Re: isearch for doc-view.el

[Tassilo Horn <tassilo@member.fsf.org>]

Juri Linkov <juri@jurta.org> writes:

>>> Isearch on the plain text will provide enough cues to visually
>>> locate the place of the found string on the rendered image.
>>
>> There are several problems with it:
>>
>>   1. The image scrolling commands only scroll on images, so you couldn't
>>      scroll down to the text with <down> or C-n.
>>
>>      The "Page 1 of 200" text isn't there anymore due to this.  Now it's
>>      in the mode-line.  (Thanks to Stefan)
>>
>>   2. My main problem with the search approach was that the real isearch
>>      happens in another buffer in the background and the doc-view buffer
>>      has to be updated accordingly by switching to the right page.  With
>>      your approach that would be even harder.  Here, you would have to
>>      do the same plus copying the page with the current match from the
>>      text representation's buffer to the doc-view buffer plus an
>>      additional isearch in the doc-view buffer to highlight the match.
>
> It seems the hardest problem here is fixing scrolling commands to take
> into account the text below the image.  Copying the text of the
> current page and starting isearch on it should be easy.

Feel free to give it a try.  I'm don't know enough about the isearch
internals to make it working in an acceptable amount of time, but I
won't be against it if it feels good.  I usability problem I see is that
because of isearch the buffer is scrolled to the matching text and the
user would have to scroll back to the image after every C-s.

Basically I'm completely satisfied with the behavior I implemented in
<87640fzs9o.fsf@baldur.tsdh.de>, but probably others are more demanding.

> No, I meant displaying only one image and text from the same image at
> once:
>                                 page1.png
>                                 page1 text
>
> But maybe for small documents (with small number of pages) displaying
> all images at once is more preferable?

I don't know if the benefits would be great enough to complicate the
code.  At least most navigation and search commends would need to be
updated accordingly.

If someone implements that and it works, it's fine with me, but I won't
put it on my todo list for now.

Bye,
Tassilo