summaryrefslogtreecommitdiff
path: root/indra/llui/llkeywords.h
diff options
context:
space:
mode:
Diffstat (limited to 'indra/llui/llkeywords.h')
-rw-r--r--indra/llui/llkeywords.h266
1 files changed, 133 insertions, 133 deletions
diff --git a/indra/llui/llkeywords.h b/indra/llui/llkeywords.h
index 2410fe7d5a..9dcdea121b 100644
--- a/indra/llui/llkeywords.h
+++ b/indra/llui/llkeywords.h
@@ -44,162 +44,162 @@ typedef LLPointer<LLTextSegment> LLTextSegmentPtr;
class LLKeywordToken
{
public:
- /**
- * @brief Types of tokens/delimters being parsed.
- *
- * @desc Tokens/delimiters that need to be identified/highlighted. All are terminated if an EOF is encountered.
- * - TT_WORD are keywords in the normal sense, i.e. constants, events, etc.
- * - TT_LINE are for entire lines (currently only flow control labels use this).
- * - TT_ONE_SIDED_DELIMITER are for open-ended delimiters which are terminated by EOL.
- * - TT_TWO_SIDED_DELIMITER are for delimiters that end with a different delimiter than they open with.
- * - TT_DOUBLE_QUOTATION_MARKS are for delimiting areas using the same delimiter to open and close.
- */
- typedef enum e_token_type
- {
- TT_UNKNOWN,
- TT_WORD,
- TT_LINE,
- TT_TWO_SIDED_DELIMITER,
- TT_ONE_SIDED_DELIMITER,
- TT_DOUBLE_QUOTATION_MARKS,
- // Following constants are more specific versions of the preceding ones
- TT_CONSTANT, // WORD
- TT_CONTROL, // WORD
- TT_EVENT, // WORD
- TT_FUNCTION, // WORD
- TT_LABEL, // LINE
- TT_SECTION, // WORD
- TT_TYPE // WORD
- } ETokenType;
-
- LLKeywordToken( ETokenType type, const LLColor4& color, const LLWString& token, const LLWString& tool_tip, const LLWString& delimiter )
- :
- mType( type ),
- mToken( token ),
- mColor( color ),
- mToolTip( tool_tip ),
- mDelimiter( delimiter ) // right delimiter
- {
- }
-
- S32 getLengthHead() const { return mToken.size(); }
- S32 getLengthTail() const { return mDelimiter.size(); }
- bool isHead(const llwchar* s) const;
- bool isTail(const llwchar* s) const;
- const LLWString& getToken() const { return mToken; }
- const LLColor4& getColor() const { return mColor; }
- ETokenType getType() const { return mType; }
- const LLWString& getToolTip() const { return mToolTip; }
- const LLWString& getDelimiter() const { return mDelimiter; }
+ /**
+ * @brief Types of tokens/delimters being parsed.
+ *
+ * @desc Tokens/delimiters that need to be identified/highlighted. All are terminated if an EOF is encountered.
+ * - TT_WORD are keywords in the normal sense, i.e. constants, events, etc.
+ * - TT_LINE are for entire lines (currently only flow control labels use this).
+ * - TT_ONE_SIDED_DELIMITER are for open-ended delimiters which are terminated by EOL.
+ * - TT_TWO_SIDED_DELIMITER are for delimiters that end with a different delimiter than they open with.
+ * - TT_DOUBLE_QUOTATION_MARKS are for delimiting areas using the same delimiter to open and close.
