blob: cc8d89a942669850933d40110471626a38f55dc3 [file] [log] [blame]
// © 2016 and later: Unicode, Inc. and others.
// License & terms of use: http://www.unicode.org/copyright.html
/*
*******************************************************************************
*
* Copyright (C) 2005-2014, International Business Machines
* Corporation and others. All Rights Reserved.
*
*******************************************************************************
*
* created on: 2005jun15
* created by: Raymond Yang
*/
#include "unicode/utypes.h"
#if !UCONFIG_NO_IDNA
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "unicode/ucnv.h"
#include "unicode/ustring.h"
#include "unicode/uidna.h"
#include "unicode/utf16.h"
#include "idnaconf.h"
static const UChar C_TAG[] = {0x3D, 0x3D, 0x3D, 0x3D, 0x3D, 0}; // =====
static const UChar C_NAMEZONE[] = {0x6E, 0x61, 0x6D, 0x65, 0x7A, 0x6F, 0x6E, 0x65, 0}; // namezone
static const UChar C_NAMEBASE[] = {0x6E, 0x61, 0x6D, 0x65, 0x62, 0x61, 0x73, 0x65, 0}; // namebase
static const UChar C_TYPE[] = {0x74, 0x79, 0x70, 0x65, 0}; // type
static const UChar C_TOASCII[] = {0x74, 0x6F, 0x61, 0x73, 0x63, 0x69, 0x69, 0}; // toascii
static const UChar C_TOUNICODE[] = {0x74, 0x6F, 0x75, 0x6E, 0x69, 0x63, 0x6F, 0x64, 0x65, 0}; // tounicode
static const UChar C_PASSFAIL[] = {0x70, 0x61, 0x73, 0x73, 0x66, 0x61, 0x69, 0x6C, 0}; // passfail
static const UChar C_PASS[] = {0x70, 0x61, 0x73, 0x73, 0}; // pass
static const UChar C_FAIL[] = {0x66, 0x61, 0x69, 0x6C, 0}; // fail
static const UChar C_DESC[] = {0x64, 0x65, 0x73, 0x63, 0}; // desc
static const UChar C_USESTD3ASCIIRULES[] = {0x55, 0x73, 0x65, 0x53, 0x54, 0x44,
0x33, 0x41, 0x53, 0x43, 0x49, 0x49, 0x52, 0x75, 0x6C, 0x65, 0x73, 0}; // UseSTD3ASCIIRules
IdnaConfTest::IdnaConfTest(){
base = NULL;
len = 0;
curOffset = 0;
type = option = passfail = -1;
namebase.setToBogus();
namezone.setToBogus();
}
IdnaConfTest::~IdnaConfTest(){
delete [] base;
}
#if !UCONFIG_NO_IDNA
/* this function is modified from RBBITest::ReadAndConvertFile()
*
*/
UBool IdnaConfTest::ReadAndConvertFile(){
char * source = NULL;
size_t source_len;
// read the test data file to memory
FILE* f = NULL;
UErrorCode status = U_ZERO_ERROR;
const char *path = IntlTest::getSourceTestData(status);
if (U_FAILURE(status)) {
errln("%s", u_errorName(status));
return FALSE;
}
const char* name = "idna_conf.txt"; // test data file
int t = static_cast<int>(strlen(path) + strlen(name) + 1);
char* absolute_name = new char[t];
strcpy(absolute_name, path);
strcat(absolute_name, name);
f = fopen(absolute_name, "rb");
delete [] absolute_name;
if (f == NULL){
dataerrln("fopen error on %s", name);
return FALSE;
}
fseek( f, 0, SEEK_END);
if ((source_len = ftell(f)) <= 0){
errln("Error reading test data file.");
fclose(f);
return FALSE;
}
source = new char[source_len];
fseek(f, 0, SEEK_SET);
if (fread(source, 1, source_len, f) != source_len) {
errln("Error reading test data file.");
delete [] source;
fclose(f);
return FALSE;
}
fclose(f);
// convert the UTF-8 encoded stream to UTF-16 stream
UConverter* conv = ucnv_open("utf-8", &status);
int dest_len = ucnv_toUChars(conv,
NULL, // dest,
0, // destCapacity,
source,
static_cast<int32_t>(source_len),
&status);
if (status == U_BUFFER_OVERFLOW_ERROR) {
// Buffer Overflow is expected from the preflight operation.
