As noted by Per Cederqvist, new_buffersize() sometimes returns the

buffer increment, and sometimes the new buffer size.  Make it do what
its name says, and fix the one place where this matters to the caller.

Also add a comment explaining why we call lseek() and then ftell().

diff --git a/Objects/fileobject.c b/Objects/fileobject.c
index d3b9894cae885980d76bcfd67245d6982578e094..b9fa0fc11465c2e3de867a82b6efea87f1efdc39 100644 (file)
--- a/Objects/fileobject.c
+++ b/Objects/fileobject.c
@@ -422,13 +422,22 @@ new_buffersize(f, currentsize)
        struct stat st;
        if (fstat(fileno(f->f_fp), &st) == 0) {
                end = st.st_size;
+               /* The following is not a bug: we really need to call lseek()
+                  *and* ftell().  The reason is that some stdio libraries
+                  mistakenly flush their buffer when ftell() is called and
+                  the lseek() call it makes fails, thereby throwing away
+                  data that cannot be recovered in any way.  To avoid this,
+                  we first test lseek(), and only call ftell() if lseek()
+                  works.  We can't use the lseek() value either, because we
+                  need to take the amount of buffered data into account.
+                  (Yet another reason why stdio stinks. :-) */
                pos = lseek(fileno(f->f_fp), 0L, SEEK_CUR);
                if (pos >= 0)
                        pos = ftell(f->f_fp);
                if (pos < 0)
                        clearerr(f->f_fp);
                if (end > pos && pos >= 0)
-                       return end - pos + 1;
+                       return currentsize + end - pos + 1;
                /* Add 1 so if the file were to grow we'd notice. */
        }
 #endif
@@ -482,7 +491,7 @@ file_read(f, args)
                if (bytesread < buffersize)
                        break;
                if (bytesrequested < 0) {
-                       buffersize = bytesread + new_buffersize(f, buffersize);
+                       buffersize = new_buffersize(f, buffersize);
                        if (_PyString_Resize(&v, buffersize) < 0)
                                return NULL;
                }