Have --self-contained use alt text when images are missing instead of failing

[lucalenardi]

Explain the problem.

Not sure if this is actually a bug or an expected behaviour, so sorry if this is not the right place to report.

Starting from a simple test.md file, with a markdown command such as ![my image](./myimage.png)

If the image exists, everything is ok, but if the image does not exist, I have the following behaviours:

1. with latex output: pandoc returns a warning message on the command line regarding the missing image and replaces the image with the alternative text "my image"
2. with html output and --self-contained option: pandoc returns an missing file error: File ./myimage.png not found in resource path and exits without completing the process using the alternative text (which is the default for html).

The same also happens with raw html tags: <img> and <embded> trying to embed a pdf file in the html.

It seems to be related to the --self-contained option, as removing it actually populate the src attribute in the html tag ignoring the file existence.

It may be useful to have pandoc generates the final output with alternative texts – or maybe just completes the generation of the output file with warnings – in contexts where an automation runs a script that involves pandoc and not all assets are already available (es. some appendix as pdf, or some images).

Pandoc version?

pandoc 2.17.1.1 on Docker image pandoc/latex.

[jgm]

I'm not sure what the best behavior is. Currently we try to guarantee that a file produced with --self-contained really is self-contained. If we left the missing src in there, it wouldn't be. But I can see the point of the request. Would be interested in feedback from others.

[mb21]

Emitting a warning instead of failing hard seems like a sensible behaviour, considering that's what pdflatex and browsers do as well.

we try to guarantee that a file produced with --self-contained really is self-contained

if the image has been deleted (or was at a URL that's a 404 now), then the file cannot be any more self-contained than that, so I'd argue that principle isn't violated.

Only thing I can think of that could break with this change is CI scripts that rely on pandoc's non-zero exit code in this case, but that's a very edge case...

[cagix]

i kind of like the behaviour when using latex as backend: emit a clear warning and produce some valid output using the alt text.

[jgm]

In the PDF module we use fillMediaBag, which catches errors in fetchItem and produces a warning.
We could add similar error handling to SelfContained.

[gabyx]

@jgm An option --warnings-as-errors would have been nice. Because, as in other languages its common to have all warnings treated as errors, at least if the users wants.

[tarleb]

@gabyx --fail-if-warnings?

[gabyx]

Sorry my bad, true that exists and I am already using. Sorry.