libtesseract segfaults with tesseract 4.0.0 (#104) · Issues · World / OpenPaperwork / pyocr

libtesseract segfaults with tesseract 4.0.0

Due to this commit, the libtesseract interface segfaults (at least when running tests).

It seems like tesseract >= 4.0.0 needs that binding programs are configured with LC_ALL=C.

I came up with a quick-and-dirty fix (see attached patch 0001-Set-locale-for-tesseract-4.0.0.patch), setting locale before initializing tesseract API and unsetting after:

in src/pyocr/libtesseract/tesseract_raw.py:343

def init(lang=None):
    assert(g_libtesseract)
    locale.setlocale(locale.LC_ALL, "C")
    handle = g_libtesseract.TessBaseAPICreate()
    locale.setlocale(locale.LC_ALL, "")

I think that since the library expect to be in C locale unsetting it after is probably a bad idea and should be set in src/pyorc/libtesseract/__init__.py but in that case there could be unexpected side issues (like lang detection in paperwork?). I don't know if the locale.setlocale function is limited to the namespace.

Due to this [commit](https://github.com/tesseract-ocr/tesseract/commit/3292484f67af8bdda23aa5e510918d0115785291), the libtesseract interface segfaults (at least when running tests).

It seems like tesseract >= 4.0.0 needs that binding programs are configured with `LC_ALL=C`.

I came up with a quick-and-dirty fix (see attached patch [0001-Set-locale-for-tesseract-4.0.0.patch](/uploads/a2c6db123baaacba97e54f559342109a/0001-Set-locale-for-tesseract-4.0.0.patch)), setting locale before initializing tesseract API and unsetting after:

in `src/pyocr/libtesseract/tesseract_raw.py:343`
```python
def init(lang=None):
    assert(g_libtesseract)
    locale.setlocale(locale.LC_ALL, "C")
    handle = g_libtesseract.TessBaseAPICreate()
    locale.setlocale(locale.LC_ALL, "")
```
I think that since the library expect to be in `C` locale unsetting it after is probably a bad idea and should be set in `src/pyorc/libtesseract/__init__.py` but in that case there could be unexpected side issues (like lang detection in paperwork?). I don't know if the `locale.setlocale` function is limited to the namespace.

Discussion
Designs