Fix wrong *length

[gvanem]

A highsize should be the upper 32-bit of a signed 64-bit value.
But nobody have ever tried this function with a >2GByte .log-file before (?)

[nnposter]

I am in agreement that the current arithmetic is unsafe and that your code would fix it. That said, I wonder whether it should not be addressed in a more modern way, which avoids the arithmetic altogether, such as:

--- a/utils.cc
+++ b/utils.cc
@@ -595,7 +595,7 @@
 char *mmapfile(char *fname, s64 *length, int openflags) {
   HANDLE fd;
   DWORD mflags, oflags;
-  DWORD lowsize, highsize;
+  LARGE_INTEGER filesize;
   char *fileptr;
 
   if (!length || !fname) {
@@ -622,11 +622,10 @@
   if (!fd)
     pfatal ("%s(%u): CreateFile()", __FILE__, __LINE__);
 
-  lowsize = GetFileSize (fd, &highsize);
-  if (lowsize == INVALID_FILE_SIZE && GetLastError() != NO_ERROR) {
-    pfatal("%s(%u): GetFileSize(), file '%s'", __FILE__, __LINE__, fname);
+  if (!GetFileSizeEx(fd, &filesize)) {
+    pfatal("%s(%u): GetFileSizeEx(), file '%s'", __FILE__, __LINE__, fname);
   }
-  *length = lowsize + highsize << sizeof(DWORD);
+  *length = (s64)filesize.QuadPart;
   if (*length < 0) {
     fatal("%s(%u): size too large, file '%s'", __FILE__, __LINE__, fname);
   }

Thoughts?

[gvanem]

Thoughts?

A GetFileSizeEx() is fine with me.

[nnposter]

Committed as r38219.
Thank you for identifying the issue and proposing a fix.