freenode/#shirakumo - IRC Chatlog

10:21:32 Shinmera My calendar is now available for pre-purchase!

10:21:34 Shinmera https://gum.co/yukari2020

10:21:34 Colleen gumroad.com/l/yukari2020 Website (HTML), Title: 紫の世界: Endless Depths 2020 Desktop Calendar

11:45:13 Shinmera I'm looking for people that aren't me and are willing to fix up uax-9

12:51:29 Shinmera I really don't get how some of the reorderings are supposed to work

12:51:42 Shinmera I see that like one value is in a place it doesn't expect

12:51:51 Shinmera but the reordering I produce is the same as the one of the reference implementation.

12:52:04 Shinmera and the levels are already correct, the test for that passes.

12:52:20 Shinmera Well, correct within the bounds of the unspecified values.

14:40:36 Shinmera Slowly whittling down the failing tests.

14:40:48 Shinmera Still dunno what to do about the reordering failures, though.

15:12:43 Shinmera Hmm, another failure that's in the reference implementation...

15:15:13 Shinmera There's a test that seems to expect me to match up two different types of brackets as the same thing.

15:16:33 Shinmera Test is: a 〈b.1〉

15:17:59 Shinmera Either it expects me to match them, or it doesn't and I'm not handling unmatched parens right (even though the ref impl does the same)

15:18:02 Shinmera sighh

15:20:45 Shinmera Total: 1'815'582

15:20:48 Shinmera Passed: 1'617'152 (89%)

15:20:50 Shinmera Failed: 198'430 (11%)

15:21:07 Shinmera Not sure how many of those failures are due to the weird ass X reordering tests.

15:25:00 Shinmera If I skip all reordering tests:

15:25:02 Shinmera Total: 953'642

15:25:05 Shinmera Passed: 953'636 (100%)

15:25:07 Shinmera Failed: 6 ( 0%)

15:25:16 Shinmera Those 6 are due to the bracket stuff from above that I don't know how to deal with

17:33:15 Shinmera Comparing the Unicode people's reference implementations it seems the Java implementation is just bugged

17:33:18 Shinmera Eg contrast

17:33:20 Shinmera https://www.unicode.org/cldr/utility/bidic.jsp?s=%D7%90+%E2%8C%A9%D7%91.1%E3%80%89&b=0&u=110&d=2

17:33:22 Colleen www.unicode.org/cldr/utilit... Website (HTML), Title: Unicode Utilities: BIDI (UBA) C Reference

17:33:29 Shinmera Resolved Levels: [1 1 1 1 1 2 1]

17:33:34 Shinmera To

17:33:36 Shinmera https://www.unicode.org/cldr/utility/bidi.jsp?a=%D7%90+%E2%8C%A9%D7%91.1%E3%80%89&p=LTR

17:33:37 Colleen www.unicode.org/cldr/utilit... Website (HTML), Title: Unicode Utilities: BIDI (UBA)

17:34:07 Shinmera Resulting Level L1 L1 L1 L1 L1 L2 L0

17:34:12 Shinmera The latter is invalid

17:34:15 Shinmera and what I get, too

17:34:36 gingerale Huh, guess that explains things then?

17:35:46 Shinmera Yeah, it just means I wasted a week trying to port an implementation that's broken

17:35:49 Shinmera which is real great

17:48:53 Shinmera The C implementation is some 10'000 lines

17:48:56 Shinmera (not joking)

18:44:53 Shinmera Okey, regarding the brackets cases

18:45:30 Shinmera it seems there was a missing special rule for U+232A U+2329 U+3008 U+3009

18:46:27 Shinmera With that rule added all the tests from the bracket cases now pass (sans reordering tests)

19:09:27 Shinmera Hm

19:09:43 Shinmera the reordering logic in both fribidi and the c reference implementation is a lot more complex

19:09:52 Shinmera so I guess that's where the last of the bugs hide