123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922 |
- // © 2016 and later: Unicode, Inc. and others.
- // License & terms of use: http://www.unicode.org/copyright.html
- /*
- *******************************************************************************
- *
- * Copyright (C) 2003-2014, International Business Machines
- * Corporation and others. All Rights Reserved.
- *
- *******************************************************************************
- * file name: uidna.cpp
- * encoding: UTF-8
- * tab size: 8 (not used)
- * indentation:4
- *
- * created on: 2003feb1
- * created by: Ram Viswanadha
- */
- #include "unicode/utypes.h"
- #if !UCONFIG_NO_IDNA
- #include "unicode/uidna.h"
- #include "unicode/ustring.h"
- #include "unicode/usprep.h"
- #include "punycode.h"
- #include "ustr_imp.h"
- #include "cmemory.h"
- #include "uassert.h"
- #include "sprpimpl.h"
- /* it is official IDNA ACE Prefix is "xn--" */
- static const char16_t ACE_PREFIX[] ={ 0x0078,0x006E,0x002d,0x002d } ;
- #define ACE_PREFIX_LENGTH 4
- #define MAX_LABEL_LENGTH 63
- /* The Max length of the labels should not be more than MAX_LABEL_LENGTH */
- #define MAX_LABEL_BUFFER_SIZE 100
- #define MAX_DOMAIN_NAME_LENGTH 255
- /* The Max length of the domain names should not be more than MAX_DOMAIN_NAME_LENGTH */
- #define MAX_IDN_BUFFER_SIZE MAX_DOMAIN_NAME_LENGTH+1
- #define LOWER_CASE_DELTA 0x0020
- #define HYPHEN 0x002D
- #define FULL_STOP 0x002E
- #define CAPITAL_A 0x0041
- #define CAPITAL_Z 0x005A
- inline static char16_t
- toASCIILower(char16_t ch){
- if(CAPITAL_A <= ch && ch <= CAPITAL_Z){
- return ch + LOWER_CASE_DELTA;
- }
- return ch;
- }
- inline static UBool
- startsWithPrefix(const char16_t* src , int32_t srcLength){
- if(srcLength < ACE_PREFIX_LENGTH){
- return false;
- }
- for(int8_t i=0; i< ACE_PREFIX_LENGTH; i++){
- if(toASCIILower(src[i]) != ACE_PREFIX[i]){
- return false;
- }
- }
- return true;
- }
- inline static int32_t
- compareCaseInsensitiveASCII(const char16_t* s1, int32_t s1Len,
- const char16_t* s2, int32_t s2Len){
-
- int32_t minLength;
- int32_t lengthResult;
- // are we comparing different lengths?
- if(s1Len != s2Len) {
- if(s1Len < s2Len) {
- minLength = s1Len;
- lengthResult = -1;
- } else {
- minLength = s2Len;
- lengthResult = 1;
- }
- } else {
- // ok the lengths are equal
- minLength = s1Len;
- lengthResult = 0;
- }
- char16_t c1,c2;
- int32_t rc;
- for(int32_t i =0;/* no condition */;i++) {
- /* If we reach the ends of both strings then they match */
- if(i == minLength) {
- return lengthResult;
- }
-
- c1 = s1[i];
- c2 = s2[i];
-
- /* Case-insensitive comparison */
- if(c1!=c2) {
- rc=(int32_t)toASCIILower(c1)-(int32_t)toASCIILower(c2);
- if(rc!=0) {
- lengthResult=rc;
- break;
- }
- }
- }
- return lengthResult;
- }
- /**
- * Ascertain if the given code point is a label separator as
- * defined by the IDNA RFC
- *
- * @param ch The code point to be ascertained
- * @return true if the char is a label separator
- * @stable ICU 2.8
- */
- static inline UBool isLabelSeparator(char16_t ch){
- switch(ch){
- case 0x002e:
- case 0x3002:
- case 0xFF0E:
- case 0xFF61:
- return true;
- default:
- return false;
- }
- }
- // returns the length of the label excluding the separator
- // if *limit == separator then the length returned does not include
- // the separtor.
