// script_english_number_* String do_english_num(String input, String(*fun)(int)) { if (is_substr(input, 0, _("<param-"))) { // a keyword parameter, of the form "<param->123</param->" size_t start = skip_tag(input, 0); if (start != String::npos) { size_t end = input.find_first_of(_('<'), start); if (end != String::npos) { String is = input.substr(start, end - start); long i = 0; if (is.ToLong(&i)) { if (i == 1) { return _("<hint-1>") + substr_replace(input, start, end, fun(i)); } else { return _("<hint-2>") + substr_replace(input, start, end, fun(i)); } } } } return _("<hint-2>") + input; } else { long i = 0; if (input.ToLong(&i)) { return fun(i); } return input; } }
void Keyword::prepare(const vector<KeywordParamP>& param_types, bool force) { if (!force && !match_re.empty()) return; parameters.clear(); // Prepare regex String regex; String text; // normal, non-regex, text vector<KeywordParamP>::const_iterator param = parameters.begin(); #if USE_CASE_INSENSITIVE_KEYWORDS regex = _("(?i)"); // case insensitive matching #endif // Parse the 'match' string for (size_t i = 0 ; i < match.size() ;) { Char c = match.GetChar(i); if (is_substr(match, i, _("<atom-param"))) { // parameter, determine type... size_t start = skip_tag(match, i), end = match_close_tag(match, i); String type = match.substr(start, end-start); // find parameter type 'type' KeywordParamP param; FOR_EACH_CONST(pt, param_types) { if (pt->name == type) { param = pt; break; } } if (!param) { // throwing an error can mean a set will not be loaded! // instead, simply disable the keyword //throw InternalError(_("Unknown keyword parameter type: ") + type); handle_error(_("Unknown keyword parameter type: ") + type); valid = false; return; } parameters.push_back(param); // modify regex : match text before param->compile(); // remove the separator from the text to prevent duplicates param->eat_separator_before(text); regex += _("(") + regex_escape(text) + _(")"); text.clear(); // modify regex : match parameter regex += _("(") + make_non_capturing(param->match) + (param->optional ? _(")?") : _(")")); i = skip_tag(match, end); // eat separator_after? param->eat_separator_after(match, i); } else {
// script_english_singular/plural/singplur String do_english(String input, String(*fun)(const String&)) { if (is_substr(input, 0, _("<param-"))) { // a keyword parameter, of the form "<param->123</param->" size_t start = skip_tag(input, 0); if (start != String::npos) { size_t end = input.find_first_of(_('<'), start); if (end != String::npos) { String is = input.substr(start, end - start); return substr_replace(input, start, end, fun(is)); } } return input; // failed } else { return fun(input); } }
struct skin_element* skin_parse(const char* document, skin_callback cb, void* cb_data) { callback = cb; callback_data = cb_data; #else struct skin_element* skin_parse(const char* document) { #endif struct skin_element* root = NULL; struct skin_element* last = NULL; const char* cursor = document; /*Keeps track of location in the document*/ skin_line = 1; skin_start = (char*)document; viewport_line = 0; skin_clear_errors(); while(*cursor != '\0') { struct skin_element* tree = skin_parse_viewport(&cursor); if(!root) { root = tree; last = root; } else { last->next = skin_buffer_to_offset(tree); last = tree; } if(!last) { skin_free_tree(root); /* Clearing any memory already used */ return NULL; } /* Making sure last is at the end */ while(IS_VALID_OFFSET(last->next)) last = skin_buffer_from_offset(last->next); } return root; } static struct skin_element* skin_parse_viewport(const char** document) { struct skin_element* root = NULL; struct skin_element* last = NULL; struct skin_element* retval = NULL; retval = skin_alloc_element(); if (!retval) return NULL; retval->type = VIEWPORT; retval->children_count = 1; retval->line = skin_line; viewport_line = skin_line; OFFSETTYPE(struct skin_element*)* children; const char* cursor = *document; /* Keeps track of location in the document */ const char* bookmark; /* Used when we need to look ahead */ int sublines = 0; /* Flag for parsing sublines */ /* Parsing out the viewport tag if there is one */ if(check_viewport(cursor)) { if (!skin_parse_tag(retval, &cursor)) return NULL; if(*cursor == '\n') { cursor++; skin_line++; } } #ifdef ROCKBOX else if (callback) { if (callback(retval, callback_data) == CALLBACK_ERROR) { skin_error(GOT_CALLBACK_ERROR, cursor); return NULL; } } #endif if (check_viewport(cursor)) { retval->children_count = 0; *document = cursor; return retval; } retval->children_count = 1; children = skin_alloc_children(1); if (!children) return NULL; do { /* First, we check to see if this line will contain sublines */ bookmark = cursor; sublines = 0; while(*cursor != '\n' && *cursor != '\0' && !(check_viewport(cursor) && cursor != *document)) { if(*cursor == MULTILINESYM) { sublines = 1; break; } else if(*cursor == TAGSYM) { skip_tag(&cursor); } else if(*cursor == COMMENTSYM) { skip_comment(&cursor); } else { /* Advancing the cursor as normal */ cursor++; } } cursor = bookmark; if(sublines) { struct skin_element* out = skin_parse_sublines(&cursor); if (!root) { root = out; last = root; } else { last->next = skin_buffer_to_offset(out); last = out; } if(!last) return NULL; } else { #ifdef ROCKBOX /* strip all leading comments */ while(*cursor == '#') { skip_comment(&cursor); skin_line++; } if (check_viewport(cursor)) break; #endif struct skin_element* out = skin_parse_line(&cursor); if (!root) { root = out; last = root; } else { last->next = skin_buffer_to_offset(out); last = out; } if(!last) return NULL; } /* Making sure last is at the end */ while(IS_VALID_OFFSET(last->next)) last = skin_buffer_from_offset(last->next); if(*cursor == '\n') { cursor++; skin_line++; } #ifdef ROCKBOX /* strip all comments */ while(*cursor == '#') { skip_comment(&cursor); skin_line++; } if (check_viewport(cursor)) break; #endif } while(*cursor != '\0' && !(check_viewport(cursor) && cursor != *document)); *document = cursor; children[0] = skin_buffer_to_offset(root); retval->children = skin_buffer_to_offset(children); return retval; } /* Auxiliary Parsing Functions */ static struct skin_element* skin_parse_line(const char**document) { return skin_parse_line_optional(document, 0); } /* * If conditional is set to true, then this will break upon encountering * SEPARATESYM. This should only be used when parsing a line inside a * conditional, otherwise just use the wrapper function skin_parse_line() */ static struct skin_element* skin_parse_line_optional(const char** document, int conditional) { const char* cursor = *document; struct skin_element* root = NULL; struct skin_element* current = NULL; struct skin_element* retval = NULL; OFFSETTYPE(struct skin_element*)* children = NULL; /* A wrapper for the line */ retval = skin_alloc_element(); if (!retval) return NULL; retval->type = LINE; retval->line = skin_line; while (*cursor == '\t') cursor++; if(*cursor != '\0' && *cursor != '\n' && *cursor != MULTILINESYM && !(conditional && (*cursor == ARGLISTSEPARATESYM || *cursor == ARGLISTCLOSESYM || *cursor == ENUMLISTSEPARATESYM || *cursor == ENUMLISTCLOSESYM))) { retval->children_count = 1; } else { retval->children_count = 0; } if(retval->children_count > 0) { children = skin_alloc_children(1); if (!children) return NULL; } #ifdef ROCKBOX if (callback) { switch (callback(retval, callback_data)) { case CALLBACK_ERROR: skin_error(GOT_CALLBACK_ERROR, cursor); return NULL; default: break; } } #endif while(*cursor != '\n' && *cursor != '\0' && *cursor != MULTILINESYM && !