Not reading image links in org-mode

[martinfowler]

Pandoc doesn't seem to be recognizing image links in org documents.

Taking this org input

* Test Images

** PDF no label

[[file:test.pdf]]

** PNG no label

[[file:test.png]]

** PDF with label

[[file:test.pdf][label]]

** PNG with label

[[file:test.png][label]]

I get the following output

foo.pdf

Only the png file is properly rendered, and then only when it has no label.

I get the same output when I produce a docx file, and something similar if I generate markdown

Test Images
===========

PDF no label
------------

[file:test.pdf](test.pdf)

PNG no label
------------

![](test.png)

PDF with label
--------------

[label](test.pdf)

PNG with label
--------------

[label](test.png)

I'm running pandoc 2.7.2 on macOs Mojave 10.14.4. I installed pandoc via homebrew

[jgm]

Looks like isImageFilename in Text.Pandoc.Readers.Org.Shared doesn't recognize the pdf extension as an image filename. I'm not sure about the label issue. @tarleb can you have a look?

[martinfowler]

I haven't done a full test of image file types, but eps has the same problem as pdf

[jgm]

(defun org-display-inline-images (&optional include-linked refresh beg end)
  "Display inline images.

An inline image is a link which follows either of these
conventions:

  1. Its path is a file with an extension matching return value
     from `image-file-name-regexp' and it has no contents.

  2. Its description consists in a single link of the previous
     type.

When optional argument INCLUDE-LINKED is non-nil, also links with
a text description part will be inlined.  This can be nice for
a quick look at those images, but it does not reflect what
exported files will look like.

When optional argument REFRESH is non-nil, refresh existing
images between BEG and END.  This will create new image displays
only if necessary.  BEG and END default to the buffer
boundaries."

image-file-name-regexp is an elisp function that returns:

"\\.\\(GIF\\|JP\\(?:E?G\\)\\|P\\(?:BM\\|GM\\|N[GM]\\|PM\\)\\|SVG\\|TIFF?\\|X\\(?:[BP]M\\)\\|gif\\|jp\\(?:e?g\\)\\|p\\(?:bm\\|gm\\|n[gm]\\|pm\\)\\|svg\\|tiff?\\|x\\(?:[bp]m\\)\\)\\'"

[tarleb]

An issue I faced while implementing this is that Emacs Org mode handles this case differently, depending on the chosen exporting format. Pandoc's org reader had to be target-format agnostic, and we chose HTML exporting as reference behavior. Exporting to HTML from Emacs results in output similar to the above, while exporting to PDF will result in an image instead of a link under "PDF no label".

Labeled image links will be output as links in all target formats.

This situation is not ideal, but I'm unsure how to improve this in a consistent way. I'm very open to suggestions.

[martinfowler]

My question would be: what do you do with markdown? [[path/to/image][image label]] in org is the same as ![label][path/to/image] in markdown. The problem (or difference) is that org-mode uses the same syntax for images as it does for web page URLs (and supports links to more things, including mail messages, elisp functions, and can be customized to add more). So I'd reckon that a good conversion first has to determine what if the target is an image, then do whatever you do for markdown. Martin Fowler http://martinfowler.com

…

On Apr 18, 2019, at 02:45, Albert Krewinkel ***@***.***> wrote: An issue I faced while implementing this is that Emacs Org mode handles this case differently, depending on the chosen exporting format. Pandoc's org reader had to be target-format agnostic, and we chose HTML exporting as reference behavior. Exporting to HTML from Emacs results in output similar to the above, while exporting to PDF will have an image instead of a link under PDF no label. Labeled image links will be output as links in all target formats. This situation is not ideal, but I'm unsure how to improve this in a consistent way. I'm very open to suggestions. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[tarleb]

I see what you mean. Org-mode syntactically conflates links and images in interesting (and to me, sometimes unintuitive) ways. I believe there is no Org mode equivalent to Markdown's ![label](path/to/image). The Org mode manual states:

An image is a link to an image file that does not have a description part.

The closes we can get is by using figures:

#+CAPTION: label
[[/path/to/image]]

Demo:

$ echo "#+CAPTION: label\n[[file:test.png]]" | pandoc -f org -t markdown
![label](test.png)

---

So I'd reckon that a good conversion first has to determine what if the target is an image, then do whatever you do for markdown.

