- From: Andy Seaborne <andy.seaborne@epimorphics.com>
- Date: Thu, 21 Mar 2013 10:47:59 +0000
- To: public-rdf-wg@w3.org
On 21/03/13 03:51, Eric Prud'hommeaux wrote: > RDF and I18N folks, we have an interesting situation where we permit > U+F900-U+FA0D to appear in local names, but advise against anything > which is not NFC. So, what do we test? The grammar is wider than the acceptable URIs in several places - it's inevitable. We're expecting URI checking to be done after parsing in a very strict implementation. So test good practice and recognize that not everywhere is completely up-to-date on everything. Andy > Everything a Turtle parser > could encouter? Currently assigned characters that are in NFC? > Identifiers consisting of a single letter 'a', under the assumption > that all others will work by extension? > > > * Gavin Carothers <gavin@carothers.name> [2013-03-20 13:17-0700] >> http://www.unicode.org/charts/PDF/U1F00.pdf U+1FFF is not a character. >> http://www.unicode.org/charts/PDF/U2150.pdf U+218F is not a character. >> No chart for code point U+2FEF could be located. Most likely this is >> because no character is assigned to this code point yet. >> http://www.unicode.org/charts/PDF/UD7B0.pdf U+D7FF is not a character. >> http://www.unicode.org/charts/PDF/UFB50.pdf U+FDCF is not a character. >> No chart for code point U+EFFFF could be located. Most likely this is >> because no character is assigned to this code point yet >> >> >> New string based on the above missing characters tested in Python 3.3 >> (earlier versions of python not supported, only one with Unicode 6.1.0) > > I banged briefly on finding an ubuntu package for Python 3.3 > (currently at 3.2). Ended up with something called perl. sigh. > > use Unicode::Normalize; > $s = "AZaz\x{00c0}\x{00d6}\x{00d8}\x{00f6}\x{00f8}\x{02ff}\x{0370}\x{037d}\x{0384}\x{1ffe}\x{200c}\x{200d}\x{2070}\x{217f}\x{2c00}\x{2fcf}\x{3001}\x{d7fb}\x{f900}\x{fdc7}\x{fdf0}\x{fffd}\x{00010000}\x{0001f52b}"; > p $s cmp NFC($s); > => 1 -- strings are different. so now to look for the first candidate: > > for (0xf900..0xfdcf) { > if (ord(Unicode::Normalize::NFC(chr($_))) == $_) { > printf("%x\n", $_); > last; > } > } > => fa0e > > # checked with > $s = "AZaz\x{00c0}\x{00d6}\x{00d8}\x{00f6}\x{00f8}\x{02ff}\x{0370}\x{037d}\x{0384}\x{1ffe}\x{200c}\x{200d}\x{2070}\x{217f}\x{2c00}\x{2fcf}\x{3001}\x{d7fb}\x{fa0e}\x{fdc7}\x{fdf0}\x{fffd}\x{00010000}\x{0001f52b}"; > p $s cmp NFC($s); > => 0 -- equivalent > > The currently unassigned characters don't impact NFC: > $s = "AZaz\x{00c0}\x{00d6}\x{00d8}\x{00f6}\x{00f8}\x{02ff}\x{0370}\x{037d}\x{037f}\x{1fff}\x{200c}\x{200d}\x{2070}\x{218f}\x{2c00}\x{2fef}\x{3001}\x{d7ff}\x{fa0e}\x{fdcf}\x{fdf0}\x{fffd}\x{10000}\x{effff}" > p $s cmp NFC($s); > => 0 -- equivalent > > >> import unicodedata >> s = >> "AZaz\u00c0\u00d6\u00d8\u00f6\u00f8\u02ff\u0370\u037d\u0384\u1ffe\u200c\u200d\u2070\u217f\u2c00\u2fcf\u3001\ud7fb\uf900\ufdc7\ufdf0\ufffd\U00010000\U0001f52b" >> >> def display_string(s): >> for c in s: >> print("""Character: {c!s} >> Codepoint: {code:x} >> Name: {name} >> Combining: {combining} >> """.format( >> c=c, >> code=ord(c), >> name=unicodedata.name(c), >> combining=unicodedata.combining(c), >> )) >> >> n = unicodedata.normalize("NFC", s) >> >> display_string(s) >> print("\n ------------------ \n ") >> display_string(n) >> >> assert n == s >> >> Yeah, they aren't the same. The offending character is f900: >> >> CJK COMPATIBILITY IDEOGRAPH-F900 which in normal form is CJK UNIFIED >> IDEOGRAPH-8C48 >> >> Finding