Recognition on x32

[qb1t0]

---

Environment

• Tesseract Version: 4.0.0-beta4
• Platform: Ubuntu 16.04 x32

Current Behavior:

Bad text recognition on x32-bit server ;C

Expected Behavior:

Good recognition.

Suggested Fix:

Additional building flags ?

[stweil]

Support for x32 might indeed be interesting, because it should be significantly faster than 32 bit code and could even be slightly faster than 64 bit code because of better cache usage. I expect that it will require some work on the Tesseract code base to fix all portability issues. See https://en.wikipedia.org/wiki/X32_ABI and https://wiki.debian.org/X32Port for a few first hints.

[stweil]

@qb1t0, is the recognition on other installations better? Could you provide an example for the bad text recognition (image, command line, results)?

[stweil]

With the test image from tesseract-ocr/tessdata#108, I always get the same results for x32 and for x86_64 when I add --psm 6.

There is a difference when running Running tesseract test.tif -: x32 shows the well know assertion contains_unichar_id(unichar_id):Error:Assert failed:in file ../../src/ccutil/unicharset.h, line 509 with both Tesseract installations.

[Shreeshrii]

@zdenop Please label as Feature Request.

[stweil]

I think it can be closed, as x32 and x86_64 deliver the same results according to my tests.