((*cursor == ARGLISTSEPARATESYM || *cursor == ARGLISTCLOSESYM || *cursor == ENUMLISTSEPARATESYM || *cursor == ENUMLISTCLOSESYM) && conditional) && !(check_viewport(cursor) && cursor != *document)) { /* Allocating memory if necessary */ if(root) { struct skin_element *next = skin_alloc_element(); if (!next) return NULL; current->next = skin_buffer_to_offset(next); current = next; } else { current = skin_alloc_element(); if (!current) return NULL; root = current; } /* Parsing the current element */ if(*cursor == TAGSYM && cursor[1] == CONDITIONSYM) { if(!skin_parse_conditional(current, &cursor)) return NULL; } else if(*cursor == TAGSYM && !find_escape_character(cursor[1])) { if(!skin_parse_tag(current, &cursor)) return NULL; } else if(*cursor == COMMENTSYM) { if(!skin_parse_comment(current, &cursor)) return NULL; } else { if(!skin_parse_text(current, &cursor, conditional)) return NULL; } } /* Moving up the calling function's pointer */ *document = cursor; if(root) { children[0] = skin_buffer_to_offset(root); retval->children = skin_buffer_to_offset(children); } return retval; } static struct skin_element* skin_parse_sublines(const char** document) { return skin_parse_sublines_optional(document, 0); } static struct skin_element* skin_parse_sublines_optional(const char** document, int conditional) { struct skin_element* retval; OFFSETTYPE(struct skin_element*)* children; const char* cursor = *document; int sublines = 1; int i; retval = skin_alloc_element(); if (!retval) return NULL; retval->type = LINE_ALTERNATOR; retval->next = skin_buffer_to_offset(NULL); retval->line = skin_line; while (*cursor == '\t') cursor++; /* First we count the sublines */ while(*cursor != '\0' && *cursor != '\n' && !((*cursor == ARGLISTSEPARATESYM || *cursor == ARGLISTCLOSESYM || *cursor == ENUMLISTSEPARATESYM || *cursor == ENUMLISTCLOSESYM) && conditional) && !(check_viewport(cursor) && cursor != *document)) { if(*cursor == COMMENTSYM) { skip_comment(&cursor); } else if(*cursor == TAGSYM) { skip_tag(&cursor); } else if(*cursor == MULTILINESYM) { sublines++; cursor++; } else { cursor++; } } /* ...and then we parse them */ retval->children_count = sublines; children = skin_alloc_children(sublines); if (!children) return NULL; cursor = *document; for(i = 0; i < sublines; i++) { children[i] = skin_buffer_to_offset(skin_parse_line_optional(&cursor, conditional)); if (children[i] < 0) return NULL; skip_whitespace(&cursor); if(*cursor != MULTILINESYM && i != sublines - 1) { skin_error(MULTILINE_EXPECTED, cursor); return NULL; } else if(i != sublines - 1) { cursor++; } } #ifdef ROCKBOX if (callback) { if (callback(retval, callback_data) == CALLBACK_ERROR) { skin_error(GOT_CALLBACK_ERROR, *document); return NULL; } } #endif *document = cursor; retval->children = skin_buffer_to_offset(children); return retval; } static int skin_parse_tag(struct skin_element* element, const char** document) { const char* cursor = *document + 1; const char* bookmark; char *open_square_bracket = NULL; char tag_name[MAX_TAG_LENGTH]; char* tag_args; const struct tag_info *tag; struct skin_tag_parameter* params = NULL; int num_args = 1; int i; int qmark = 0; /* Flag for the all-or-none option */ int optional = 0; /* Checking the tag name */ for (i=0; cursor[i] && i<MAX_TAG_LENGTH; i++) tag_name[i] = cursor[i]; /* First we check the two characters after the '%', then a single char */ tag = NULL; i = MAX_TAG_LENGTH; while (!tag && i > 1) { tag_name[i-1] = '\0'; tag = find_tag(tag_name); i--; } if(!tag) { skin_error(ILLEGAL_TAG, cursor); return 0; } cursor += i; /* Copying basic tag info */ if(element->type != CONDITIONAL && element->type != VIEWPORT) element->type = TAG; element->tag = tag; tag_args = tag->params; element->line = skin_line; /* Checking for the * flag */ if(tag_args[0] == '?') { qmark = 1; tag_args++; } /* If this tag has no arguments, we can bail out now */ if(strlen(tag_args) == 0 || (tag_args[0] == '|' && *cursor != ARGLISTOPENSYM) || (qmark && *cursor != ARGLISTOPENSYM)) { #ifdef ROCKBOX if (callback) { if (callback(element, callback_data) == CALLBACK_ERROR) { skin_error(GOT_CALLBACK_ERROR, cursor); return 0; } } #endif *document = cursor; return 1; } /* Checking the number of arguments and allocating args */ if(*cursor != ARGLISTOPENSYM && tag_args[0] != '|') { skin_error(ARGLIST_EXPECTED, cursor); return 0; } else { cursor++; } bookmark = cursor; while(*cursor != '\n' && *cursor != '\0' && *cursor != ARGLISTCLOSESYM) { /* Skipping over escaped characters */ if(*cursor == TAGSYM && *(cursor+1) != ARGLISTSEPARATESYM) { skip_tag(&cursor); } else if(*cursor == COMMENTSYM) { skip_comment(&cursor); } else if(*cursor == ARGLISTSEPARATESYM) { num_args++; cursor++; } else { cursor++; } } cursor = bookmark; /* Restoring the cursor */ element->params_count = num_args; params = skin_alloc_params(num_args); if (!params) return 0; /* Now we have to actually parse each argument */ for(i = 0; i < num_args; i++) { char type_code; /* Making sure we haven't run out of arguments */ if(*tag_args == '\0') { skin_error(TOO_MANY_ARGS, cursor); return 0; } /* Checking for the optional bar */ if(*tag_args == '|') { optional = 1; tag_args++; } /* Scanning the arguments */ skip_whitespace(&cursor); /* Checking for comments */ if(*cursor == COMMENTSYM) skip_comment(&cursor); if (*tag_args == '[') { /* we need to guess which type of param it is. * guess using this priority: * default > decimal/integer > single tag/code > string */ int j=0; bool canbedefault = false, last_char_is_percent = false; bool haspercent = false, number = true, hasdecimal = false; char temp_params[8]; open_square_bracket = tag_args; tag_args++; while (*tag_args != ']') { if (*tag_args >= 'a' && *tag_args <= 'z') canbedefault = true; temp_params[j++] = tolower(*tag_args++); } temp_params[j] = '\0'; j = 0; while (cursor[j] && cursor[j] != ',' && cursor[j] != ')') { haspercent = haspercent || (cursor[j] == '%'); hasdecimal = hasdecimal || (cursor[j] == '.'); number = number && (isdigit(cursor[j]) || (cursor[j] == '.') || (cursor[j] == '-') || (cursor[j] == '%')); j++; } last_char_is_percent = cursor[j-1] == '%'; type_code = '?'; if (canbedefault && *cursor == DEFAULTSYM && !isdigit(cursor[1])) { type_code = 'i'; } else if (number && hasdecimal && strchr(temp_params, 'd')) { type_code = 'd'; } else if (number && last_char_is_percent && strchr(temp_params, 'p')) { type_code = 'p'; } else if (number && (strchr(temp_params, 'i') || strchr(temp_params, 'd'))) { type_code = strchr(temp_params, 'i') ? 'i' : 'd'; } else if (haspercent && (strchr(temp_params, 't') || strchr(temp_params, 'c'))) { type_code = strchr(temp_params, 't') ? 