At a fundamental level, this extension grants access to a number of functions provided by the libidn and libidn2 libraries. Not all of the functions provided by these libraries are exposed.
Whenever possible, the code will transform the encoding of the data in the database to UTF-8 before further processing.
Installation in postgresql is fairly simple:
CREATE EXTENSION idn;
as part of the `sql/idn.sql`_ file. Please let somebody know if the documentation diverges from the actual (tested) API.
Many examples taken from http://www.unicode.org/faq/idn.html.
check a string to see if it trips the PR29 check:
select idn_pr29_check(u&'d\0061t\+000061'::text); idn_pr29_check ---------------- t (1 row) select idn_pr29_check(u&'d\0B47\0300\0B3E'::text); idn_pr29_check ---------------- f (1 row)
idna (2003) encoding and decoding:
select idn_idna_decode('xn--bcher-kva.de'); idn_idna_decode ----------------- bücher.de (1 row) select idn_idna_encode('bücher.de'); idn_idna_encode ------------------ xn--bcher-kva.de (1 row) select idn_idna_encode(u&'b\0075\0308cher.de'); idn_idna_encode ------------------ xn--bcher-kva.de (1 row) select idn_idna_decode('xn--bcher-kva.de', 'IDNA_FLAG_NONE'); idn_idna_decode ----------------- bücher.de (1 row) select idn_idna_encode('bücher.de', 'IDNA_FLAG_NONE'); idn_idna_encode ------------------ xn--bcher-kva.de (1 row) select idn_idna_encode(u&'b\0075\0308cher.de', 'IDNA_FLAG_NONE'); idn_idna_encode ------------------ xn--bcher-kva.de (1 row)using the IDNA_USE_STD3_ASCII_RULES flag:
First, an example without the flag:
select idn_idna_encode('foo.ba_r.baz', 'IDNA_FLAG_NONE'); idn_idna_encode ----------------- foo.ba_r.baz (1 row)And now with the flag:
select idn_idna_encode('foo.ba_r.baz', 'IDNA_FLAG_USE_STD3_ASCII_RULES'); WARNING: Error encountered converting from IDNA2003 to ASCII: Non-digit/letter/hyphen in input idn_idna_encode ----------------- (1 row) select idn_idna_encode('foo.ba_r.ba--z', 'IDNA_FLAG_USE_STD3_ASCII_RULES'); WARNING: Error encountered converting from IDNA2003 to ASCII: Non-digit/letter/hyphen in input idn_idna_encode ----------------- (1 row) -- check the not-equal sign select idn_idna_encode(u&'foo.bar.\2260baz', 'IDNA_FLAG_NONE'); -- success idn_idna_encode ---------------------- foo.bar.xn--baz-dl2a (1 row) select idn_idna_encode(u&'foo.bar.\2260baz', 'IDNA_FLAG_USE_STD3_ASCII_RULES'); -- fail idn_idna_encode ---------------------- foo.bar.xn--baz-dl2a (1 row)punycode encoding and decoding:
select idn_punycode_encode('bücher.de'); idn_punycode_encode --------------------- bcher.de-65a (1 row) select idn_punycode_encode(u&'b\0075\0308cher.de'); idn_punycode_encode --------------------- bucher.de-hkf (1 row) select idn_punycode_decode('bcher.de-65a'); idn_punycode_decode --------------------- bücher.de (1 row)check NFKC normalization:
Start by showing that the decomposed form is not equal to the composed form:
select u&'\0065\0301' = u&'\00e9'; ?column? ---------- f (1 row)
And now show the NFKC normalization:
select idn_utf8_nfkc_normalize(u&'\0065\0301') = u&'\00e9'; ?column? ---------- t (1 row)
Show that superscript '9' gets normalized to just '9':
select idn_utf8_nfkc_normalize(u&'\2079') = '9'; ?column? ---------- t (1 row)
Some comparisons between IDNA 2003 and 2008 follow.