something in the F900ish range is left to Eric. Script above can be >> modified until it passes. >> >> Cheers, >> Gavin >> >> >> >> >> >> On Wed, Mar 20, 2013 at 12:13 PM, Eric Prud'hommeaux <eric@w3.org> wrote: >> >>> * Andy Seaborne <andy.seaborne@epimorphics.com> [2013-03-20 17:36+0000] >>>> The TTL has U+037E but ... >>>> >>>> PN_CHARS_BASE has a hole specifically for that >>>> >>>> [#x0370-#x037D] | [#x037F-#x1FFF] >>>> >>>> => not a legal char. >>> >>> Yeah, I screwed that up. I should have gone the other way 'cause it's at >>> the bottom of a range (unlike all the other unassigned chars). Attached are >>> the same tests with s/37f/384/. Could you chop off after the "AZaz" and see >>> if that works and do a binary search to see what it's complaining about? >>> >>> I18N folks, could you tell me why an NFC validator is objecting to this >>> (beautiful) IRI and if there's some validator I can use for testing:? >>> <http://a.example/AZazÀÖØöø˿Ͱͽ΄῾⁰↉Ⰰ⿕、ퟻ豈ﷇﷰ�𐀀> >>> The goal is to test as much as possible the valid input to < >>> http://www.w3.org/TR/turtle/#grammar-production-PrefixedName>. In turtle, >>> the localName gets appended to the namespace, hence the url above. The >>> >>> [163s] PN_CHARS_BASE ::= [A-Z] | [a-z] | [#x00C0-#x00D6] | >>> [#x00D8-#x00F6] | [#x00F8-#x02FF] | [#x0370-#x037D] | [#x037F-#x1FFF] | >>> [#x200C-#x200D] | [#x2070-#x218F] | [#x2C00-#x2FEF] | [#x3001-#xD7FF] | >>> [#xF900-#xFDCF] | [#xFDF0-#xFFFD] | [#x10000-#xEFFFF] >>> >>> production is taken from <http://www.w3.org/TR/REC-xml/#NT-NameStartChar>: >>> >>> [4] NameStartChar ::= ":" | [A-Z] | "_" | [a-z] | [#xC0-#xD6] | >>> [#xD8-#xF6] | [#xF8-#x2FF] | [#x370-#x37D] | [#x37F-#x1FFF] | >>> [#x200C-#x200D] | [#x2070-#x218F] | [#x2C00-#x2FEF] | [#x3001-#xD7FF] | >>> [#xF900-#xFDCF] | [#xFDF0-#xFFFD] | [#x10000-#xEFFFF] >>> >>> >>> >>>> Removing it (Greek question mark), I then get: >>>> >>>> WARN [line: 2, col: 43] Bad IRI: >>>> <http://a.example/AZaz???????????????????????> Code: 46/NOT_NFC in >>>> PATH: The IRI is not in Unicode Normal Form C. >>>> WARN [line: 2, col: 43] Bad IRI: >>>> <http://a.example/AZaz???????????????????????> Code: 47/NOT_NFKC in >>>> PATH: The IRI is not in Unicode Normal Form KC. >>>> WARN [line: 2, col: 43] Bad IRI: >>>> <http://a.example/AZaz???????????????????????> Code: >>>> 56/COMPATIBILITY_CHARACTER in PATH: TODO >>>> >>>> with or without the last char. >>>> >>>>> I poked around looking for composing characters in the PN_CHARS_BASE >>>>> character ranges. \u02ff MODIFIER LETTER LOW LEFT ARROW seemed like it >>>>> could be a culprit, but fileformat.info claims it's not in a combining >>>>> class. Likewise \ufffd REPLACEMENT CHARACTER >>>>> >>>>> There are a bunch of yet-unassigned characters which could be confusing >>>>> a vigilent IRI checkr. I've mapped those to the highest currently- >>>>> assigned characters in their respective range (per fileformat.info): >>>>> >>>>> \u037f 37e >>>>> \u1fff 1ffe >>>>> \u218f 2189 >>>>> \u2fef 2fd5 >>>>> \ud7ff d7fb >>>>> \ufdcf fdc7 >>>>> \U000effff e01ef >>>>> >>>>> attached is a variant of >>>>> localName_with_PN_CHARS_BASE_character_boundaries.{nt,ttl} >>>>> with the values substituted. (I pass this modified test so there >>>>> shouldn't be any typos in it.) If it still doesn't work, try chopping >>>>> off the last character 'cause it's a variation selector which ostensibly >>>>> is NF{,K}{C,D} valid, but may not have been when jjc wrote your checker. >>>>> >>>>> >>>> >>> >>> -- >>> -ericP >>> >>> >
Received on Thursday, 21 March 2013 10:49:19 UTC