't' : 'c'; } else if (strchr(temp_params, 's')) { type_code = 's'; } if (type_code == '?') { skin_error(INSUFFICIENT_ARGS, cursor); return 0; } } else type_code = *tag_args; /* Storing the type code */ params[i].type_code = type_code; /* Checking a nullable argument for null. */ if(*cursor == DEFAULTSYM && !isdigit(cursor[1])) { if(islower(type_code)) { params[i].type = DEFAULT; cursor++; } else { skin_error(DEFAULT_NOT_ALLOWED, cursor); return 0; } } else if(tolower(type_code) == 'i') { /* Scanning an int argument */ if(!isdigit(*cursor) && *cursor != '-') { skin_error(INT_EXPECTED, cursor); return 0; } params[i].type = INTEGER; params[i].data.number = scan_int(&cursor); } else if(tolower(type_code) == 'd' || tolower(type_code) == 'p') { int val = 0; bool have_point = false; bool have_tenth = false; while ( isdigit(*cursor) || *cursor == '.' ) { if (*cursor != '.') { val *= 10; val += *cursor - '0'; if (have_point) { have_tenth = true; cursor++; break; } } else have_point = true; cursor++; } if (have_tenth == false) val *= 10; if (tolower(type_code) == 'd') params[i].type = DECIMAL; else { params[i].type = PERCENT; cursor++; /* skip trailing % sign */ } params[i].data.number = val; } else if(tolower(type_code) == 's' || tolower(type_code) == 'f') { /* Scanning a string argument */ params[i].type = STRING; params[i].data.text = skin_buffer_to_offset(scan_string(&cursor)); } else if(tolower(type_code) == 'c') { /* Recursively parsing a code argument */ params[i].type = CODE; params[i].data.code = skin_buffer_to_offset(skin_parse_code_as_arg(&cursor)); if(params[i].data.code < 0) return 0; } else if (tolower(type_code) == 't') { struct skin_element* child = skin_alloc_element(); child->type = TAG; if (!skin_parse_tag(child, &cursor)) return 0; child->next = skin_buffer_to_offset(NULL); params[i].type = CODE; params[i].data.code = skin_buffer_to_offset(child); } skip_whitespace(&cursor); if(*cursor != ARGLISTSEPARATESYM && i < num_args - 1) { skin_error(SEPARATOR_EXPECTED, cursor); return 0; } else if(*cursor != ARGLISTCLOSESYM && i == num_args - 1) { skin_error(CLOSE_EXPECTED, cursor); return 0; } else { cursor++; } if (*(tag_args + 1) == '*') { if (i+1 == num_args) tag_args += 2; else if (open_square_bracket && *tag_args == ']') { tag_args = open_square_bracket; open_square_bracket = NULL; } } else tag_args++; /* Checking for the optional bar */ if(*tag_args == '|') { optional = 1; tag_args++; } } element->params = skin_buffer_to_offset(params); /* Checking for a premature end */ if(*tag_args != '\0' && !optional) { skin_error(INSUFFICIENT_ARGS, cursor); return 0; } #ifdef ROCKBOX if (callback) { if (callback(element, callback_data) == CALLBACK_ERROR) { skin_error(GOT_CALLBACK_ERROR, *document); return 0; } } #endif *document = cursor; return 1; } /* * If the conditional flag is set true, then parsing text will stop at an * ARGLISTSEPARATESYM. Only set that flag when parsing within a conditional */ static int skin_parse_text(struct skin_element* element, const char** document, int conditional) { const char* cursor = *document; int length = 0; int dest; char *text = NULL; /* First figure out how much text we're copying */ while(*cursor != '\0' && *cursor != '\n' && *cursor != MULTILINESYM && *cursor != COMMENTSYM && !((*cursor == ARGLISTSEPARATESYM || *cursor == ARGLISTCLOSESYM || *cursor == ENUMLISTSEPARATESYM || *cursor == ENUMLISTCLOSESYM) && conditional)) { /* Dealing with possibility of escaped characters */ if(*cursor == TAGSYM) { if(find_escape_character(cursor[1])) cursor++; else break; } length++; cursor++; } cursor = *document; /* Copying the text into the element struct */ element->type = TEXT; element->line = skin_line; element->next = skin_buffer_to_offset(NULL); text = skin_alloc_string(length); element->data = skin_buffer_to_offset(text); if (element->data < 0) return 0; for(dest = 0; dest < length; dest++) { /* Advancing cursor if we've encountered an escaped character */ if(*cursor == TAGSYM) cursor++; text[dest] = *cursor; cursor++; } text[length] = '\0'; #ifdef ROCKBOX if (callback) { if (callback(element, callback_data) == CALLBACK_ERROR) { skin_error(GOT_CALLBACK_ERROR, *document); return 0; } } #endif *document = cursor; return 