Re: Getting rid of some more lseek() calls

On Thu, Feb 13, 2020 at 02:51:44PM +1300, Thomas Munro wrote:
> Ok, how about this?

Alvaro's point sounds sensible to me.  I like the approach you are
taking in 0001.  At least it avoids more issues with WIN32 and stat()
(I hope to work on that at some point, we'll see..).

+/*
+ * pg_file_size --- return the size of a file
+ */
+int64
+pg_file_size(int fd)
+{
This routine has nothing really dependent on the backend.  Would it
make sense to put it in a different place where it can be used by the
frontend?  The function should include at least a comment about why we
have a special path for Windows, aka not falling into the trap of the
4GB limit for stat().

The commit message of 0001 mentions pg_read(), and that should be
pg_pread().

There are two combinations of lseek/read that could be replaced: one
in pg_receivewal.c:FindStreamingStart(), and one in
SimpleXLogPageRead() for parsexlog.c as of pg_rewind.

Patch 0002 looks good to me.  This actually removes a confusion when
failing to seek the end of the file as the offset referenced to would
be 0.  Patch 0003 is also a very good thing.
--
Michael