Auto generate unicode property tests.

[zherczeg]

I have created a new test (test26) which is auto-generated by a script in maint. This test should be recreated after a unicode update to a new version.

[zherczeg]

I wanted to generate negative tests, but I got unexpected (I could says disturbing) results.

For example Scripts.txt says:
A8F2..A8F7 ; Devanagari # Lo [6] DEVANAGARI SIGN SPACING CANDRABINDU..DEVANAGARI SIGN CANDRABINDU AVAGRAHA

And ScriptExtensions.txt says:
A8F3 ; Deva Taml # Lo DEVANAGARI SIGN CANDRABINDU VIRAMA

My question is: if a character is defined as Devanagari, why it is defined as a script extension for Devanagari? This is completely illogical and looks like a bug in the data files. Does anybody know the reason? If we cannot trust the Script Extension data, we need to rewrite the generator completely :(

[zherczeg]

I think the script overlapping problem does not cause any practical problem at the moment, but it is good to keep in mind. The generator removes the not-working negative tests by parsing the Scripts.txt. It does not try to find another negative test though, so currently there is no negative test for "Cyrillic" script.

[zherczeg]

Reworked the script to include normal scripts. This way we can check that our hand maintained datasets are correct.