1; } static int skin_parse_conditional(struct skin_element* element, const char** document) { const char* cursor = *document + 1; /* Starting past the "%" */ const char* bookmark; int children = 1; int i; #ifdef ROCKBOX bool feature_available = true; const char *false_branch = NULL; const char *conditional_end = NULL; #endif OFFSETTYPE(struct skin_element*)* children_array = NULL; /* Some conditional tags allow for target feature checking, * so to handle that call the callback as usual with type == TAG * then call it a second time with type == CONDITIONAL and check the return * value */ element->type = TAG; element->line = skin_line; /* Parsing the tag first */ if(!skin_parse_tag(element, &cursor)) return 0; element->type = CONDITIONAL; #ifdef ROCKBOX if (callback) { switch (callback(element, callback_data)) { case FEATURE_NOT_AVAILABLE: feature_available = false; break; case CALLBACK_ERROR: return 0; default: break; } } #endif /* Counting the children */ if(*(cursor++) != ENUMLISTOPENSYM) { skin_error(ARGLIST_EXPECTED, cursor); return 0; } bookmark = cursor; while(*cursor != ENUMLISTCLOSESYM && *cursor != '\0') { if(*cursor == COMMENTSYM) { skip_comment(&cursor); } else if(*cursor == TAGSYM) { skip_tag(&cursor); } else if(*cursor == ENUMLISTSEPARATESYM) { children++; cursor++; if (*cursor == '\n') cursor++; #ifdef ROCKBOX if (false_branch == NULL && !feature_available) { false_branch = cursor; children--; } #endif } else { cursor++; } } #ifdef ROCKBOX if (*cursor == ENUMLISTCLOSESYM && false_branch == NULL && !feature_available) { false_branch = cursor+1; children--; } if (element->tag->flags&FEATURE_TAG) { if (feature_available && children > 1) children--; } conditional_end = cursor; /* if we are skipping the true branch fix that up */ cursor = false_branch ? false_branch : bookmark; #else cursor = bookmark; #endif /* Parsing the children */ /* Feature tags could end up having 0 children which breaks * the render in dangerous ways. Minor hack, but insert an empty * child. (e.g %?xx<foo> when xx isnt available ) */ if (children == 0) { const char* emptyline= ""; children = 1; children_array = skin_alloc_children(children); if (!children_array) return 0; element->children_count = children; children_array[0] = skin_buffer_to_offset(skin_parse_code_as_arg(&emptyline)); } else { children_array = skin_alloc_children(children); if (!children_array) return 0; element->children_count = children; for(i = 0; i < children; i++) { if (*cursor == '\n') { skin_line++; cursor++; } children_array[i] = skin_buffer_to_offset(skin_parse_code_as_arg(&cursor)); if (children_array[i] < 0) return 0; skip_whitespace(&cursor); #ifdef ROCKBOX if ((element->tag->flags&FEATURE_TAG) && feature_available) cursor = conditional_end; #endif if(i < children - 1 && *cursor != ENUMLISTSEPARATESYM) { skin_error(SEPARATOR_EXPECTED, cursor); return 0; } else if(i == children - 1 && *cursor != ENUMLISTCLOSESYM) { skin_error(CLOSE_EXPECTED, cursor); return 0; } else { cursor++; } } } *document = cursor; element->children = skin_buffer_to_offset(children_array); return 1; } static int skin_parse_comment(struct skin_element* element, const char** document) { const char* cursor = *document; #ifndef ROCKBOX char* text = NULL; #endif int length; /* * Finding the index of the ending newline or null-terminator * The length of the string of interest doesn't include the leading #, the * length we need to reserve is the same as the index of the last character */ for(length = 0; cursor[length] != '\n' && cursor[length] != '\0'; length++); element->type = COMMENT; element->line = skin_line; #ifdef ROCKBOX element->data = INVALID_OFFSET; #else element->data = text = skin_alloc_string(length); if (!element->data) return 0; /* We copy from one char past cursor to leave out the # */ memcpy((void*)text, (void*)(cursor + 1), sizeof(char) * (length-1)); text[length - 1] = '\0'; #endif if(cursor[length] == '\n') skin_line++; *document += (length); /* Move cursor up past # and all text */ if(**document == '\n') (*document)++; return 1; } static struct skin_element* skin_parse_code_as_arg(const char** document) { int sublines = 0; const char* cursor = *document; /* Checking for sublines */ while(*cursor != '\n' && *cursor != '\0' && *cursor != ENUMLISTSEPARATESYM && *cursor != ARGLISTSEPARATESYM && *cursor != ENUMLISTCLOSESYM && *cursor != ARGLISTCLOSESYM) { if(*cursor == MULTILINESYM) { sublines = 1; break; } else if(*cursor == TAGSYM) { skip_tag(&cursor); } else { /* Advancing the cursor as normal */ cursor++; } } if(sublines) return skin_parse_sublines_optional(document, 1); else return skin_parse_line_optional(document, 1); }
aku_Status SeriesParser::to_normal_form(const char* begin, const char* end, char* out_begin, char* out_end, const char** keystr_begin, const char** keystr_end) { // Verify args if (end < begin) { return AKU_EBAD_ARG; } if (out_end < out_begin) { return AKU_EBAD_ARG; } int series_name_len = end - begin; if (series_name_len > AKU_LIMITS_MAX_SNAME) { return AKU_EBAD_DATA; } if (series_name_len > (out_end - out_begin)) { return AKU_EBAD_ARG; } char* it_out = out_begin; const char* it = begin; // Get metric name it = skip_space(it, end); it = copy_until(it, end, ' ', &it_out); it = skip_space(it, end); if (it == end) { // At least one tag should be specified return AKU_EBAD_DATA; } *keystr_begin = it_out; // Get pointers to the keys const char* tags[AKU_LIMITS_MAX_TAGS]; auto ix_tag = 0u; bool error = false; while(it < end && ix_tag < AKU_LIMITS_MAX_TAGS) { tags[ix_tag] = it; it = skip_tag(it, end, &error); it = skip_space(it, end); if (!error) { ix_tag++; } else { break; } } if (error) { // Bad string return AKU_EBAD_DATA; } if (ix_tag == 0) { // User should specify at least one tag return AKU_EBAD_DATA; } std::sort(tags, tags + ix_tag, [tags, end](const char* lhs, const char* rhs) { // lhs should be always less thenn rhs auto lenl = 0u; auto lenr = 0u; if (lhs < rhs) { lenl = rhs - lhs; lenr = end - rhs; } else { lenl = end - lhs; lenr = lhs - rhs; } auto it = 0u; while(true) { if (it >= lenl || it >= lenr) { return it < lenl; } if (lhs[it] == '=' || rhs[it] == '=') { return lhs[it] == '='; } if (lhs[it] < rhs[it]) { return true; } else if (lhs[it] > rhs[it]) { return false; } it++; } return true; }); // Copy tags to output string for (auto i = 0u; i < ix_tag; i++) { // insert space *it_out++ = ' '; // insert tag const char* tag = tags[i]; copy_until(tag, end, ' ', &it_out); } *keystr_begin = skip_space(*keystr_begin, out_end); *keystr_end = it_out; return AKU_SUCCESS; }
static void dump_info_headers( const char *input, uint length ) /*************************************************************/ { const uint_8 *p; uint_32 abbrev_code; uint_32 abbrev_offset; uint_8 * abbrev; uint_32 tag; uint_32 unit_length; uint_32 tag_offset; const uint_8 *unit_base; info_state state; bool found; p = input; state.addr_size = 0; found = false; while( p - input < length ) { state.cu_header = p - input; unit_length = get_u32( (uint_32 *)p ); unit_base = p + sizeof( uint_32 ); Wdputs( "Length: " ); Puthex( unit_length, 8 ); Wdputslc( "\nVersion: " ); Puthex( get_u16( (uint_16 *)(p + 4) ), 4 ); Wdputslc( "\nAbbrev: " ); abbrev_offset = get_u32( (uint_32 *)(p + 6) ); Puthex( abbrev_offset, 8 ); state.addr_size = *(p + 10); Wdputslc( "\nAddress Size " ); Puthex( *(p + 10), 2 ); Wdputslc( "\n" ); p += 