+ */
+ typedef enum e_token_type
+ {
+ TT_UNKNOWN,
+ TT_WORD,
+ TT_LINE,
+ TT_TWO_SIDED_DELIMITER,
+ TT_ONE_SIDED_DELIMITER,
+ TT_DOUBLE_QUOTATION_MARKS,
+ // Following constants are more specific versions of the preceding ones
+ TT_CONSTANT, // WORD
+ TT_CONTROL, // WORD
+ TT_EVENT, // WORD
+ TT_FUNCTION, // WORD
+ TT_LABEL, // LINE
+ TT_SECTION, // WORD
+ TT_TYPE // WORD
+ } ETokenType;
+
+ LLKeywordToken( ETokenType type, const LLColor4& color, const LLWString& token, const LLWString& tool_tip, const LLWString& delimiter )
+ :
+ mType( type ),
+ mToken( token ),
+ mColor( color ),
+ mToolTip( tool_tip ),
+ mDelimiter( delimiter ) // right delimiter
+ {
+ }
+
+ S32 getLengthHead() const { return mToken.size(); }
+ S32 getLengthTail() const { return mDelimiter.size(); }
+ bool isHead(const llwchar* s) const;
+ bool isTail(const llwchar* s) const;
+ const LLWString& getToken() const { return mToken; }
+ const LLColor4& getColor() const { return mColor; }
+ ETokenType getType() const { return mType; }
+ const LLWString& getToolTip() const { return mToolTip; }
+ const LLWString& getDelimiter() const { return mDelimiter; }
#ifdef _DEBUG
- void dump();
+ void dump();
#endif
private:
- ETokenType mType;
- LLWString mToken;
- LLColor4 mColor;
- LLWString mToolTip;
- LLWString mDelimiter;
+ ETokenType mType;
+ LLWString mToken;
+ LLColor4 mColor;
+ LLWString mToolTip;
+ LLWString mDelimiter;
};
class LLKeywords
{
public:
- LLKeywords();
- ~LLKeywords();
+ LLKeywords();
+ ~LLKeywords();
- void clearLoaded() { mLoaded = false; }
- LLColor4 getColorGroup(const std::string& key_in);
- bool isLoaded() const { return mLoaded; }
+ void clearLoaded() { mLoaded = false; }
+ LLColor4 getColorGroup(const std::string& key_in);
+ bool isLoaded() const { return mLoaded; }
- void findSegments(std::vector<LLTextSegmentPtr> *seg_list,
- const LLWString& text,
- class LLTextEditor& editor,
+ void findSegments(std::vector<LLTextSegmentPtr> *seg_list,
+ const LLWString& text,
+ class LLTextEditor& editor,
LLStyleConstSP style);
- void initialize(LLSD SyntaxXML);
- void processTokens();
-
- // Add the token as described
- void addToken(LLKeywordToken::ETokenType type,
- const std::string& key,
- const LLColor4& color,
- const std::string& tool_tip = LLStringUtil::null,
- const std::string& delimiter = LLStringUtil::null);
-
- // This class is here as a performance optimization.
- // The word token map used to be defined as std::map<LLWString, LLKeywordToken*>.
- // This worked, but caused a performance bottleneck due to memory allocation and string copies
- // because it's not possible to search such a map without creating an LLWString.
- // Using this class as the map index instead allows us to search using segments of an existing
- // text run without copying them first, which greatly reduces overhead in LLKeywords::findSegments().
- class WStringMapIndex
- {
- public:
- // copy constructor
- WStringMapIndex(const WStringMapIndex& other);
- // constructor from a string (copies the string's data into the new object)
- WStringMapIndex(const LLWString& str);
- // constructor from pointer and length
- // NOTE: does NOT copy data, caller must ensure that the lifetime of the pointer exceeds that of the new object!
- WStringMapIndex(const llwchar *start, size_t length);
- ~WStringMapIndex();
- bool operator<(const WStringMapIndex &other) const;
- private:
- void copyData(const llwchar *start, size_t length);
- const llwchar *mData;
- size_t mLength;
- bool mOwner;
-
-
- LLColor4 mColor;
- };
-
- typedef std::map<WStringMapIndex, LLKeywordToken*> word_token_map_t;
- typedef word_token_map_t::const_iterator keyword_iterator_t;
- keyword_iterator_t begin() const { return mWordTokenMap.begin(); }
- keyword_iterator_t end() const { return mWordTokenMap.end(); }
-
- typedef std::map<WStringMapIndex, LLColor4> group_color_map_t;
- typedef group_color_map_t::const_iterator color_iterator_t;
- group_color_map_t mColorGroupMap;
+ void initialize(LLSD SyntaxXML);
+ void processTokens();
+
+ // Add the token as described
+ void addToken(LLKeywordToken::ETokenType type,
+ const std::string& key,
+ const LLColor4& color,
+ const std::string& tool_tip = LLStringUtil::null,
+ const std::string& delimiter = LLStringUtil::null);
+
+ // This class is here as a performance optimization.