- static inline int32_t
- getNextSeparator(char16_t *src, int32_t srcLength,
- char16_t **limit, UBool *done){
- if(srcLength == -1){
- int32_t i;
- for(i=0 ; ;i++){
- if(src[i] == 0){
- *limit = src + i; // point to null
- *done = true;
- return i;
- }
- if(isLabelSeparator(src[i])){
- *limit = src + (i+1); // go past the delimiter
- return i;
-
- }
- }
- }else{
- int32_t i;
- for(i=0;i<srcLength;i++){
- if(isLabelSeparator(src[i])){
- *limit = src + (i+1); // go past the delimiter
- return i;
- }
- }
- // we have not found the delimiter
- // if(i==srcLength)
- *limit = src+srcLength;
- *done = true;
- return i;
- }
- }
- static inline UBool isLDHChar(char16_t ch){
- // high runner case
- if(ch>0x007A){
- return false;
- }
- //[\\u002D \\u0030-\\u0039 \\u0041-\\u005A \\u0061-\\u007A]
- if( (ch==0x002D) ||
- (0x0030 <= ch && ch <= 0x0039) ||
- (0x0041 <= ch && ch <= 0x005A) ||
- (0x0061 <= ch && ch <= 0x007A)
- ){
- return true;
- }
- return false;
- }
- static int32_t
- _internal_toASCII(const char16_t* src, int32_t srcLength,
- char16_t* dest, int32_t destCapacity,
- int32_t options,
- UStringPrepProfile* nameprep,
- UParseError* parseError,
- UErrorCode* status)
- {
- // TODO Revisit buffer handling. The label should not be over 63 ASCII characters. ICU4J may need to be updated too.
- char16_t b1Stack[MAX_LABEL_BUFFER_SIZE], b2Stack[MAX_LABEL_BUFFER_SIZE];
- //initialize pointers to stack buffers
- char16_t *b1 = b1Stack, *b2 = b2Stack;
- int32_t b1Len=0, b2Len,
- b1Capacity = MAX_LABEL_BUFFER_SIZE,
- b2Capacity = MAX_LABEL_BUFFER_SIZE ,
- reqLength=0;
- int32_t namePrepOptions = ((options & UIDNA_ALLOW_UNASSIGNED) != 0) ? USPREP_ALLOW_UNASSIGNED: 0;
- UBool* caseFlags = nullptr;
-
- // the source contains all ascii codepoints
- UBool srcIsASCII = true;
- // assume the source contains all LDH codepoints
- UBool srcIsLDH = true;
- int32_t j=0;
- //get the options
- UBool useSTD3ASCIIRules = (UBool)((options & UIDNA_USE_STD3_RULES) != 0);
- int32_t failPos = -1;
-
- if(srcLength == -1){
- srcLength = u_strlen(src);
- }
-
- if(srcLength > b1Capacity){
- b1 = (char16_t*) uprv_malloc(srcLength * U_SIZEOF_UCHAR);
- if(b1==nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- b1Capacity = srcLength;
- }
- // step 1
- for( j=0;j<srcLength;j++){
- if(src[j] > 0x7F){
- srcIsASCII = false;
- }
- b1[b1Len++] = src[j];
- }
-
- // step 2 is performed only if the source contains non ASCII
- if(srcIsASCII == false){
-
- // step 2
- b1Len = usprep_prepare(nameprep, src, srcLength, b1, b1Capacity, namePrepOptions, parseError, status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- // redo processing of string
- // we do not have enough room so grow the buffer
- if(b1 != b1Stack){
- uprv_free(b1);
- }
- b1 = (char16_t*) uprv_malloc(b1Len * U_SIZEOF_UCHAR);
- if(b1==nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- *status = U_ZERO_ERROR; // reset error
-
- b1Len = usprep_prepare(nameprep, src, srcLength, b1, b1Len, namePrepOptions, parseError, status);
- }
- }
- // error bail out
- if(U_FAILURE(*status)){
- goto CLEANUP;
- }
- if(b1Len == 0){
- *status = U_IDNA_ZERO_LENGTH_LABEL_ERROR;
- goto CLEANUP;
- }
- // for step 3 & 4
- srcIsASCII = true;
- for( j=0;j<b1Len;j++){
- // check if output of usprep_prepare is all ASCII
- if(b1[j] > 0x7F){
- srcIsASCII = false;
- }else if(isLDHChar(b1[j])==false){ // if the char is in ASCII range verify that it is an LDH character
- srcIsLDH = false;
- failPos = j;
- }
- }
- if(useSTD3ASCIIRules){
- // verify 3a and 3b
- // 3(a) Verify the absence of non-LDH ASCII code points; that is, the
- // absence of 0..2C, 2E..2F, 3A..40, 5B..60, and 7B..7F.
