At 01:27 AM 3/15/2007 -0500, William A. Rowe, Jr. wrote:
>Jonathan Gilbert wrote:
[snip]
>> Index: file_io/win32/readwrite.c
>> ===================================================================
>> --- file_io/win32/readwrite.c (revision 515905)
>> +++ file_io/win32/readwrite.c (working copy)
>> @@ -326,9 +326,60 @@
>> rv = APR_SUCCESS;
>> }
>> else {
>> - (*nbytes) = 0;
>> rv = apr_get_os_error();
>> - if (rv == APR_FROM_OS_ERROR(ERROR_IO_PENDING)) {
>> + if (rv == APR_FROM_OS_ERROR(ERROR_INSUFFICIENT_BUFFER)) {
[snip]
>> + }
>> + else if (rv == APR_FROM_OS_ERROR(ERROR_IO_PENDING)) {
[snip]
>> }
>> +
>> + (*nbytes) = 0;
>
>We can't return 0 bytes written, too many apps will presume 0 means file
>closed, other signals tripped, error etc. We should do at least one
>short-write before returning in either case.
The return of 0 bytes written is only if it actually returns an error
state. The ERROR_INSUFFICIENT_BUFFER branch of the 'if' is, in all cases,
exited with a 'return', so that assignment to nbytes does not get executed
(that's why it was moved down there). The assignment is part of existing
behaviour that was present before I made the patch.
Jonathan Gilbert