+ // The word token map used to be defined as std::map<LLWString, LLKeywordToken*>.
+ // This worked, but caused a performance bottleneck due to memory allocation and string copies
+ // because it's not possible to search such a map without creating an LLWString.
+ // Using this class as the map index instead allows us to search using segments of an existing
+ // text run without copying them first, which greatly reduces overhead in LLKeywords::findSegments().
+ class WStringMapIndex
+ {
+ public:
+ // copy constructor
+ WStringMapIndex(const WStringMapIndex& other);
+ // constructor from a string (copies the string's data into the new object)
+ WStringMapIndex(const LLWString& str);
+ // constructor from pointer and length
+ // NOTE: does NOT copy data, caller must ensure that the lifetime of the pointer exceeds that of the new object!
+ WStringMapIndex(const llwchar *start, size_t length);
+ ~WStringMapIndex();
+ bool operator<(const WStringMapIndex &other) const;
+ private:
+ void copyData(const llwchar *start, size_t length);
+ const llwchar *mData;
+ size_t mLength;
+ bool mOwner;
+
+
+ LLColor4 mColor;
+ };
+
+ typedef std::map<WStringMapIndex, LLKeywordToken*> word_token_map_t;
+ typedef word_token_map_t::const_iterator keyword_iterator_t;
+ keyword_iterator_t begin() const { return mWordTokenMap.begin(); }
+ keyword_iterator_t end() const { return mWordTokenMap.end(); }
+
+ typedef std::map<WStringMapIndex, LLColor4> group_color_map_t;
+ typedef group_color_map_t::const_iterator color_iterator_t;
+ group_color_map_t mColorGroupMap;
#ifdef _DEBUG
- void dump();
+ void dump();
#endif
protected:
- void processTokensGroup(const LLSD& Tokens, const std::string& Group);
- void insertSegment(std::vector<LLTextSegmentPtr>& seg_list,
- LLTextSegmentPtr new_segment,
- S32 text_len,
- const LLColor4 &defaultColor,
- class LLTextEditor& editor);
- void insertSegments(const LLWString& wtext,
- std::vector<LLTextSegmentPtr>& seg_list,
- LLKeywordToken* token,
- S32 text_len,
- S32 seg_start,
- S32 seg_end,
- LLStyleConstSP style,
- LLTextEditor& editor);
+ void processTokensGroup(const LLSD& Tokens, const std::string& Group);
+ void insertSegment(std::vector<LLTextSegmentPtr>& seg_list,
+ LLTextSegmentPtr new_segment,
+ S32 text_len,
+ const LLColor4 &defaultColor,
+ class LLTextEditor& editor);
+ void insertSegments(const LLWString& wtext,
+ std::vector<LLTextSegmentPtr>& seg_list,
+ LLKeywordToken* token,
+ S32 text_len,
+ S32 seg_start,
+ S32 seg_end,
+ LLStyleConstSP style,
+ LLTextEditor& editor);
void insertSegment(std::vector<LLTextSegmentPtr>& seg_list, LLTextSegmentPtr new_segment, S32 text_len, LLStyleConstSP style, LLTextEditor& editor );
- bool mLoaded;
- LLSD mSyntax;
- word_token_map_t mWordTokenMap;
- typedef std::deque<LLKeywordToken*> token_list_t;
- token_list_t mLineTokenList;
- token_list_t mDelimiterTokenList;
+ bool mLoaded;
+ LLSD mSyntax;
+ word_token_map_t mWordTokenMap;
+ typedef std::deque<LLKeywordToken*> token_list_t;
+ token_list_t mLineTokenList;
+ token_list_t mDelimiterTokenList;
- typedef std::map<std::string, std::string> element_attributes_t;
- typedef element_attributes_t::const_iterator attribute_iterator_t;
- element_attributes_t mAttributes;
- std::string getAttribute(const std::string& key);
+ typedef std::map<std::string, std::string> element_attributes_t;
+ typedef element_attributes_t::const_iterator attribute_iterator_t;
+ element_attributes_t mAttributes;
+ std::string getAttribute(const std::string& key);
- std::string getArguments(LLSD& arguments);
+ std::string getArguments(LLSD& arguments);
};
#endif // LL_LLKEYWORDS_H