At 01:09 PM 3/21/2003, Jeff Trawick wrote:
>I'm a bit nervous about the existing/new special handling for descriptors 0-2.
Ok, here is a patch that uses the target file's inherit and noclose
flag bits for apr_file_dup2(), and leaves apr_file_dup() results as
never inherited and always registering a cleanup.
Would this patch satisfy your concerns? Does this patch resolve
the new misbehavior?
Bill
Index: file_io/unix/filedup.c
===================================================================
RCS file: /home/cvs/apr/file_io/unix/filedup.c,v
retrieving revision 1.60
diff -u -r1.60 filedup.c
--- file_io/unix/filedup.c 19 Mar 2003 10:17:26 -0000 1.60
+++ file_io/unix/filedup.c 21 Mar 2003 17:31:58 -0000
@@ -117,16 +117,25 @@
/* make sure unget behavior is consistent */
(*new_file)->ungetchar = old_file->ungetchar;
- /* apr_file_dup() clears the inherit attribute for normal files,
- * but sets the inherit attribute for std[out,in,err] fd's.
- * The user must call apr_file_inherit_[un]set() on the dupped
- * apr_file_t when desired.
- */
- if ((*new_file)->filedes <= 2) {
- (*new_file)->flags = old_file->flags | APR_INHERIT;
+ if (which_dup == 1) {
+ /* apr_file_dup() retains all old_file flags with the exceptions
+ * of APR_INHERIT and APR_FILE_NOCLEANUP.
+ * The user must call apr_file_inherit_set() on the dupped
+ * apr_file_t when desired.
+ */
+ (*new_file)->flags = old_file->flags
+ & ~(APR_INHERIT | APR_FILE_NOCLEANUP);
+ }
}
- else {
- (*new_file)->flags = old_file->flags & ~APR_INHERIT;
+ else /* which_dup == 2 */ {
+ /* apr_file_dup2() must respect the original settings for the
+ * new_file->flags inheritence. If there is no cleanup, we
+ * should be finished already, otherwise fall through to the
+ * apr_pool_cleanup_register
+ */
+ if ((*new_file)->flags & APR_FILE_NOCLEANUP) {
+ return APR_SUCCESS;
+ }
}
apr_pool_cleanup_register((*new_file)->pool, (void *)(*new_file),