- // 3(b) Verify the absence of leading and trailing hyphen-minus; that
- // is, the absence of U+002D at the beginning and end of the
- // sequence.
- if( srcIsLDH == false /* source at this point should not contain anyLDH characters */
- || b1[0] == HYPHEN || b1[b1Len-1] == HYPHEN){
- *status = U_IDNA_STD3_ASCII_RULES_ERROR;
- /* populate the parseError struct */
- if(srcIsLDH==false){
- // failPos is always set the index of failure
- uprv_syntaxError(b1,failPos, b1Len,parseError);
- }else if(b1[0] == HYPHEN){
- // fail position is 0
- uprv_syntaxError(b1,0,b1Len,parseError);
- }else{
- // the last index in the source is always length-1
- uprv_syntaxError(b1, (b1Len>0) ? b1Len-1 : b1Len, b1Len,parseError);
- }
- goto CLEANUP;
- }
- }
- // Step 4: if the source is ASCII then proceed to step 8
- if(srcIsASCII){
- if(b1Len <= destCapacity){
- u_memmove(dest, b1, b1Len);
- reqLength = b1Len;
- }else{
- reqLength = b1Len;
- goto CLEANUP;
- }
- }else{
- // step 5 : verify the sequence does not begin with ACE prefix
- if(!startsWithPrefix(b1,b1Len)){
- //step 6: encode the sequence with punycode
- // do not preserve the case flags for now!
- // TODO: Preserve the case while implementing the RFE
- // caseFlags = (UBool*) uprv_malloc(b1Len * sizeof(UBool));
- // uprv_memset(caseFlags,true,b1Len);
- b2Len = u_strToPunycode(b1,b1Len,b2,b2Capacity,caseFlags, status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- // redo processing of string
- /* we do not have enough room so grow the buffer*/
- b2 = (char16_t*) uprv_malloc(b2Len * U_SIZEOF_UCHAR);
- if(b2 == nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- *status = U_ZERO_ERROR; // reset error
-
- b2Len = u_strToPunycode(b1,b1Len,b2,b2Len,caseFlags, status);
- }
- //error bail out
- if(U_FAILURE(*status)){
- goto CLEANUP;
- }
- // TODO : Reconsider while implementing the case preserve RFE
- // convert all codepoints to lower case ASCII
- // toASCIILower(b2,b2Len);
- reqLength = b2Len+ACE_PREFIX_LENGTH;
- if(reqLength > destCapacity){
- *status = U_BUFFER_OVERFLOW_ERROR;
- goto CLEANUP;
- }
- //Step 7: prepend the ACE prefix
- u_memcpy(dest, ACE_PREFIX, ACE_PREFIX_LENGTH);
- //Step 6: copy the contents in b2 into dest
- u_memcpy(dest+ACE_PREFIX_LENGTH, b2, b2Len);
- }else{
- *status = U_IDNA_ACE_PREFIX_ERROR;
- //position of failure is 0
- uprv_syntaxError(b1,0,b1Len,parseError);
- goto CLEANUP;
- }
- }
- // step 8: verify the length of label
- if(reqLength > MAX_LABEL_LENGTH){
- *status = U_IDNA_LABEL_TOO_LONG_ERROR;
- }
- CLEANUP:
- if(b1 != b1Stack){
- uprv_free(b1);
- }
- if(b2 != b2Stack){
- uprv_free(b2);
- }
- uprv_free(caseFlags);
-
- return u_terminateUChars(dest, destCapacity, reqLength, status);
- }
- static int32_t
- _internal_toUnicode(const char16_t* src, int32_t srcLength,
- char16_t* dest, int32_t destCapacity,
- int32_t options,
- UStringPrepProfile* nameprep,
- UParseError* parseError,
- UErrorCode* status)
- {
- //get the options
- //UBool useSTD3ASCIIRules = (UBool)((options & UIDNA_USE_STD3_RULES) != 0);
- int32_t namePrepOptions = ((options & UIDNA_ALLOW_UNASSIGNED) != 0) ? USPREP_ALLOW_UNASSIGNED: 0;
- // TODO Revisit buffer handling. The label should not be over 63 ASCII characters. ICU4J may need to be updated too.
