual
>>>> section linked below:
>>>> (info "(gnus) Required Back End Functions")
>>>
>>> Please do look at how they're called, as well; for a fuller picture.
>>>
>>>> To me the code just reads like that that for nnvirtual
>>>> the groups are activated this way only because
>>>> it doesn't have request list.
>>>>
>>>>>
>>>>> No, I mean, the point of that code seems to be that nnvirtual is
>>>>> activated _last_, i.e. after any component groups of other backends. And
>>>>> it's not just select methods: even Foreign Groups are included, no?
>>>>>
>>>>>> Should verify that nnvirtual has arguments? So far it is possible
>>>>>> to add nnvirtual to select methods with "" but this is invalid.
>>>>>
>>>>> I'm not sure what you mean, but it does have arguments: the components
>>>>> regexp, for one.
>>>>>
>>>>
>>>> It does have arguments exactly but you can add nnvirtual without
>>>> arguments or technically an empty argument to select methods.
>>>>
>>>>>> If nnvirtual isn't added to select methods nothing happens besides
>>>>>> the regular activation.
>>>>>
>>>>> Couldn't someone have added one? Say, with the above argument?
>>>>
>>>> I guess so but that isn't the intended way of using nnvirtual,
>>>> also it wouldn't take care of the request of the setup such
>>>> as creating the associated group.
>>>
>>> Well now you've made me look at it... :-)
>>>
>>> Try this:
>>>
>>> - In *Groups*: j nnvirtual:Test RET
>>> - S l 3 RET
>>> - e "nnvirtual:Test" C-c C-c
>>> - q
>>> - Add to -secondary-select-methods:
>>>
>>>   (nnvirtual "Test"
>>> 		   (nnvirtual-component-regexp
>>>                    ;; Put in some group you have.
>>> 		   "^$\\|\\(^nnml\\+archive:sent$\\)"))
>>>
>>> M-x gnus
>>>
>>>> Regular activation meant in this context that the request-list function
>>>> is called but without a argument the nnvirtual-server doesn't contain
>>>> any groups.
>>>
>>> OK. But I don't think I've come across that term.
>>>
>>> Also, your -request-list makes the component groups subscribable from
>>> the Server buffer. That seems wrong too: the backend ought to have only
>>> one (umbrella) group, no?
>
>> It only has one group containing other groups yes. The backend is not
>> supposed to end up in the Server but is only there if it falsely was
>> added to the select-methods.
>>
>> Anyway I need a backtrace from the case where the search did not work
>> with nnml groups.
>
> No no, you misunderstand me: It was just to demonstrate that:
>
>> that isn't the intended way of using nnvirtual
>
> ...can't be said. So please re-verify anything (there might've been)
> about the patch predicated on that. I did not encounter errors.
>
> I know you're exploiting the fact that the server and group are one, but
> perhaps it's being too clever by half. Any such unintended effects of
> that easy road ought to be acceptable to other users.

Yeah that's why I posted the patch here so more can try it with
different setups.

A review from a Gnus maintainer or contributor would also be awesome.