11; while( p - unit_base < unit_length ) { tag_offset = p - input; p = DecodeULEB128( p, &abbrev_code ); if( abbrev_code == 0 ) continue; abbrev = find_abbrev( abbrev_offset, abbrev_code ); if( abbrev == NULL ) { Wdputs( "can't find abbreviation " ); Puthex( abbrev_code, 8 ); Wdputslc( "\n" ); break; } if( p >= input + length ) break; abbrev = DecodeULEB128( abbrev, &tag ); abbrev++; state.abbrev = abbrev; state.p = p; if( tag_offset == 0x59a125 ) { found = true; } if( found ) { Wdputs( "Offset: " ); Puthex( tag_offset, 8 ); Wdputs( " Code: " ); Puthex( abbrev_code, 8 ); Wdputslc( "\n" ); Wdputs( " " ); getTAG( tag ); Wdputslc( "\n" ); if( !dump_tag( &state ) )break; } else { skip_tag( &state ); } p = state.p; } if( found )break; } }
/** A hint is formed by * 1. an insertion of a parameter, <param-..>...</param->. * 2. a <hint-1> or <hint-2> tag * * Hints have the following meaning: * - "<hint-1>xxx(yyy)zzz" ---> "xxxzzz" (singular) * - "<hint-2>xxx(yyy)zzz" ---> "xxxyyyzzz" (plural) * - "[^., ]a <param-..>[aeiou]" ---> "\1 an \2" (articla 'an', case insensitive) * - "<hint-?>" ---> "" (remove <hint>s afterwards) * * Note: there is no close tags for hints */ String process_english_hints(const String& str) { String ret; ret.reserve(str.size()); // have we seen a <hint-1/2>? // 1 for singular, 2 for plural int singplur = 0; for (size_t i = 0 ; i < str.size() ; ) { Char c = str.GetChar(i); if (is_substr(str, i, _("<hint-"))) { Char h = str.GetChar(i + 6); // hint code if (h == _('1')) { singplur = 1; } else if (h == _('2')) { singplur = 2; } i = skip_tag(str, i); } else if (is_substr(str, i, _("<param-"))) { size_t after = skip_tag(str, i); if (after != String::npos) { Char c = str.GetChar(after); if (singplur == 1 && is_substr(str, after, _("a</param-"))) { // a -> an, where the a originates from english_number_a(1) size_t after_end = skip_tag(str,after+1); if (after_end == String::npos) { throw Error(_("Incomplete </param> tag")); } if (after_end != String::npos && after_end + 1 < str.size() && isSpace(str.GetChar(after_end)) && is_vowel(str.GetChar(after_end+1))) { ret.append(str,i,after-i); ret += _("an"); i = after + 1; continue; } } else if (is_vowel(c) && ret.size() >= 2) { // a -> an? // is there "a" before this? String last = ret.substr(ret.size() - 2); if ( (ret.size() == 2 || !isAlpha(ret.GetChar(ret.size() - 3))) && (last == _("a ") || last == _("A ")) ) { ret.insert(ret.size() - 1, _('n')); } } else if (is_substr(str, after, _("</param-")) && ret.size() >= 1 && ret.GetChar(ret.size() - 1) == _(' ')) { // empty param, drop space before it ret.resize(ret.size() - 1); } } ret.append(str,i,min(after,str.size())-i); i = after; } else if (is_substr(str, i, _("<singular>"))) { // singular -> keep, plural -> drop size_t start = skip_tag(str, i); size_t end = match_close_tag(str, start); if (singplur == 1 && end != String::npos) { ret += str.substr(start, end - start); } singplur = 0; i = skip_tag(str, end); } else if (is_substr(str, i, _("<plural>"))) { // singular -> drop, plural -> keep size_t start = skip_tag(str, i); size_t end = match_close_tag(str, start); if (singplur == 2 && end != String::npos) { ret += str.substr(start, end - start); } singplur = 0; i = skip_tag(str, end); } else if (c == _('(') && singplur) { // singular -> drop (...), plural -> keep it size_t end = str.find_first_of(_(')'), i); if (end != String::npos) { if (singplur == 2) { ret += str.substr(i + 1, end - i - 1); } i = end + 1; } else { // handle like normal ret += c; ++i; } singplur = 0; } else if (c == _('<')) { size_t after = skip_tag(str, i); ret.append(str,i,min(after,str.size())-i); i = after; } else { ret += c; ++i; } } return ret; }