- char16_t b1Stack[MAX_LABEL_BUFFER_SIZE], b2Stack[MAX_LABEL_BUFFER_SIZE], b3Stack[MAX_LABEL_BUFFER_SIZE];
- //initialize pointers to stack buffers
- char16_t *b1 = b1Stack, *b2 = b2Stack, *b1Prime=nullptr, *b3=b3Stack;
- int32_t b1Len = 0, b2Len, b1PrimeLen, b3Len,
- b1Capacity = MAX_LABEL_BUFFER_SIZE,
- b2Capacity = MAX_LABEL_BUFFER_SIZE,
- b3Capacity = MAX_LABEL_BUFFER_SIZE,
- reqLength=0;
- UBool* caseFlags = nullptr;
- UBool srcIsASCII = true;
- /*UBool srcIsLDH = true;
- int32_t failPos =0;*/
- // step 1: find out if all the codepoints in src are ASCII
- if(srcLength==-1){
- srcLength = 0;
- for(;src[srcLength]!=0;){
- if(src[srcLength]> 0x7f){
- srcIsASCII = false;
- }/*else if(isLDHChar(src[srcLength])==false){
- // here we do not assemble surrogates
- // since we know that LDH code points
- // are in the ASCII range only
- srcIsLDH = false;
- failPos = srcLength;
- }*/
- srcLength++;
- }
- }else if(srcLength > 0){
- for(int32_t j=0; j<srcLength; j++){
- if(src[j]> 0x7f){
- srcIsASCII = false;
- break;
- }/*else if(isLDHChar(src[j])==false){
- // here we do not assemble surrogates
- // since we know that LDH code points
- // are in the ASCII range only
- srcIsLDH = false;
- failPos = j;
- }*/
- }
- }else{
- return 0;
- }
-
- if(srcIsASCII == false){
- // step 2: process the string
- b1Len = usprep_prepare(nameprep, src, srcLength, b1, b1Capacity, namePrepOptions, parseError, status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- // redo processing of string
- /* we do not have enough room so grow the buffer*/
- b1 = (char16_t*) uprv_malloc(b1Len * U_SIZEOF_UCHAR);
- if(b1==nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- *status = U_ZERO_ERROR; // reset error
-
- b1Len = usprep_prepare(nameprep, src, srcLength, b1, b1Len, namePrepOptions, parseError, status);
- }
- //bail out on error
- if(U_FAILURE(*status)){
- goto CLEANUP;
- }
- }else{
- //just point src to b1
- b1 = (char16_t*) src;
- b1Len = srcLength;
- }
- // The RFC states that
- // <quote>
- // ToUnicode never fails. If any step fails, then the original input
- // is returned immediately in that step.
- // </quote>
- //step 3: verify ACE Prefix
- if(startsWithPrefix(b1,b1Len)){
- //step 4: Remove the ACE Prefix
- b1Prime = b1 + ACE_PREFIX_LENGTH;
- b1PrimeLen = b1Len - ACE_PREFIX_LENGTH;
- //step 5: Decode using punycode
- b2Len = u_strFromPunycode(b1Prime, b1PrimeLen, b2, b2Capacity, caseFlags,status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- // redo processing of string
- /* we do not have enough room so grow the buffer*/
- b2 = (char16_t*) uprv_malloc(b2Len * U_SIZEOF_UCHAR);
- if(b2==nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- *status = U_ZERO_ERROR; // reset error
- b2Len = u_strFromPunycode(b1Prime, b1PrimeLen, b2, b2Len, caseFlags, status);
- }
- //step 6:Apply toASCII
- b3Len = uidna_toASCII(b2, b2Len, b3, b3Capacity, options, parseError, status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- // redo processing of string
- /* we do not have enough room so grow the buffer*/
- b3 = (char16_t*) uprv_malloc(b3Len * U_SIZEOF_UCHAR);
- if(b3==nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- *status = U_ZERO_ERROR; // reset error
- b3Len = uidna_toASCII(b2,b2Len,b3,b3Len,options,parseError, status);
- }
- //bail out on error
- if(U_FAILURE(*status)){
- goto CLEANUP;
- }
- //step 7: verify
- if(compareCaseInsensitiveASCII(b1, b1Len, b3, b3Len) !=0){
- // Cause the original to be returned.