LATIN SMALL LETTER SHARP S encodes differently:
select idn_idna_encode(u&'\00DF') = 'ss'; ?column? ---------- t (1 row) select idn2_lookup(u&'\00DF') = 'xn--zca'; ?column? ---------- t (1 row)
show register vs. lookup for IDNA 2008:
select idn2_register(u&'\00DF', NULL, 'IDN2_FLAG_NONE') = 'xn--zca'; ?column? ---------- t (1 row)
more examples from the FAQ:
-- U+03C2 GREEK SMALL LETTER FINAL SIGMA -- U+200C ZERO WIDTH NON-JOINER -- U+200D ZERO WIDTH JOINER -- (öbb.at) -- should be same in 2003/2008/UTS46 select idn_idna_encode(u&'\00f6bb.at') = 'xn--bb-eka.at'; ?column? ---------- t (1 row) select idn2_lookup(u&'\00f6bb.at') = 'xn--bb-eka.at'; ?column? ---------- t (1 row) -- FIXME: libidn2 says this is disallowed for registration. -- AFAICT, the Unicode FAQ page doesn't say anything about -- it one way or the other. select idn2_register(u&'\00f6bb.at', NULL, 'IDN2_FLAG_NONE'); -- disallowed? WARNING: Error encountered performing idn2 register: string contains a disallowed character idn2_register ------------------ (1 row) -- (ÖBB.at) -- 2003 allows (w/case change) -- UTS46 allows (w/case change) -- 2008 disallows select idn_idna_encode(u&'\00d6BB.at') = 'xn--bb-eka.at'; ?column? ---------- t (1 row) select idn2_lookup(u&'\00d6BB.at'); -- should disallow WARNING: Error encountered performing idn2 lookup: string contains a disallowed character idn2_lookup ---------------- (1 row) select idn2_register(u&'\00d6BB.at', NULL, 'IDN2_FLAG_NONE'); -- should disallow WARNING: Error encountered performing idn2 register: string contains a disallowed character idn2_register ------------------ (1 row) -- (√.com) -- 2003 + UTS46 allow, 2008 disallows select idn_idna_encode(u&'\221a.com') = 'xn--19g.com'; ?column? ---------- t (1 row) select idn2_lookup(u&'\221a.com'); -- should disallow WARNING: Error encountered performing idn2 lookup: string contains a disallowed character idn2_lookup ---------------- (1 row) select idn2_register(u&'\221a.com', NULL, 'IDN2_FLAG_NONE'); -- should disallow WARNING: Error encountered performing idn2 register: string contains a disallowed character idn2_register ------------------ (1 row) -- (faß.de) -- valid across all three, but difference answers for 2008 select idn_idna_encode(u&'fa\00df.de') = 'fass.de'; -- 2003 ?column? ---------- t (1 row) select idn2_lookup(u&'fa\00df.de') = 'xn--fa-hia.de'; -- 2008 ?column? ---------- t (1 row) -- FIXME: libidn2 says this is disallowed for registration. -- AFAICT, the Unicode FAQ page doesn't say anything about -- it one way or the other. select idn2_register(u&'fa\00df.de', NULL, 'IDN2_FLAG_NONE'); -- disallowed for reg? WARNING: Error encountered performing idn2 register: string contains a disallowed character idn2_register ------------------ (1 row) -- FIXME: unknown what UTS46 does -- (ԛәлп.com) -- valid for 2003 lookup, but not registration -- valid for 2008 + UTS46 select idn_idna_encode(u&'\051b\04d9\043b\043f.com'); -- should fail (2003) WARNING: Error encountered converting from IDNA2003 to ASCII: String preparation failed idn_idna_encode ----------------- (1 row) select idn_idna_encode(u&'\051b\04d9\043b\043f.com', 'IDNA_FLAG_ALLOW_UNASSIGNED') = 'xn--k1ai47bhi.com'; -- 2003 ?column? ---------- t (1 row) select idn2_lookup(u&'\051b\04d9\043b\043f.com') = 'xn--k1ai47bhi.com'; -- 2008 ?column? ---------- t (1 row) -- FIXME: libidn2 says this is disallowed for registration. -- AFAICT, the Unicode FAQ page doesn't say anything about -- it one way or the other. select idn2_register(u&'\051b\04d9\043b\043f.com', NULL, 'IDN2_FLAG_NONE'); -- disallowed for 2008 reg? WARNING: Error encountered performing idn2 register: string contains a disallowed character idn2_register ------------------ (1 row) -- (Ⱥbby.com) -- valid for 2003 lookup, not registration -- valid for UTS46 (w/case change) -- NOT valid for 2008 select idn_idna_encode(u&'\023abby.com'); -- should fail WARNING: Error encountered converting from IDNA2003 to ASCII: String preparation failed idn_idna_encode ----------------- (1 row) select idn_idna_encode(u&'\023abby.com', 'IDNA_FLAG_ALLOW_UNASSIGNED') = 'xn--bby-spb.com'; -- 2003 ?column? ---------- t (1 row) select idn2_lookup(u&'\023abby.com', 'IDN2_FLAG_NONE'); -- 2008, fails WARNING: Error encountered performing idn2 lookup: string contains a disallowed character idn2_lookup ---------------- (1 row) select idn2_register(u&'\023abby.com', NULL, 'IDN2_FLAG_NONE'); -- 2008, fails WARNING: Error encountered performing idn2 register: string contains a disallowed character idn2_register ------------------ (1 row)
- The TLD checking code is not exposed. Primarily, this is due to the fact that the tables are hard-coded into the library and may be out-of-date. With the appropriate warning/caveat noted, however, it may be reasonable to include them.
- the punycode functions in libidn expose a facility for case-folding. Since PostgreSQL already includes case-folding smarts, the value-add wasn't deemed worth the extra complexity cost.
- document the stringprep portion