status = U_ZERO_ERROR;
UChar * dest = NULL;
dest = new UChar[ dest_len + 1];
ucnv_toUChars(conv, dest, dest_len + 1, source, static_cast<int32_t>(source_len), &status);
// Do not know the "if possible" behavior of ucnv_toUChars()
// Do it by ourself.
dest[dest_len] = 0;
len = dest_len;
base = dest;
delete [] source;
ucnv_close(conv);
return TRUE; // The buffer will owned by caller.
}
errln("UConverter error: %s", u_errorName(status));
delete [] source;
ucnv_close(conv);
return FALSE;
}
int IdnaConfTest::isNewlineMark(){
static const UChar LF = 0x0a;
static const UChar CR = 0x0d;
UChar c = base[curOffset];
// CR LF
if ( c == CR && curOffset + 1 < len && base[curOffset + 1] == LF){
return 2;
}
// CR or LF
if ( c == CR || c == LF) {
return 1;
}
return 0;
}
/* Read a logical line.
*
* All lines ending in a backslash (\) and immediately followed by a newline
* character are joined with the next line in the source file forming logical
* lines from the physical lines.
*
*/
UBool IdnaConfTest::ReadOneLine(UnicodeString& buf){
if ( !(curOffset < len) ) return FALSE; // stream end
static const UChar BACKSLASH = 0x5c;
buf.remove();
int t = 0;
while (curOffset < len){
if ((t = isNewlineMark())) { // end of line
curOffset += t;
break;
}
UChar c = base[curOffset];
if (c == BACKSLASH && curOffset < len -1){ // escaped new line mark
if ((t = isNewlineMark())){
curOffset += 1 + t; // BACKSLAH and NewlineMark
continue;
}
}
buf.append(c);
curOffset++;
}
return TRUE;
}
//
//===============================================================
//
/* Explain <xxxxx> tag to a native value
*
* Since <xxxxx> is always larger than the native value,
* the operation will replace the tag directly in the buffer,
* and, of course, will shift tail elements.
*/
void IdnaConfTest::ExplainCodePointTag(UnicodeString& buf){
buf.append((UChar)0); // add a terminal NULL
UChar* bufBase = buf.getBuffer(buf.length());
UChar* p = bufBase;
while (*p != 0){
if ( *p != 0x3C){ // <
*bufBase++ = *p++;
} else {
p++; // skip <
UChar32 cp = 0;
for ( ;*p != 0x3E; p++){ // >
if (0x30 <= *p && *p <= 0x39){ // 0-9
cp = (cp * 16) + (*p - 0x30);
} else if (0x61 <= *p && *p <= 0x66){ // a-f
cp = (cp * 16) + (*p - 0x61) + 10;
} else if (0x41 <= *p && *p <= 0x46) {// A-F
cp = (cp * 16) + (*p - 0x41) + 10;
}
// no else. hope everything is good.
