Package r-tesseract and upgrade tesseract-ocr to 5.x about guix-science

[PromyLOPh]

Hm, both tesseract-ocr and r-tesseract seems like they could be included in Guix proper and thus they should go there. I’m not sure about the training data files though. According to the repo they seem to be open source, with the source available in theory, but no instructions how to build them. Would be nice to have them in Guix proper too though, since tesseract is quite useless without data.

edit: It seems the LSTM source data is in this repo: https://github.com/tesseract-ocr/langdata_lstm

[wdkrnls]

My packaging foo is quite weak, but is packaging this training data a use case for the copy-build-system? This just reminded me of something like a vim plugin which uses that. I'm also not sure what to do about the fact that this git repository of traineddata files is huge. The eng.traineddata file 22.4 Mb by itself and there are dozens of languages. Can git-fetch handle partial shallow sparse clones?

I saw this for tesseract 4 which looks like a build plan: https://github.com/tesseract-ocr/tesstrain

[PromyLOPh]

I’m not sure tesstrain would work. As far as I know it expects image files, which the langdata repo does not contain. The latter relies on something to generate images from fonts and the supplied text. Also apparently it takes weeks to train the LSTM, so maybe not a good idea to build it from source.

But yeah, the binary training data is a case for copy-build-system. I think big is not that much of a problem, but I’m not sure about shallow/sparse clones.