Pandoc's readers generally have no insights into the target format and output options, but the reader (i.e., parser) already has to decide which links are actually images. With Emacs Org-mode, PDF and eps are images when outputting PDF via LaTeX, but cannot be displayed as images when exporting to HTML. The reader has to make a choice between image or link, without information on the target format's capabilities; we follow Emacs' logic for non-PDF output for this.

A better way would be to preserve this information (link or image) in the intermediate document AST. Unfortunately, we currently have no way of expressing image when used with LaTeX, link otherwise in the AST. @jgm proposed a fix for this in #547. I'm afraid there is not much we can do about this discrepancy between pandoc and Emacs unless we perform these (extensive) changes.

[martinfowler]

Ah, with that explanation the current behavior makes sense. A consequence of the differing capabilities of HTML and PDF and org-mode's origins in a text only environment. It sounds like the only thing that can realistically be done is some documentation in the org-mode features section. I haven't done much with org mode and pdf, was only really experimenting. There is a non-pandoc way for org mode to generate pdf, I wonder if more people who do this go that route instead, so haven't noticed this issue. But this is handy to know, as I'm still experimenting with org-mode and rmarkdown for some reports, and it was only with the latter that I realized that pandoc did stand-alone HTML, which is better for what I needed than the pdf I was trying out. Martin Fowler http://martinfowler.com

…

On Apr 18, 2019, at 15:09, Albert Krewinkel ***@***.***> wrote: I see what you mean. Org-mode syntactically conflates links and images in interesting (and to me, sometimes unintuitive) ways. I believe there is no Org mode equivalent to Markdown's . The Org mode manual states: An image is a link to an image file that does not have a description part. The closes we can get is by using figures: #+CAPTION: label [[/path/to/image]] Demo: $ echo "#+CAPTION: label\n[[file:test.png]]" | pandoc -f org -t markdown  So I'd reckon that a good conversion first has to determine what if the target is an image, then do whatever you do for markdown. Pandoc's readers generally have no insights into the target format and output options, but the reader (i.e., parser) already has to decide which links are an image. With Emacs Org-mode, PDF and eps are images when outputting PDF via LaTeX, but cannot be displayed as images when exporting to HTML. The reader has to make a choice between image or link, without information on the target format's capabilities; we follow Emacs' logic for non-PDF output for this. A better way would be to preserve this information (link or image) in the intermediate document AST. Unfortunately, we currently have no way of expressing image when used with LaTeX, link otherwise in the AST. @jgm proposed a fix for this in #547. I'm afraid there is not much we can do about this discrepancy between pandoc and Emacs unless we perform these (extensive) changes. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[jgm]

Pandoc will parse

[[file:test.pdf]]

from org as

Link ("",[],[]) [Str "file:test.pdf"] ("test.pdf","")

So one approach would be to use a lua filter to transform the AST before rendering to LaTeX/PDF.

-- fixpdfs.lua
function Link(el)
  if el.title == '' and el.target:match('.*[.]pdf') and pandoc.utils.stringify(el.content) == "file:" .. el.target then
      return pandoc.Image(el.content, el.target)
  end
end

Invoke using

pandoc my.org -o my.pdf --lua-filter fixpdfs.lua

[martinfowler]

That looks like it would fit the bill. When I get a chance to try it I will. Thanks for outlining that, I hadn't delved into the lua filters yet. Martin Fowler http://martinfowler.com

…

On Apr 19, 2019, at 14:41, John MacFarlane ***@***.***> wrote: Pandoc will parse [[file:test.pdf]] from org as Link ("",[],[]) [Str "file:test.pdf"] ("test.pdf","") So one approach would be to use a lua filter to transform the AST before rendering to LaTeX/PDF. -- fixpdfs.lua function Link(el ) if el.title == '' and el.target:match('.*[.]pdf') and pandoc.utils.stringify(el.content) == "file:" .. el.target then return pandoc.Image(el.content, el.target ) end end Invoke using pandoc my.org -o my.pdf --lua-filter fixpdfs.lua — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[tbruckmaier]

This seems also to be the case with uppercase image extensions:

$ echo "[[file:test.png]]" | pandoc -f org -t html
<p><img src="test.png" /></p>

$ echo "[[file:test.PNG]]" | pandoc -f org -t html
<p><a href="test.PNG">file:test.PNG</a></p>

$ pandoc --version
pandoc 2.9.2
Compiled with pandoc-types 1.20, texmath 0.12.0.1, skylighting 0.8.3.2

Is the list of valid image extensions configurable in any way?

[tarleb]

@tbruckmaier that's different to what's discussed here, please open a new issue.