Don't create a Bigint object first for every integer while parsing the sources, instead first check…

vm/os_linux.cc

Index: vm/os_linux.cc
===================================================================
--- vm/os_linux.cc	(revision 9558)
+++ vm/os_linux.cc	(working copy)
@@ -161,6 +161,30 @@
 }
 
 
+bool OS::Strtoll(const char* str, int64_t* value) {
+  ASSERT(str != NULL && strlen(str) > 0 && value != NULL);
+  bool negative_value = false;
+  int32_t base = 10;
+  if (str[0] == '-') {
+    negative_value = true;
+    str += 1;
+  }
+  if ((strlen(str) > 2) && (str[0] == '0') &&

[sra1, 2012/07/11 21:46:43]

strlen scan not necessary since the conditions guard against  reading past the
NUL.  What you have is equivalent to:

if ((str[0] == '0') &&
    (str[1] == 'x' || str[1] == 'X') &&
    (str[2] != 0)) {

[siva, 2012/07/12 18:28:23]

Done.

+      (str[1] == 'x' || str[1] == 'X')) {
+    base = 16;
+  }
+  errno = 0;
+  *value = strtoll(str, NULL, base);

[sra1, 2012/07/11 21:46:43]

You should pass in &endptr and verify endptr != str && *endptr == 0.

[siva, 2012/07/12 18:28:23]

I am not sure I understand the value of this additional check, it is possible
for 
(endptr != str && *endptr == 0)
to be true and errno could still be ERANGE

We are not interested here in reporting why the function failed etc.

On 2012/07/11 21:46:43, sra1 wrote:

[sra1, 2012/07/12 20:12:15]

You don't get ERANGE or EINVAL for trailing junk.
The only way to detect lack of trailing junk is to check that *endptr == 0.

const char *str = '123X';
int base = 10;
...
const char *endptr;
int64_t value = stroll('123X', &endptr, base);
if (endptr == str || *endptr != 0) return false;  // necessary check.

[siva, 2012/07/16 19:45:47]

Good point, I have addressed this in a different CL.

On 2012/07/12 20:12:15, sra1 wrote:

+  if (errno == 0) {
+    if (negative_value) {
+      *value = -(*value);
+    }
+    return true;
+  }
+  return false;
+}
+
+
 void OS::PrintErr(const char* format, ...) {
   va_list args;
   va_start(args, format);