Index: icu46/source/test/letest/xmlreader.cpp |
=================================================================== |
--- icu46/source/test/letest/xmlreader.cpp (revision 0) |
+++ icu46/source/test/letest/xmlreader.cpp (revision 0) |
@@ -0,0 +1,259 @@ |
+/* |
+ ******************************************************************************* |
+ * |
+ * Copyright (C) 1999-2007, International Business Machines |
+ * Corporation and others. All Rights Reserved. |
+ * |
+ ******************************************************************************* |
+ */ |
+ |
+#include "unicode/utypes.h" |
+#include "unicode/uclean.h" |
+#include "unicode/uchar.h" |
+#include "unicode/unistr.h" |
+#include "unicode/uscript.h" |
+#include "unicode/putil.h" |
+#include "unicode/ctest.h" |
+ |
+#include "layout/LETypes.h" |
+#include "layout/LEScripts.h" |
+ |
+#include "letsutil.h" |
+#include "letest.h" |
+ |
+#include "xmlreader.h" |
+ |
+#include "xmlparser.h" |
+ |
+#include <stdlib.h> |
+#include <stdio.h> |
+#include <string.h> |
+ |
+//U_NAMESPACE_USE |
+ |
+#define CH_COMMA 0x002C |
+ |
+static le_uint32 *getHexArray(const UnicodeString &numbers, int32_t &arraySize) |
+{ |
+ int32_t offset = -1; |
+ |
+ arraySize = 1; |
+ while((offset = numbers.indexOf(CH_COMMA, offset + 1)) >= 0) { |
+ arraySize += 1; |
+ } |
+ |
+ le_uint32 *array = NEW_ARRAY(le_uint32, arraySize); |
+ char number[16]; |
+ le_int32 count = 0; |
+ le_int32 start = 0, end = 0; |
+ le_int32 len = 0; |
+ |
+ // trim leading whitespace |
+ while(u_isUWhiteSpace(numbers[start])) { |
+ start += 1; |
+ } |
+ |
+ while((end = numbers.indexOf(CH_COMMA, start)) >= 0) { |
+ len = numbers.extract(start, end - start, number, ARRAY_SIZE(number), US_INV); |
+ number[len] = '\0'; |
+ start = end + 1; |
+ |
+ sscanf(number, "%x", &array[count++]); |
+ |
+ // trim whitespace following the comma |
+ while(u_isUWhiteSpace(numbers[start])) { |
+ start += 1; |
+ } |
+ } |
+ |
+ // trim trailing whitespace |
+ end = numbers.length(); |
+ while(u_isUWhiteSpace(numbers[end - 1])) { |
+ end -= 1; |
+ } |
+ |
+ len = numbers.extract(start, end - start, number, ARRAY_SIZE(number), US_INV); |
+ number[len] = '\0'; |
+ sscanf(number, "%x", &array[count]); |
+ |
+ return array; |
+} |
+ |
+static float *getFloatArray(const UnicodeString &numbers, int32_t &arraySize) |
+{ |
+ int32_t offset = -1; |
+ |
+ arraySize = 1; |
+ while((offset = numbers.indexOf(CH_COMMA, offset + 1)) >= 0) { |
+ arraySize += 1; |
+ } |
+ |
+ float *array = NEW_ARRAY(float, arraySize); |
+ char number[32]; |
+ le_int32 count = 0; |
+ le_int32 start = 0, end = 0; |
+ le_int32 len = 0; |
+ |
+ // trim leading whitespace |
+ while(u_isUWhiteSpace(numbers[start])) { |
+ start += 1; |
+ } |
+ |
+ while((end = numbers.indexOf(CH_COMMA, start)) >= 0) { |
+ len = numbers.extract(start, end - start, number, ARRAY_SIZE(number), US_INV); |
+ number[len] = '\0'; |
+ start = end + 1; |
+ |
+ sscanf(number, "%f", &array[count++]); |
+ |
+ // trim whiteapce following the comma |
+ while(u_isUWhiteSpace(numbers[start])) { |
+ start += 1; |
+ } |
+ } |
+ |
+ while(u_isUWhiteSpace(numbers[start])) { |
+ start += 1; |
+ } |
+ |
+ // trim trailing whitespace |
+ end = numbers.length(); |
+ while(u_isUWhiteSpace(numbers[end - 1])) { |
+ end -= 1; |
+ } |
+ |
+ len = numbers.extract(start, end - start, number, ARRAY_SIZE(number), US_INV); |
+ number[len] = '\0'; |
+ sscanf(number, "%f", &array[count]); |
+ |
+ return array; |
+} |
+ |
+U_CDECL_BEGIN |
+void readTestFile(const char *testFilePath, TestCaseCallback callback) |
+{ |
+#if !UCONFIG_NO_REGULAR_EXPRESSIONS |
+ UErrorCode status = U_ZERO_ERROR; |
+ UXMLParser *parser = UXMLParser::createParser(status); |
+ UXMLElement *root = parser->parseFile(testFilePath, status); |
+ |
+ if (root == NULL) { |
+ log_err("Could not open the test data file: %s\n", testFilePath); |
+ delete parser; |
+ return; |
+ } |
+ |
+ UnicodeString test_case = UNICODE_STRING_SIMPLE("test-case"); |
+ UnicodeString test_text = UNICODE_STRING_SIMPLE("test-text"); |
