Changeset 286062 in webkit

Timestamp:

Nov 19, 2021, 7:57:39 AM (5 years ago)

Author:

commit-queue@webkit.org

Message:

Remove allocation in JSON::Value::parseJSON
​https://bugs.webkit.org/show_bug.cgi?id=233346

Patch by Alex Christensen <​achristensen@webkit.org> on 2021-11-19
Reviewed by Yusuke Suzuki.

Parse the characters as Latin1 characters if we have an 8 bit string rather than converting them to UTF-16 just to be parsed.

• wtf/JSONValues.cpp:

(WTF::JSONImpl::Value::parseJSON):

Location:

trunk/Source/WTF

Files:

• ChangeLog (modified) (1 diff)
• wtf/JSONValues.cpp (modified) (12 diffs)

trunk/Source/WTF/ChangeLog

@@ +1 @@
+2021-11-19  Alex Christensen  <achristensen@webkit.org>
+
+        Remove allocation in JSON::Value::parseJSON
+        https://bugs.webkit.org/show_bug.cgi?id=233346
+
+        Reviewed by Yusuke Suzuki.
+
+        Parse the characters as Latin1 characters if we have an 8 bit string rather than converting them to UTF-16 just to be parsed.
+
+        * wtf/JSONValues.cpp:
+        (WTF::JSONImpl::Value::parseJSON):
+
 2021-11-18  Antoine Quint  <graouts@webkit.org>
 

trunk/Source/WTF/wtf/JSONValues.cpp

@@ -62 +62 @@
 const char* const falseToken = "false";
 
-bool parseConstToken(const UChar* start, const UChar* end, const UChar** tokenEnd, const char* token)
+template<typename CodeUnit>
+bool parseConstToken(const CodeUnit* start, const CodeUnit* end, const CodeUnit** tokenEnd, const char* token)
 {
     while (start < end && *token != '\0' && *start++ == *token++) { }
@@ -73 +74 @@
 }
 
-bool readInt(const UChar* start, const UChar* end, const UChar** tokenEnd, bool canHaveLeadingZeros)
+template<typename CodeUnit>
+bool readInt(const CodeUnit* start, const CodeUnit* end, const CodeUnit** tokenEnd, bool canHaveLeadingZeros)
 {
     if (start == end)
@@ -95 +97 @@
 }
 
-bool parseNumberToken(const UChar* start, const UChar* end, const UChar** tokenEnd)
+template<typename CodeUnit>
+bool parseNumberToken(const CodeUnit* start, const CodeUnit* end, const CodeUnit** tokenEnd)
 {
     // We just grab the number here. We validate the size in DecodeNumber.
@@ -102 +105 @@
         return false;
 
-    UChar c = *start;
+    CodeUnit c = *start;
     if ('-' == c)
         ++start;
@@ -146 +149 @@
 }
 
-bool readHexDigits(const UChar* start, const UChar* end, const UChar** tokenEnd, int digits)
+template<typename CodeUnit>
+bool readHexDigits(const CodeUnit* start, const CodeUnit* end, const CodeUnit** tokenEnd, int digits)
 {
     if (end - start < digits)
@@ -160 +164 @@
 }
 
-bool parseStringToken(const UChar* start, const UChar* end, const UChar** tokenEnd)
+template<typename CodeUnit>
+bool parseStringToken(const CodeUnit* start, const CodeUnit* end, const CodeUnit** tokenEnd)
 {
     while (start < end) {
-        UChar c = *start++;
+        CodeUnit c = *start++;
         if ('\\' == c && start < end) {
             c = *start++;
@@ -198 +203 @@
 }
 
-Token parseToken(const UChar* start, const UChar* end, const UChar** tokenStart, const UChar** tokenEnd)
+template<typename CodeUnit>
+Token parseToken(const CodeUnit* start, const CodeUnit* end, const CodeUnit** tokenStart, const CodeUnit** tokenEnd)
 {
     while (start < end && isSpaceOrNewline(*start))
@@ -262 +268 @@
 }
 
-bool decodeString(const UChar* start, const UChar* end, StringBuilder& output)
+template<typename CodeUnit>
+bool decodeString(const CodeUnit* start, const CodeUnit* end, StringBuilder& output)
 {
     while (start < end) {
@@ -317 +324 @@
 }
 
-bool decodeString(const UChar* start, const UChar* end, String& output)
+template<typename CodeUnit>
+bool decodeString(const CodeUnit* start, const CodeUnit* end, String& output)
 {
     if (start == end) {
@@ -336 +344 @@
 }
 
-RefPtr<JSON::Value> buildValue(const UChar* start, const UChar* end, const UChar** valueTokenEnd, int depth)
+template<typename CodeUnit>
+RefPtr<JSON::Value> buildValue(const CodeUnit* start, const CodeUnit* end, const CodeUnit** valueTokenEnd, int depth)
 {
     if (depth > stackLimit)
@@ -342 +351 @@
 
     RefPtr<JSON::Value> result;
-    const UChar* tokenStart;
-    const UChar* tokenEnd;
+    const CodeUnit* tokenStart;
+    const CodeUnit* tokenEnd;
     Token token = parseToken(start, end, &tokenStart, &tokenEnd);
     switch (token) {
@@ -507 +516 @@
 RefPtr<Value> Value::parseJSON(const String& json)
 {
-    // FIXME: This whole file should just use StringView instead of UChar/length and avoid upconverting.
-    auto characters = StringView(json).upconvertedCharacters();
-    const UChar* start = characters;
-    const UChar* end = start + json.length();
-    const UChar* tokenEnd;
-    auto result = buildValue(start, end, &tokenEnd, 0);
-    if (!result)
-        return nullptr;
-
-    for (const UChar* valueEnd = tokenEnd; valueEnd < end; ++valueEnd) {
-        if (!isSpaceOrNewline(*valueEnd))
+    auto containsNonSpace = [] (const auto* begin, const auto* end) {
+        if (!begin)
+            return false;
+        for (const auto* it = begin; it < end; it++) {
+            if (!isSpaceOrNewline(*it))
+                return true;
+        }
+        return false;
+    };
+
+    RefPtr<Value> result;
+    if (json.is8Bit()) {
+        const LChar* start = json.characters8();
+        const LChar* end = start + json.length();
+        const LChar* tokenEnd { nullptr };
+        result = buildValue(start, end, &tokenEnd, 0);
+        if (containsNonSpace(tokenEnd, end))
             return nullptr;
-    }
-
+    } else {
+        const UChar* start = json.characters16();
+        const UChar* end = start + json.length();
+        const UChar* tokenEnd { nullptr };
+        result = buildValue(start, end, &tokenEnd, 0);
+        if (containsNonSpace(tokenEnd, end))
+            return nullptr;
+    }
     return result;
 }