"gb18030 ranges" have problematic definitions about encoding

[annevk]

Why would the gb18030 encoder algorithm ever hit the gb18030 ranges index for U+8000? It's in the gb18030 index.

[peteroupc]

Well then, here is another example of a problematic code point, and this time it doesn't appear in "gb18030 index": U+E5E5.

58853 ---> 19043, 0x9FA6 ---> 19043 + 58853 - 40870 ---> 37026

but:

37026 ---> 33550, 0xE865 ---> 59493 + 37026 - 33550 ---> 62969

and 69292 (F5F9 differs from E5E5).

[annevk]

Hmm, not immediately sure why that would fail. I assume the decoder/encoder has the same problem here? Doesn't seem like the byte conversion part should matter much.

[annevk]

@r12a any ideas? It seems that for your tests at http://www.w3.org/International/tests/repo/results/encoding-dbl-byte.en browsers mostly pass.

[r12a]

i didn't test all the BMP for gb18030, just selected ranges, so i missed that particular character. However, there are now two new tests that do test it, and find it (and only that one in the PUA) to be problematic on Firefox and Chrome (didn't try any others).

http://www.w3.org/International/tests/repo/encoding/legacy-mb-schinese/gb18030/gb18030-encode-form-other-pua.html

http://www.w3.org/International/tests/repo/encoding/legacy-mb-schinese/gb18030/gb18030-decode-other-pua.html

i won't integrate these tests into the i18n test suite fully until we decide what needs to be done about this.

i don't have any clear idea about why this fails, although peter's suggestion seems plausible (note the that problem character is the first one in the PUA (that isn't in the index)).

(http://r12a.github.io/apps/encodings/ also shows the behaviour @peteroupc describes)

[vyv03354]

Also, if #22 does not change the mapping for 0x8135F437 from U+1E3F to U+E7C7, another code point would have the problem.

[annevk]

@r12a I'm confused by the outcome of your tests since it suggests browsers simply do not encode U+E5E5 at all despite gb18030 supposedly being a UTF (both Chrome and Firefox emit an "HTML entity").

[vyv03354]

For Firefox, this is intentional because some sites expected a space for gbk 0xA3A0.
https://bugzilla.mozilla.org/show_bug.cgi?id=131837
But this bug is archaic. I'm fine with changing back the mapping to U+E5E5 to align with GB18030 spec/IE/Edge.

[annevk]

Ah, that is the problem. "This matches the GB18030-2000 standard for code points encoded as two bytes, except for 0xA3 0xA0 which maps to U+3000 to be compatible with deployed content." Except we never took care of mapping U+E5E5 to an error in the encoder.

[peteroupc]

I don't agree with emitting an encoder error for the code point 0xE5E5; like "vyv03354", I also agree with changing back the mapping for this code point, if that is indeed what the GB18030 standard says.

[annevk]

Since most browsers emit an error, it seems safer to just do that. Especially since WebKit "fixed" this in 2008, six years after Gecko did. Seems likely it might still be problematic.

[annevk]

I created a PR for my proposal in #25. I would appreciate review before landing this.