1 files changed, 35 insertions, 0 deletions
diff --git a/admin/notes/exit-value b/admin/notes/exit-value
new file mode 100644
index 00000000000..e59bb614b9b
--- /dev/null
+++ b/admin/notes/exit-value
@@ -0,0 +1,35 @@
+ttn 2004-05-09
+
+The exit value of a program returning to the shell on unixoid systems is
+typically 0 for success, and non-0 (such as 1) for failure.  For vms it is
+odd (1,3,5...) for success, even (0,2,4...) for failure.
+
+This holds from the point of view of the "shell" (in quotes because vms has a
+different dispatch model that is not explained further here).
+
+From the point of view of the program, nowadays stdlib.h on both type of
+systems provides macros `EXIT_SUCCESS' and `EXIT_FAILURE' that should DTRT.
+
+NB: The numerical values of these macros DO NOT need to fulfill the the exit
+value requirements outlined in the first paragraph!  That is the job of the
+`exit' function.  Thus, this kind of construct shows misunderstanding:
+
+   #ifdef VMS
+      exit (1);
+   #else
+      exit (0);
+   #endif
+
+Values aside from EXIT_SUCCESS and EXIT_FAILURE are tricky.
+
+
+
+ttn 2004-05-12
+
+Values aside from EXIT_SUCCESS and EXIT_FAILURE can be used to indicate
+finer gradations of failure.  If this is the only information available
+to the caller, clamping such values to EXIT_FAILURE loses information.
+If there are other ways to indicate the problem to the caller (such as
+a message to stderr) it may be ok to clamp.  In all cases, it is the
+relationship between the program and its caller that must be examined.
+[Insert ZAMM quote here.]