To save me digging through and testing, why does there have to be an additional new line added at the end? Is it the while read or xargs?

I'm not sure about xargs. But read only splits strings at newlines, so it ignores the final line that does not finish with a LF.

There are tricks to also handle the final line in a loop, such as while read line || [ -n "$line" ]; do ...; done, but well, it feels that's better just to add a final LF :)

That link was fascinating, thank you. Let me ponder it further.

First - thank you so much for taking the time to do this and raise such a well documented PR.

Second, sorry it's taken until now to reply.

I have given this a LOT of thought. And I have decided not to merge the PR. The main problem for me is the additional dependencies of img2pdf. There are users who really don't want additional dependencies and even go so far as to edit the Dockerfile to reduce the image size. It should not stop you from running your code - although you will need to adapt your config.local.js to set your pipelines.

What I have done is raise #569 which will make it so your pipelines will work easily.

What I am also going to do is see about merging in the while read filename... code - so I am not yet closing this PR as I want to cherrypick.

Thanks for your message. img2pdf indeed is quite a large dependency, especially as it pulls the whole Python runtime. I understand that you are wary including it.

• As you said, it does not prevent me from using it, I'll just have to include my own pipelines in config.local.js.
  I am using Docker though. I had tried to make an extended image using a Dockerfile that reads FROM sbs20/scanservjs:... ; RUN apt-get update && apt-get install -y img2pdf. However, when running docker build, apt-get fails for some reason (I cannot reproduce as I cannot access my raspberry pi right now). Something about missing permissions to perform whatever late installation step. Does this ring a bell to you? Do you know how I could still install img2pdf in an extended Docker image, without building the whole image from scratch?
• Not upstreaming these changes is too bad after all, because it makes scanservjs not work out-of-the-box for raspberry pi users that have a scanner with ADF. That's too bad because scanservjs is, by far, the best (and only?) solution I found to scan on a headless raspberry pi.
  Img2pdf documentation, in its last paragraph, suggests tesseract is able to create PDFs without re-encoding. That would be great because it already is in the Docker image, but I could not make it work without having it insist on doing any OCR. Do you think it is possible though? You're probably more familiar with tesseract as I am.
  Otherwise, do you think you would be OK to either
  • publish two images (say, the "regular" scanservjs:$version image, plus an image tagged scanservjs:$version-with-extras)?
  • Even without img2pdf, there still is the "file-by-file conversion" from this PR that does not require any extra dependency, that seems useful for lower-end machines (and that, I think, does not cause any harm on more up-to-date machines).
  • At least write instructions in documentation files for raspberry pi users?

Out of curiosity, what machine are you using scanservjs on?
Thanks for your feedback!

Something about missing permissions to perform whatever late installation step. Does this ring a bell to you?

It all depends on what the ... is in FROM sbs20/scanservjs:.... If you're using this then it switches user and you won't have sufficient privs. You should be using this instead. If it's not that, then I don't know I'm afraid.

But I could not make it work without having it insist on doing any OCR. Do you think it is possible though?

No idea I'm afraid - I am actually fairly unfamiliar with Tesseract, sorry!

publish two images (say, the "regular" scanservjs:$version image, plus an image tagged scanservjs:$version-with-extras)

I would gladly link to someone else's image. I am less keen to add and maintain yet another image. I had never intended to support Docker - it adds a lot of support overhead, but somehow did anyway; this project is already placing heavy demands on my time.

there still is the "file-by-file conversion" from this PR that does not require any extra dependency

Yep - and I may yet include that, I just couldn't find the time to get it done and fully tested today. That's what I meant above with What I am also going to do is see about merging in the while read filename... code. I wanted to make the change in #569 which would at least mean you could use standard configuration changes.

At least write instructions in documentation files for raspberry pi users

Indeed. I do always try to do this - but I would welcome contributions.