Text2image SIGSEGV on Linux

[Shreeshrii]

@Shreeshrii @stweil : please create separate issue for command that crash on linux, so we can track it.

That command also crashes with SIGSEGV on Linux. This is a bug which needs a fix.

text2image --fonts_dir= --text ./langdata/san.training_text --outputbase san.exp-1 --ptsize=32 --strip_unrenderable_words --fontconfig_refresh_config_file=false --leading=32 --char_spacing=0.0 --exposure=-1 --find_fonts --min_coverage=.9 --degrade_image=1 --underline_start_prob=.05 --underline_continuation_prob=.01

Ref: Issue 396

[zdenop]

It does not crash on openSUSE linux:

text2image_log.txt

[stweil]

I noticed that the crash is a "feature", caused by an assertion if langdata/san.training_text does not exist. Tesseract forces a SIGSEGV for assertions to improve debug information.

IMHO missing data is a typical user error which should be reported, but not throw an assertion. So replacing the assertion by a conditional tprintf(...); exit(1); in training/fileio.cpp might be the correct solution. If you agree, I'll send a pull request.

[amitdo]

@stweil, right!
See #345, #349

[zdenop]

@stweil OK

[Shreeshrii]

FYI, in my testing on Windows 10, the crash was unrelated to 'langdata/san.training_text does not exist' since the file was very much there.

I will test further after new windows binary is made available with all these new patches.

Thanks for looking into this.

[stweil]

@amitdo, thanks for the pointers. PR #402 now does something similar for fileio.