- *status = U_IDNA_VERIFICATION_ERROR;
- goto CLEANUP;
- }
- //step 8: return output of step 5
- reqLength = b2Len;
- if(b2Len <= destCapacity) {
- u_memmove(dest, b2, b2Len);
- }
- }
- else{
- // See the start of this if statement for why this is commented out.
- // verify that STD3 ASCII rules are satisfied
- /*if(useSTD3ASCIIRules == true){
- if( srcIsLDH == false // source contains some non-LDH characters
- || src[0] == HYPHEN || src[srcLength-1] == HYPHEN){
- *status = U_IDNA_STD3_ASCII_RULES_ERROR;
- // populate the parseError struct
- if(srcIsLDH==false){
- // failPos is always set the index of failure
- uprv_syntaxError(src,failPos, srcLength,parseError);
- }else if(src[0] == HYPHEN){
- // fail position is 0
- uprv_syntaxError(src,0,srcLength,parseError);
- }else{
- // the last index in the source is always length-1
- uprv_syntaxError(src, (srcLength>0) ? srcLength-1 : srcLength, srcLength,parseError);
- }
- goto CLEANUP;
- }
- }*/
- // just return the source
- //copy the source to destination
- if(srcLength <= destCapacity){
- u_memmove(dest, src, srcLength);
- }
- reqLength = srcLength;
- }
- CLEANUP:
- if(b1 != b1Stack && b1!=src){
- uprv_free(b1);
- }
- if(b2 != b2Stack){
- uprv_free(b2);
- }
- uprv_free(caseFlags);
- // The RFC states that
- // <quote>
- // ToUnicode never fails. If any step fails, then the original input
- // is returned immediately in that step.
- // </quote>
- // So if any step fails lets copy source to destination
- if(U_FAILURE(*status)){
- //copy the source to destination
- if(dest && srcLength <= destCapacity){
- // srcLength should have already been set earlier.
- U_ASSERT(srcLength >= 0);
- u_memmove(dest, src, srcLength);
- }
- reqLength = srcLength;
- *status = U_ZERO_ERROR;
- }
- return u_terminateUChars(dest, destCapacity, reqLength, status);
- }
- U_CAPI int32_t U_EXPORT2
- uidna_toASCII(const char16_t* src, int32_t srcLength,
- char16_t* dest, int32_t destCapacity,
- int32_t options,
- UParseError* parseError,
- UErrorCode* status){
-
- if(status == nullptr || U_FAILURE(*status)){
- return 0;
- }
- if((src==nullptr) || (srcLength < -1) || (destCapacity<0) || (!dest && destCapacity > 0)){
- *status = U_ILLEGAL_ARGUMENT_ERROR;
- return 0;
- }
- UStringPrepProfile* nameprep = usprep_openByType(USPREP_RFC3491_NAMEPREP, status);
-
- if(U_FAILURE(*status)){
- return -1;
- }
-
- int32_t retLen = _internal_toASCII(src, srcLength, dest, destCapacity, options, nameprep, parseError, status);
-
- /* close the profile*/
- usprep_close(nameprep);
-
- return retLen;
- }
- U_CAPI int32_t U_EXPORT2
- uidna_toUnicode(const char16_t* src, int32_t srcLength,
- char16_t* dest, int32_t destCapacity,
- int32_t options,
- UParseError* parseError,
- UErrorCode* status){
- if(status == nullptr || U_FAILURE(*status)){
- return 0;
- }
- if( (src==nullptr) || (srcLength < -1) || (destCapacity<0) || (!dest && destCapacity > 0)){
- *status = U_ILLEGAL_ARGUMENT_ERROR;
- return 0;
- }
- UStringPrepProfile* nameprep = usprep_openByType(USPREP_RFC3491_NAMEPREP, status);
-
- if(U_FAILURE(*status)){
- return -1;
- }
-
- int32_t retLen = _internal_toUnicode(src, srcLength, dest, destCapacity, options, nameprep, parseError, status);
- usprep_close(nameprep);
