[tracker] tests,libtracker-fts: always run full unicode tests
- From: Aleksander Morgado <aleksm src gnome org>
- To: commits-list gnome org
- Cc:
- Subject: [tracker] tests,libtracker-fts: always run full unicode tests
- Date: Thu, 3 Oct 2013 14:39:28 +0000 (UTC)
commit 30d64d6bd2b7421e82cfdd41ec8c79ae891f49d1
Author: Aleksander Morgado <aleksander lanedo com>
Date: Thu Oct 3 16:38:45 2013 +0200
tests,libtracker-fts: always run full unicode tests
We no longer have the glib/pango parser, so just enable all unicode tests.
tests/libtracker-fts/tracker-parser-test.c | 29 ----------------------------
1 files changed, 0 insertions(+), 29 deletions(-)
---
diff --git a/tests/libtracker-fts/tracker-parser-test.c b/tests/libtracker-fts/tracker-parser-test.c
index dcdfbac..9d3c2c9 100644
--- a/tests/libtracker-fts/tracker-parser-test.c
+++ b/tests/libtracker-fts/tracker-parser-test.c
@@ -26,23 +26,6 @@
#include <libtracker-fts/tracker-parser.h>
-/* Note
- * Currently, three different types of parsers are defined in libtracker-fts:
- * - GNU libunistring-based parser, up to 20% faster than the others, and full
- * unicode compliant.
- * - libicu-based parser, full unicode compliant but slower as it needs
- * conversions to/from UChars (UTF-16 encoded strings)
- * - glib/pango parser, not fully unicode compliant as it doesn't work properly
- * with decomposed strings (NFD normalized), doesn't make a unicode-based
- * word-breaking and doesn't make full-word casefolding.
- *
- * Some of the tests, thus, will be DISABLED for the GLIB/PANGO parser.
- */
-#undef FULL_UNICODE_TESTS
-#if defined HAVE_LIBUNISTRING || defined HAVE_LIBICU
-#define FULL_UNICODE_TESTS
-#endif
-
/* -------------- COMMON FOR ALL TESTS ----------------- */
/* Fixture object type */
@@ -258,11 +241,9 @@ static const TestDataExpectedWord test_data_normalization[] = {
{ "école", "ecole", FALSE, TRUE },
{ "ÉCOLE", "ecole", FALSE, TRUE },
{ "École", "ecole", FALSE, TRUE },
-#ifdef FULL_UNICODE_TESTS /* glib/pango doesn't like NFD strings */
{ "e" "\xCC\x81" "cole", "ecole", FALSE, TRUE },
{ "E" "\xCC\x81" "COLE", "ecole", FALSE, TRUE },
{ "E" "\xCC\x81" "cole", "ecole", FALSE, TRUE },
-#endif
{ NULL, NULL, FALSE, FALSE }
};
@@ -277,7 +258,6 @@ static const TestDataExpectedWord test_data_unaccent[] = {
{ "ὶ", "ι", FALSE, TRUE }, /* greek small iotta with U+0300, composed */
{ "Ὼ", "ω", FALSE, TRUE }, /* greek capital omega with U+0300, composed */
{ "ὼ", "ω", FALSE, TRUE }, /* greek small omega with U+0300, composed */
-#ifdef FULL_UNICODE_TESTS /* glib/pango does not like NFD strings */
{ "Ὰ", "α", FALSE, TRUE }, /* capital alpha with U+0300, decomposed */
{ "ὰ", "α", FALSE, TRUE }, /* small alpha with U+0300, decomposed */
{ "Ὶ", "ι", FALSE, TRUE }, /* capital iotta with U+0300, decomposed */
@@ -286,7 +266,6 @@ static const TestDataExpectedWord test_data_unaccent[] = {
{ "ὼ", "ω", FALSE, TRUE }, /* small omega with U+0300, decomposed */
{ "aN͡Ga", "anga", FALSE, TRUE }, /* 0x0361 affects to two characters */
{ "aNG͡a", "anga", FALSE, TRUE }, /* 0x0361 affects to two characters */
-#endif
{ "Murciélago", "murciélago", FALSE, FALSE },
{ "camión", "camión", FALSE, FALSE },
{ "desagüe", "desagüe", FALSE, FALSE },
@@ -305,32 +284,24 @@ static const TestDataExpectedWord test_data_casefolding[] = {
{ "gross", "gross", FALSE, TRUE },
{ "GROSS", "gross", FALSE, TRUE },
{ "GrOsS", "gross", FALSE, TRUE },
-#ifdef FULL_UNICODE_TESTS /* glib/pango doesn't do full-word casefolding */
{ "groß", "gross", FALSE, TRUE },
-#endif
{ NULL, NULL, FALSE, FALSE }
};
/* Number of expected words tests */
static const TestDataExpectedNWords test_data_nwords[] = {
-#ifdef FULL_UNICODE_TESTS /* glib/pango assumes ' is a word breaker */
{ "The quick (\"brown\") fox can’t jump 32.3 feet, right?", TRUE, 8 },
{ "The quick (\"brown\") fox can’t jump 32.3 feet, right?", FALSE, 10 },
-#endif
/* Note: as of 0.9.15, the dot is always a word breaker, even between
* numbers. */
{ "filename.txt", TRUE, 2 },
{ ".hidden.txt", TRUE, 2 },
{ "noextension.", TRUE, 1 },
{ "ホモ・サピエンス", TRUE, 2 }, /* katakana */
-#ifdef FULL_UNICODE_TESTS /* glib/pango doesn't work properly with chinese */
{ "喂人类", TRUE, 2 }, /* chinese */
-#endif
{ "Американские суда находятся в международных водах.", TRUE, 6 }, /* russian */
-#ifdef FULL_UNICODE_TESTS /* glib/pango doesn't work properly with chinese */
{ "Bần chỉ là một anh nghèo xác", TRUE, 7 }, /* vietnamese */
{ "ホモ・サピエンス 喂人类 katakana, chinese, english", TRUE, 7 }, /* mixed */
-#endif
{ NULL, FALSE, 0 }
};
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]