}
p++; // skip >
if (U_IS_BMP(cp)){
*bufBase++ = cp;
} else {
*bufBase++ = U16_LEAD(cp);
*bufBase++ = U16_TRAIL(cp);
}
}
}
*bufBase = 0; // close our buffer
buf.releaseBuffer();
}
void IdnaConfTest::Call(){
if (type == -1 || option == -1 || passfail == -1 || namebase.isBogus() || namezone.isBogus()){
errln("Incomplete record");
} else {
UErrorCode status = U_ZERO_ERROR;
UChar result[200] = {0,}; // simple life
const UChar *p = namebase.getTerminatedBuffer();
const int p_len = namebase.length();
if (type == 0 && option == 0){
uidna_IDNToASCII(p, p_len, result, 200, UIDNA_USE_STD3_RULES, NULL, &status);
} else if (type == 0 && option == 1){
uidna_IDNToASCII(p, p_len, result, 200, UIDNA_ALLOW_UNASSIGNED, NULL, &status);
} else if (type == 1 && option == 0){
uidna_IDNToUnicode(p, p_len, result, 200, UIDNA_USE_STD3_RULES, NULL, &status);
} else if (type == 1 && option == 1){
uidna_IDNToUnicode(p, p_len, result, 200, UIDNA_ALLOW_UNASSIGNED, NULL, &status);
}
if (passfail == 0){
if (U_FAILURE(status)){
id.append(" should pass, but failed. - ");
id.append(u_errorName(status));
errcheckln(status, id);
} else{
if (namezone.compare(result, -1) == 0){
// expected
logln(UnicodeString("namebase: ") + prettify(namebase) + UnicodeString(" result: ") + prettify(result));
} else {
id.append(" no error, but result is not as expected.");
errln(id);
}
}
} else if (passfail == 1){
if (U_FAILURE(status)){
// expected
// TODO: Uncomment this when U_IDNA_ZERO_LENGTH_LABEL_ERROR is added to u_errorName
//logln("Got the expected error: " + UnicodeString(u_errorName(status)));
} else{
if (namebase.compare(result, -1) == 0){
// garbage in -> garbage out
logln(UnicodeString("ICU will not recognize malformed ACE-Prefixes or incorrect ACE-Prefixes. ") + UnicodeString("namebase: ") + prettify(namebase) + UnicodeString(" result: ") + prettify(result));
} else {
id.append(" should fail, but not failed. ");
id.append(u_errorName(status));
errln(id);
}
}
}
}
type = option = passfail = -1;
namebase.setToBogus();
namezone.setToBogus();
id.remove();
return;
}
void IdnaConfTest::Test(void){
if (!ReadAndConvertFile())return;
UnicodeString s;
UnicodeString key;
UnicodeString value;
// skip everything before the first "=====" and "=====" itself
do {
if (!ReadOneLine(s)) {
errln("End of file prematurely found");
break;
}
}
while (s.compare(C_TAG, -1) != 0); //"====="
while(ReadOneLine(s)){
s.trim();
key.remove();
value.remove();
if (s.compare(C_TAG, -1) == 0){ //"====="
Call();
} else {
// explain key:value
int p = s.indexOf((UChar)0x3A); // :
key.setTo(s,0,p).trim();
value.setTo(s,p+1).trim();
if (key.compare(C_TYPE, -1) == 0){
if (value.compare(C_TOASCII, -1) == 0) {
type = 0;
} else if (value.compare(C_TOUNICODE, -1) == 0){
type = 1;
}
} else if (key.compare(C_PASSFAIL, -1) == 0){
if (value.compare(C_PASS, -1) == 0){
passfail = 0;
} else if (value.compare(C_FAIL, -1) == 0){
passfail = 1;
}
} else if (key.compare(C_DESC, -1) == 0){
if (value.indexOf(C_USESTD3ASCIIRULES, u_strlen(C_USESTD3ASCIIRULES), 0) == -1){
option = 1; // not found
} else {
option = 0;
}
id.setTo(value, 0, value.indexOf((UChar)0x20)); // space
} else if (key.compare(C_NAMEZONE, -1) == 0){
ExplainCodePointTag(value);
namezone.setTo(value);
} else if (key.compare(C_NAMEBASE, -1) == 0){
ExplainCodePointTag(value);
namebase.setTo(value);
}
// just skip other lines
}
}
Call(); // for last record
}
#else
void IdnaConfTest::Test(void)
{
// test nothing...
}
#endif
void IdnaConfTest::runIndexedTest( int32_t index, UBool exec, const char* &name, char* /*par*/){
switch (index) {
TESTCASE(0,Test);
default: name = ""; break;
}
}
#endif