-
- return retLen;
- }
- U_CAPI int32_t U_EXPORT2
- uidna_IDNToASCII( const char16_t *src, int32_t srcLength,
- char16_t* dest, int32_t destCapacity,
- int32_t options,
- UParseError *parseError,
- UErrorCode *status){
- if(status == nullptr || U_FAILURE(*status)){
- return 0;
- }
- if((src==nullptr) || (srcLength < -1) || (destCapacity<0) || (!dest && destCapacity > 0)){
- *status = U_ILLEGAL_ARGUMENT_ERROR;
- return 0;
- }
- int32_t reqLength = 0;
- UStringPrepProfile* nameprep = usprep_openByType(USPREP_RFC3491_NAMEPREP, status);
-
- if(U_FAILURE(*status)){
- return 0;
- }
- //initialize pointers
- char16_t *delimiter = (char16_t*)src;
- char16_t *labelStart = (char16_t*)src;
- char16_t *currentDest = (char16_t*) dest;
- int32_t remainingLen = srcLength;
- int32_t remainingDestCapacity = destCapacity;
- int32_t labelLen = 0, labelReqLength = 0;
- UBool done = false;
- for(;;){
- labelLen = getNextSeparator(labelStart,remainingLen, &delimiter,&done);
- labelReqLength = 0;
- if(!(labelLen==0 && done)){// make sure this is not a root label separator.
-
- labelReqLength = _internal_toASCII( labelStart, labelLen,
- currentDest, remainingDestCapacity,
- options, nameprep,
- parseError, status);
-
- if(*status == U_BUFFER_OVERFLOW_ERROR){
-
- *status = U_ZERO_ERROR; // reset error
- remainingDestCapacity = 0;
- }
- }
-
- if(U_FAILURE(*status)){
- break;
- }
-
- reqLength +=labelReqLength;
- // adjust the destination pointer
- if(labelReqLength < remainingDestCapacity){
- currentDest = currentDest + labelReqLength;
- remainingDestCapacity -= labelReqLength;
- }else{
- // should never occur
- remainingDestCapacity = 0;
- }
- if(done){
- break;
- }
- // add the label separator
- if(remainingDestCapacity > 0){
- *currentDest++ = FULL_STOP;
- remainingDestCapacity--;
- }
- reqLength++;
- labelStart = delimiter;
- if(remainingLen >0 ){
- remainingLen = (int32_t)(srcLength - (delimiter - src));
- }
- }
- if(reqLength > MAX_DOMAIN_NAME_LENGTH){
- *status = U_IDNA_DOMAIN_NAME_TOO_LONG_ERROR;
- }
- usprep_close(nameprep);
-
- return u_terminateUChars(dest, destCapacity, reqLength, status);
- }
- U_CAPI int32_t U_EXPORT2
- uidna_IDNToUnicode( const char16_t* src, int32_t srcLength,
- char16_t* dest, int32_t destCapacity,
- int32_t options,
- UParseError* parseError,
- UErrorCode* status){
-
- if(status == nullptr || U_FAILURE(*status)){
- return 0;
- }
- if((src==nullptr) || (srcLength < -1) || (destCapacity<0) || (!dest && destCapacity > 0)){
- *status = U_ILLEGAL_ARGUMENT_ERROR;
- return 0;
- }
- int32_t reqLength = 0;
- UStringPrepProfile* nameprep = usprep_openByType(USPREP_RFC3491_NAMEPREP, status);
-
- if(U_FAILURE(*status)){
- return 0;
- }
- //initialize pointers
- char16_t *delimiter = (char16_t*)src;
- char16_t *labelStart = (char16_t*)src;
- char16_t *currentDest = (char16_t*) dest;
- int32_t remainingLen = srcLength;
- int32_t remainingDestCapacity = destCapacity;
- int32_t labelLen = 0, labelReqLength = 0;
- UBool done = false;
- for(;;){
- labelLen = getNextSeparator(labelStart,remainingLen, &delimiter,&done);
-
- // The RFC states that
- // <quote>
- // ToUnicode never fails. If any step fails, then the original input
- // is returned immediately in that step.
