Allow efficient WebURL -> GURL conversions

This CL adds a constructor to GURL that takes a std::string instead of a char*
and a length. This constructor allows Blink to produce a temporary std::string
containing the URL's spec that we can move directly into the GURL, avoiding a
copy.

Once I update Blink, this constructor will help us avoid spending 5% of our
total time rendering http://sina.com.cn on KURL -> GURL conversions.

R=brettw
BUG=261412

Review URL: https://chromiumcodereview.appspot.com/20127002

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@213885 0039d316-1c4b-4281-b951-d872f2087c98

2 files changed, 25 insertions, 2 deletions

diff --git a/url/gurl.cc b/url/gurl.cc
index 3dd8463..09b8abb 100644
--- a/url/gurl.cc
+++ b/url/gurl.cc
@@ -116,6 +116,19 @@ GURL::GURL(const char* canonical_spec, size_t canonical_spec_len,
       is_valid_(is_valid),
       parsed_(parsed),
       inner_url_(NULL) {
+  InitializeFromCanonicalSpec();
+}
+
+GURL::GURL(std::string canonical_spec,
+           const url_parse::Parsed& parsed, bool is_valid)
+    : is_valid_(is_valid),
+      parsed_(parsed),
+      inner_url_(NULL) {
+  spec_.swap(canonical_spec);
+  InitializeFromCanonicalSpec();
+}
+
+void GURL::InitializeFromCanonicalSpec() {
   if (is_valid_ && SchemeIsFileSystem()) {
     inner_url_ =
         new GURL(spec_.data(), parsed_.Length(), *parsed_.inner_parsed(), true);
@@ -127,9 +140,9 @@ GURL::GURL(const char* canonical_spec, size_t canonical_spec_len,
   // and we can't always canonicalize then reproducabely.
   if (is_valid_) {
     url_parse::Component scheme;
-    if (!url_util::FindAndCompareScheme(canonical_spec, canonical_spec_len,
+    if (!url_util::FindAndCompareScheme(spec_.data(), spec_.length(),
                                         "filesystem", &scheme) ||
-        scheme.begin == parsed.scheme.begin) {
+        scheme.begin == parsed_.scheme.begin) {
       // We can't do this check on the inner_url of a filesystem URL, as
       // canonical_spec actually points to the start of the outer URL, so we'd
       // end up with infinite recursion in this constructor.
diff --git a/url/gurl.h b/url/gurl.h
index 6e58249..da5f736 100644
--- a/url/gurl.h
+++ b/url/gurl.h
@@ -41,6 +41,14 @@ class URL_EXPORT GURL {
   // information associated with the URL, which must be correct and consistent.
   GURL(const char* canonical_spec, size_t canonical_spec_len,
        const url_parse::Parsed& parsed, bool is_valid);
+  // Notice that we take the canonical_spec by value so that we can convert
+  // from WebURL without copying the string. When we call this constructor
+  // we pass in a temporary std::string, which lets the compiler skip the
+  // copy and just move the std::string into the function argument. In the
+  // implementation, we use swap to move the data into the GURL itself,
+  // which means we end up with zero copies.
+  GURL(std::string canonical_spec,
+       const url_parse::Parsed& parsed, bool is_valid);
 
   ~GURL();
 
@@ -335,6 +343,8 @@ class URL_EXPORT GURL {
   }
 
  private:
+  void InitializeFromCanonicalSpec();
+
   // Returns the substring of the input identified by the given component.
   std::string ComponentString(const url_parse::Component& comp) const {
     if (comp.len <= 0)