Re: master ab23fa4: Sort by recency in flex completion style when no flexy stuff happening

Re: master ab23fa4: Sort by recency in flex completion style when no flexy stuff happening

[Dmitry Gutov]

On 15.08.2021 15:33, Jo�o T�vora wrote:
>      Sort by recency in flex completion style when no flexy stuff happening
>      
>      Fixes: bug#49888
>      
>      * minibuffer.el (completion--flex-adjust-metadata): Fall back to usual
>      alphanumeric, length, recency strategy if no minibuffer input.  There
>      is still a bug indicated by the nearby FIXMEs, though.

On the balance, it's a good change, but I hope we can agree that it's 
not really the proper place to choose the default sorting method.

It would be great if sorting were implemented in a more composable way.

Re: master ab23fa4: Sort by recency in flex completion style when no flexy stuff happening

[João Távora]

It is composable, in fact it's inside some "compose-flex-sort-fn` helper
function to take into account the table's sorting.

if you mean the problem of code duplication, however, then I also agree.
See the FIXMEs explaining the situation.

If you mean something else, please clarify.

João

Re: master ab23fa4: Sort by recency in flex completion style when no flexy stuff happening

[Dmitry Gutov]

On 17.08.2021 10:20, João Távora wrote:
> It is composable, in fact it's inside some "compose-flex-sort-fn` helper
> function to take into account the table's sorting.
> 
> if you mean the problem of code duplication, however, then I also agree.
> See the FIXMEs explaining the situation.
> 
> If you mean something else, please clarify.

It does "compose" things, but it makes all the decisions, so it's 
impossible to control from the outside.

One of the things I'd like to do is to sort taking into account both 
flex sorting and the previous occurrences of the completions in the 
buffer (by adding the scores together, or somehow else combining them).

I see there are other possible uses described in bug#45260. Perhaps this 
discussion could move there.

But what I imagined is some middle layer above the completion styles 
which would do the sorting. The above discussion is based on the 
existing practice, however, where the tables provide the sorting logic 
(and not, say, sorting scores which can be combined somewhere else).

Re: master ab23fa4: Sort by recency in flex completion style when no flexy stuff happening

[João Távora]

[-- Attachment #1: Type: text/plain, Size: 1468 bytes --]

On Tue, Aug 17, 2021, 13:57 Dmitry Gutov <dgutov@yandex.ru> wrote:

> On 17.08.2021 10:20, João Távora wrote:
> > It is composable, in fact it's inside some "compose-flex-sort-fn` helper
> > function to take into account the table's sorting.
> >
> > if you mean the problem of code duplication, however, then I also agree.
> > See the FIXMEs explaining the situation.
> >
> > If you mean something else, please clarify.
>
> It does "compose" things, but it makes all the decisions, so it's
> impossible to control from the outside.
>

I guess you could add a variable or two there, but the way it works is by
design.

>
> One of the things I'd like to do is to sort taking into account both
> flex sorting and the previous occurrences of the completions in the
> buffer (by adding the scores together, or somehow else combining them).
>

That sounds interesting, maybe it's a good idea for a new completion style,
which may share some code, maybe a lot of code, with flex.

I see there are other possible uses described in bug#45260. Perhaps this
> discussion could move there.
>

Put me in Cc, if you want.

But what I imagined is some middle layer above the completion styles
> which would do the sorting. The above discussion is based on the
> existing practice, however, where the tables provide the sorting logic
> (and not, say, sorting scores which can be combined somewhere else).
>

Sounds interesting.

João

>

[-- Attachment #2: Type: text/html, Size: 2765 bytes --]

Re: master ab23fa4: Sort by recency in flex completion style when no flexy stuff happening

[João Távora]

Dmitry Gutov <dgutov@yandex.ru> writes:

> On 15.08.2021 15:33, Jo�o T�vora wrote:
>>      Sort by recency in flex completion style when no flexy stuff happening
>>           Fixes: bug#49888
>>           * minibuffer.el (completion--flex-adjust-metadata): Fall
>> back to usual
>>      alphanumeric, length, recency strategy if no minibuffer input.  There
>>      is still a bug indicated by the nearby FIXMEs, though.
>
> On the balance, it's a good change, but I hope we can agree that it's
> not really the proper place to choose the default sorting method.

I've now pushed an adjusment for this fix that is functionally
equivalent but duplicates no code of completion-all-sorted-completions
at all.

Thus, the "default" sorting method can be chosen there to be whatever.
The flex style won't specify any display-sort-function/cycle sort
function unless there is "flexy stuff happening".

João