+ UnicodeString test_font = UNICODE_STRING_SIMPLE("test-font"); |
+ UnicodeString result_glyphs = UNICODE_STRING_SIMPLE("result-glyphs"); |
+ UnicodeString result_indices = UNICODE_STRING_SIMPLE("result-indices"); |
+ UnicodeString result_positions = UNICODE_STRING_SIMPLE("result-positions"); |
+ |
+ // test-case attributes |
+ UnicodeString id_attr = UNICODE_STRING_SIMPLE("id"); |
+ UnicodeString script_attr = UNICODE_STRING_SIMPLE("script"); |
+ UnicodeString lang_attr = UNICODE_STRING_SIMPLE("lang"); |
+ |
+ // test-font attributes |
+ UnicodeString name_attr = UNICODE_STRING_SIMPLE("name"); |
+ UnicodeString ver_attr = UNICODE_STRING_SIMPLE("version"); |
+ UnicodeString cksum_attr = UNICODE_STRING_SIMPLE("checksum"); |
+ |
+ const UXMLElement *testCase; |
+ int32_t tc = 0; |
+ |
+ while((testCase = root->nextChildElement(tc)) != NULL) { |
+ if (testCase->getTagName().compare(test_case) == 0) { |
+ char *id = getCString(testCase->getAttribute(id_attr)); |
+ char *script = getCString(testCase->getAttribute(script_attr)); |
+ char *lang = getCString(testCase->getAttribute(lang_attr)); |
+ char *fontName = NULL; |
+ char *fontVer = NULL; |
+ char *fontCksum = NULL; |
+ const UXMLElement *element; |
+ int32_t ec = 0; |
+ int32_t charCount = 0; |
+ int32_t typoFlags = 3; // kerning + ligatures... |
+ UScriptCode scriptCode; |
+ le_int32 languageCode = -1; |
+ UnicodeString text, glyphs, indices, positions; |
+ int32_t glyphCount = 0, indexCount = 0, positionCount = 0; |
+ TestResult expected = {0, NULL, NULL, NULL}; |
+ |
+ uscript_getCode(script, &scriptCode, 1, &status); |
+ if (LE_FAILURE(status)) { |
+ log_err("invalid script name: %s.\n", script); |
+ goto free_c_strings; |
+ } |
+ |
+ if (lang != NULL) { |
+ languageCode = getLanguageCode(lang); |
+ |
+ if (languageCode < 0) { |
+ log_err("invalid language name: %s.\n", lang); |
+ goto free_c_strings; |
+ } |
+ } |
+ |
+ while((element = testCase->nextChildElement(ec)) != NULL) { |
+ UnicodeString tag = element->getTagName(); |
+ |
+ // TODO: make sure that each element is only used once. |
+ if (tag.compare(test_font) == 0) { |
+ fontName = getCString(element->getAttribute(name_attr)); |
+ fontVer = getCString(element->getAttribute(ver_attr)); |
+ fontCksum = getCString(element->getAttribute(cksum_attr)); |
+ |
+ } else if (tag.compare(test_text) == 0) { |
+ text = element->getText(TRUE); |
+ charCount = text.length(); |
+ } else if (tag.compare(result_glyphs) == 0) { |
+ glyphs = element->getText(TRUE); |
+ } else if (tag.compare(result_indices) == 0) { |
+ indices = element->getText(TRUE); |
+ } else if (tag.compare(result_positions) == 0) { |
+ positions = element->getText(TRUE); |
+ } else { |
+ // an unknown tag... |
+ char *cTag = getCString(&tag); |
+ |
+ log_info("Test %s: unknown element with tag \"%s\"\n", id, cTag); |
+ freeCString(cTag); |
+ } |
+ } |
+ |
+ expected.glyphs = (LEGlyphID *) getHexArray(glyphs, glyphCount); |
+ expected.indices = (le_int32 *) getHexArray(indices, indexCount); |
+ expected.positions = getFloatArray(positions, positionCount); |
+ |
+ expected.glyphCount = glyphCount; |
+ |
+ if (glyphCount < charCount || indexCount != glyphCount || positionCount < glyphCount * 2 + 2) { |
+ log_err("Test %s: inconsistent input data: charCount = %d, glyphCount = %d, indexCount = %d, positionCount = %d\n", |
+ id, charCount, glyphCount, indexCount, positionCount); |
+ goto free_expected; |
+ }; |
+ |
+ (*callback)(id, fontName, fontVer, fontCksum, scriptCode, languageCode, text.getBuffer(), charCount, &expected); |
+ |
+free_expected: |
+ DELETE_ARRAY(expected.positions); |
+ DELETE_ARRAY(expected.indices); |
+ DELETE_ARRAY(expected.glyphs); |
+ |
+free_c_strings: |
+ freeCString(fontCksum); |
+ freeCString(fontVer); |
+ freeCString(fontName); |
+ freeCString(lang); |
+ freeCString(script); |
+ freeCString(id); |
+ } |
+ } |
+ |
+ delete root; |
+ delete parser; |
+#endif |
+} |
+U_CDECL_END |
Property changes on: icu46/source/test/letest/xmlreader.cpp |
___________________________________________________________________ |
Added: svn:eol-style |
+ LF |