Make TessdataManager able to save archive using LibArchive

[sadra-barikbin]

• Made TessDataManager able to save archive using LibArchive
• Added -t option to combine_tessdata to transform proprietary .traineddata to archive file.

[sadra-barikbin]

@stweil , shall I add a test?

[stweil]

With this change TessdataManager::SaveFile will always write traineddata files in ZIP format which are incompatible with Tesseract binaries which were build without LibArchive. I'm afraid that would cause problems for a lot of people.

[egorpugin]

I thought libarchive can deduce archive types.

[stweil]

Yes, it can, but the current code uses archive_write_set_format_zip instead of archive_write_set_format_by_name, so it will always write a ZIP file. And of course libarchive cannot write the proprietary traineddata format.

[stweil]

Added -t option to combine_tessdata to transform proprietary .traineddata to archive file.

I'd prefer a more general solution which allows different target formats. In addition it should allow writing to a different output file and use long options. So the syntax might look like this:

Usage: combine_tessdata --convert [--format TARGET_FORMAT] INFILE [OUTFILE]

TARGET_FORMAT would default to zip, but should allow any typical file extension of archive files and also traineddata for conversions into the proprietary Tesseract format.

[egorpugin]

Isn't automatic format simpler?
Libarchive writes files based on their extension.
If we write .zip - it will be .zip.
.tar.gz - tar.gz
.tar.xz or lz - ...

[stweil]

That's right, and this feature of LibArchive would also be used to implement my suggested solution.

If we implement support for combine_tessdata --convert eng.traineddata eng.zip, that would create a filename which is currently unsupported by tesseract, so an additional renaming mv eng.zip eng.traineddata would be required. Should we extend the Tesseract code to support more extensions than the current .traineddata, so -l eng.zip would work? Or do you think of another solution? Maybe eng.traineddata.zip?

[zdenop]

Is there also intention to read such converted data by tesseract?

If yes, than please be careful about changing extension: it will break a lot of workflows that looks for available/installed languages (AFAIR also GetAvailableLanguagesAsVector a.k.a tesseract --list-langs)

Nowadays it is quite common to use private file extension instated of indication it is archive (e.h. xlsx, odt are zip archives)

On other hand: if file extension will not be changed and tesseract will be build without libarchive support, that has to be improved error handling why tesseract is not able read traineddata...

[sadra-barikbin]

I thought this feature is decided on to be implemented. Now it seems it's an arguable one. My own requirement i.e. inspecting config file in the .traineddata file and possibly overwriting it is fulfilled with the sequence of combine_tessdata -e and combine_tessdata -o. I don't see a strong point in having this feature either. Instead, I could work on PageXML renderer if you agree.