- // </quote>
- // _internal_toUnicode will copy the label.
- /*if(labelLen==0 && done==false){
- *status = U_IDNA_ZERO_LENGTH_LABEL_ERROR;
- break;
- }*/
-
- labelReqLength = _internal_toUnicode(labelStart, labelLen,
- currentDest, remainingDestCapacity,
- options, nameprep,
- parseError, status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- *status = U_ZERO_ERROR; // reset error
- remainingDestCapacity = 0;
- }
- if(U_FAILURE(*status)){
- break;
- }
-
- reqLength +=labelReqLength;
- // adjust the destination pointer
- if(labelReqLength < remainingDestCapacity){
- currentDest = currentDest + labelReqLength;
- remainingDestCapacity -= labelReqLength;
- }else{
- // should never occur
- remainingDestCapacity = 0;
- }
- if(done){
- break;
- }
- // add the label separator
- // Unlike the ToASCII operation we don't normalize the label separators
- if(remainingDestCapacity > 0){
- *currentDest++ = *(labelStart + labelLen);
- remainingDestCapacity--;
- }
- reqLength++;
- labelStart = delimiter;
- if(remainingLen >0 ){
- remainingLen = (int32_t)(srcLength - (delimiter - src));
- }
- }
- if(reqLength > MAX_DOMAIN_NAME_LENGTH){
- *status = U_IDNA_DOMAIN_NAME_TOO_LONG_ERROR;
- }
- usprep_close(nameprep);
-
- return u_terminateUChars(dest, destCapacity, reqLength, status);
- }
- U_CAPI int32_t U_EXPORT2
- uidna_compare( const char16_t *s1, int32_t length1,
- const char16_t *s2, int32_t length2,
- int32_t options,
- UErrorCode* status){
- if(status == nullptr || U_FAILURE(*status)){
- return -1;
- }
- char16_t b1Stack[MAX_IDN_BUFFER_SIZE], b2Stack[MAX_IDN_BUFFER_SIZE];
- char16_t *b1 = b1Stack, *b2 = b2Stack;
- int32_t b1Len, b2Len, b1Capacity = MAX_IDN_BUFFER_SIZE, b2Capacity = MAX_IDN_BUFFER_SIZE;
- int32_t result=-1;
-
- UParseError parseError;
- b1Len = uidna_IDNToASCII(s1, length1, b1, b1Capacity, options, &parseError, status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- // redo processing of string
- b1 = (char16_t*) uprv_malloc(b1Len * U_SIZEOF_UCHAR);
- if(b1==nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- *status = U_ZERO_ERROR; // reset error
-
- b1Len = uidna_IDNToASCII(s1,length1,b1,b1Len, options, &parseError, status);
-
- }
- b2Len = uidna_IDNToASCII(s2,length2, b2,b2Capacity, options, &parseError, status);
- if(*status == U_BUFFER_OVERFLOW_ERROR){
- // redo processing of string
- b2 = (char16_t*) uprv_malloc(b2Len * U_SIZEOF_UCHAR);
- if(b2==nullptr){
- *status = U_MEMORY_ALLOCATION_ERROR;
- goto CLEANUP;
- }
- *status = U_ZERO_ERROR; // reset error
-
- b2Len = uidna_IDNToASCII(s2, length2, b2, b2Len, options, &parseError, status);
-
- }
- // when toASCII is applied all label separators are replaced with FULL_STOP
- result = compareCaseInsensitiveASCII(b1,b1Len,b2,b2Len);
- CLEANUP:
- if(b1 != b1Stack){
- uprv_free(b1);
- }
- if(b2 != b2Stack){
- uprv_free(b2);
- }
- return result;
- }
- #endif /* #if !UCONFIG_NO_IDNA */
|