Re: yank-media: allow users to limit image types that can be inserted

[Visuwesh <visuweshm@gmail.com>]

[வியாழன் அக்டோபர் 31, 2024] Eli Zaretskii wrote:

>> From: Visuwesh <visuweshm@gmail.com>
>> Cc: yantar92@posteo.net,  pinmacs@cas.cat,  rpluim@gmail.com,
>>   emacs-devel@gnu.org
>> Date: Tue, 29 Oct 2024 16:59:34 +0530
>> 
>> > My suggestion does not require any need to know those ugly details.
>> > It just suggests to remove from the list the handlers a mode doesn't
>> > want.  Removing the, say, image/png handler from the list does not
>> > require any knowledge how that handler accesses the clipboard nor how
>> > it extracts PNG images from the clipboard.  It just requires to know
>> > the (trivial) fact that an image/png handler can interpret the
>> > clipboard data as a PNG image.
>> >
>> > So I don't think I understand your reasoning.  What did I miss?
>> 
>> That the user does want PNG images is a "soft preference."  If the
>> clipboard only has image/png, the user would have the image/png data
>> instead of none at all.  You offered a solution for this: bind
>> yank-media--registered-handlers in a custom command but having a
>> variable would make it easy for the user to have a _global_ preference
>> across major-modes.
>
> This (and the patch you suggest) completes the full circle of this
> discussion: it started when I said that such a global preference makes
> very little sense to me, based on my experience with yanking different
> data types in other applications.  Users have no reason for such
> preferences, as they almost never have enough reasons to prefer one
> type over the others _globally_.  The preference only makes sense in
> each specific case, and then asking the user to choose is exactly TRT,
> which we already do.

At least for images, I see a value in having a global preference.  You
also seem to agree on this point.

>> >> The major-mode authors would simply write handlers for all relevant
>> >> data types and leave it to the user to choose the preferred type if
>> >> more than one of them is handled by the major-mode.
>> >
>> > AFAIU, we were talking about situations where the major mode "knows
>> > better" than the user, and doesn't want to leave the choice to users.
>> 
>> No, the major-mode does not "know better" than user.
>
> It might, though.  And it already does, in fact: the few modes which
> support yank-media only register some of the handlers, but not others.
> Did you ask yourself why message.el registers only the handler for
> images, but not, say, for text/html?  A major mode which doesn't
> support embedded images will probably refrain from installing image/*
> handlers, etc.

Sure but it doesn't help when the major-mode can support multiple image
formats but the user wants a specific format always.  

>> It simply wants to respect the "soft" preferences of the user.
>
> If we think such global preferences might make sense, we need to
> describe and discuss the use cases and reasons for having such
> preferences.  Until now, the only reason I've heard was the desire to
> save some typing, i.e. let Emacs yank the one "preferred" media type
> without asking.  If that is the reason, we could have instead a
> feature whereby "the best" or "the most appropriate" media type is
> yanked without asking; for example if yank-media is invoked with a
> prefix argument.  (Or maybe the other way around: without the argument
> yank-media would yank "the best" media type.)  This is what many other
> applications do, except that they do it when you type Ctrl-V, whereas
> we decided, and for good reasons, not to change how C-y work (though I
> don't see why some optional feature could not make C-y call
> yank-media).

I agree that it would be good to have yank-media choose the most
appropriate format itself.

>> > I don't understand how.  TARGETS include stuff like image/png and
>> > text/html; how does looking at that defeat any abstractions, and what
>> > abstractions are those?  We cannot consider TARGETS to be an opaque
>> > object anyway because then we won't be able to ask the user which of
>> > the MIME types she wants to yank, nor apply any advance preferences of
>> > the user.
>> 
>> The MIME type the user wants to yank is asked by _yank-media_ currently.
>> The major-mode has no part in this conversation, and this is exactly
>> what we want to change.
>
> Major mode _does_ have a part: it registers only the handlers it wants
> to support.  So it determines, up front, which media types will be
> available, even if more of them are in the clipboard.

I agree but its participation could be improved.  When you copy a cell
from LibreOffice's Excel clone, it puts both image/png and a TSV data
type in your clipboard.  Taking the example of Org which has a handler
registered for image/png: if it registers a handler for the TSV data
type, the user would be asked to choose between image/png and TSV when
she copies a table cell.  The intent of registering image/png is a
hindrance here.  Having a separate command for yanking a table cell
would remove the convenience of having a single command that does TRT.

>> Without the patch applied, copying an image from Firefox and using
>> yank-media in an Org buffer asks me if I want image/png or image/jpeg.
>
> Please tell me why on earth would you prefer PNG to JPEG when yanking,
> or vice versa.  The result is an image displayed in the buffer, which
> is neither PNG nor JPEG.  Why does it matter _for_you_personally_
> which intermediate format will Emacs use as part of the yanking
> process?

You answered the question yourself:

    For example, JPEG is generally inferior because it's lossy, so we
    could have that hypothetical variant of yank-media which doesn't ask
    questions to always prefer PNG to JPEG if both are possible.

>> With that patch applied, and after evaluating 
>> 
>>     (setq-local
>>        yank-media-preferred-type-function
>>        (lambda (types)
>>          (if (memq 'image/png types)
>>              (list 'image/png)
>>            types)))
>> 
>> in an Org buffer, or message-mode buffer, I don't get asked that
>> question.  Afterwards, I copy a JPEG image using xclip then do
>> yank-media. This yanks the image without asking me anything.  This would
>> not be possible had Org or message-mode only registered a handler for
>> image/png.
>
> If we want to avoid the question when several image formats are
> available, we could teach Org, or Emacs in general, which formats to
> prefer.  For example, JPEG is generally inferior because it's lossy,
> so we could have that hypothetical variant of yank-media which doesn't
> ask questions to always prefer PNG to JPEG if both are possible.  This
> is IMO better than asking the users to decide that for Emacs.

I agree, but can we, at least, agree that it should be easy to override
it completely?  What do you think about adding a patch like I proposed
but also having a "sensible" default definition of
yank-media-preferred-type-function? (for some definition of sensible.)

> Anyway, since with this message we've made a full circle, let me
> summarize my opinions on this:
>
>   . I think we should add to Emacs rules for preferring one media
>     format over the others when several are available and supported
>   . Such rules should be customized by major modes based on their
>     features and preferences (e.g., a mode that has no support for
>     faces might prefer plain text to other textual formats)

(1) and (2) would be best done with the help of user feedback.  A patch
like the one I proposed would help to kick-start it.  IMO, coming up
with such rules is a long game that requires user feedback (like the one
that started this whole thread).  Due to the number of users and its
relevance, Org is a good ground for obtaining user feedback on this
subject.

>   . I think we should have a variant of yank-media that yanks "the
>     best" of the available formats without asking, based on the above
>     rules

I like your idea of reusing C-u for this.

> If after all of this people still want a global "preference", I won't
> mount the barricades to fight that, although, as I explained, such a
> global preference makes little sense to me, and sounds like an
> inferior replacement for the missing built-in preference rules I think
> we should have.

The built-in preference rules is already a global "preference" from
where I stand.  Only that, the OP wanted control over these rules.