Fix PR_ImplodeTime for Linux x64.

This fixes the following error when building for x64 targets:

base/third_party/nspr/prtime.cc: In function 'PRTime PR_ImplodeTime(const 
PRExplodedTime*)':
base/third_party/nspr/prtime.cc:173: error: integer overflow in expression
base/third_party/nspr/prtime.cc:176: error: integer overflow in expression

PRTime is long long (8 bytes) on ia32, and long (8 bytes) on x64.  On ia32,
LONG_MAX (4 bytes) converted to microseconds fits in the PRTime type, but
on x64 LONG_MAX is 8 bytes and so overflows the PRTime type on conversion.

Avoid these issues by only returning INT_MAX.  On ia32 this is the correct
value, and does not change behaviour.  On x64 the call to timegm() will always
succeed due to the increased range of time_t, so the error condition of -1
should never be reached.

Fix a bug in the conditional expression for 1 second before epoch.

Patch by Joel Stanley

Original review URL: http://codereview.chromium.org/160511

R=dean,wtc
BUG=18231
TEST=new unit test
Review URL: http://codereview.chromium.org/164366

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@23186 0039d316-1c4b-4281-b951-d872f2087c98

1 files changed, 11 insertions, 0 deletions

diff --git a/base/pr_time_unittest.cc b/base/pr_time_unittest.cc
index 416094c..6dda299 100644
--- a/base/pr_time_unittest.cc
+++ b/base/pr_time_unittest.cc
@@ -205,6 +205,17 @@ TEST_F(PRTimeTest, ParseTimeTestEpochNeg1) {
   EXPECT_EQ(-1, parsed_time.ToTimeT());
 }
 
+// If time_t is 32 bits, a date after year 2038 will overflow time_t and
+// cause timegm() to return -1.  The parsed time should not be 1 second
+// before epoch.
+TEST_F(PRTimeTest, ParseTimeTestEpochNotNeg1) {
+  Time parsed_time;
+
+  EXPECT_EQ(true, Time::FromString(L"Wed Dec 31 23:59:59 GMT 2100",
+                                   &parsed_time));
+  EXPECT_NE(-1, parsed_time.ToTimeT());
+}
+
 TEST_F(PRTimeTest, ParseTimeTestEpochNeg2) {
   Time parsed_time;