Ftr minimal digit

[bnguyenvanyen]

So this has got a number of changes :

• Builders inherit from BaseBuilder (easy to drop if you want to)
• Builders' class attributes become instance attributes
• DigitBuilder moved to builders.py
• DigitLineBoxBuilder for digits inherits from LineBoxBuilder
• libtesseract is available with version 3.3
• Tests have been restructured to reduce code redundancy. Some mixins are defined in tests/tests_base.py then multiple inheritance is used for the actual tests.
• Tests have been added for the new builders.
• test_version in tests_libtesseract passes with version (3,4,1).

On my system (libtesseract3.4.1) most of the tests pass except :

• The cuneiform tests since I don't have it installed
• The libtesseract words and lines japanese tests
• The libtesseract TestBasicDoc test
• The line digits test since the files test-digits.lines don't exist

[jflesch]

Wow, nice ! There is obviously a lot of work here :) However, it's going to take me a while to review it. I'll add comments as I read the code. Please do not update the code until I'm finished (following the discussions would be a mess otherwise).

edit: oh actually, I didn't notice that Github added a button "finish review" to not publish every comment immediately :)

[bnguyenvanyen]

Hi, thank you for your reactivity ! I put in the minor tweaks.

I'm now working on a way to make the builder system more flexible (I rewrote libtesseract.image_to_string) but it would break any code inheriting from the existing builders, so it will probably have to stay on my fork.

If some day you're cleaning up the builders and breaking backward-compatibility anyway you might think about it !

Cheers

[jflesch]

My bad, I totally forgot to review this pull request, sorry :(

Anyway, it's perfect, I'm merging it. Thank you very much for this contribution :)

[jflesch]

It will be included in 0.4.3. I'll release today (if the tests are fine)