*Repeat and non-zero return codes

*Repeat’s handling of non-zero return codes is very broken. Should it stop or should it continue? Should a non-zero return code cause *Repeat itself to exit with a non-zero return code?

For commands that exit with errors, the behaviour is pretty clear: If the -continue flag isn’t specified, Repeat will stop, print the error (or report via Wimp_ReportError), and then exit with a return code of 1.

However for commands that exit with non-zero return codes, things are less certain. system (actually, _kernel_system) only stores an error if the command exits via the error environment handler, so the error *Repeat gets from _kernel_last_oserror will be undefined. It could be null, which will then cause *Repeat to stop and exit with a zero return code, or if -continue is used, will cause it to crash with a null pointer dereference. If it’s non-null (and -continue isn’t used), *Repeat will stop and exit with a non-zero return code, but the error it reports will be nonsense.

An extra wrinkle is that because system makes calls to DDEUtils to manage long command lines, if DDEUtils isn’t loaded, _kernel_last_oserror will always have a “SWI &4258B not found” error stored, so commands which exit with a non-zero return code will always cause Repeat to find that error and report it. So machines with DDEUtils vs. without will get differing behaviour from Repeat when commands exit with non-zero return codes.

I would have thought that *Repeat should only stop on errors; for non-zero return codes the user can choose an appropriate Sys$RCLimit so that an error is generated.

I would think that after 20+ years the current behaviour is set. If you want to change the behaviour add a second API.

Repeat’s inconsistent behaviour is responsible for an immense amount of grief, when trying to diagnose incomplete booting, due to something failing in PreDesk.

I would prefer this to be fixed, as any change in behaviour will generally only become apparent when something is already broken.

The *Repeat documentation doesn’t mention the “-continue” parameter.

Terminate and report the file that returned a non-zero, exiting with said value? I’d expect -continue to ignore all errors, including non-zero returns. Add a specific error so it can be distinguished.

Q if system is dependent on DDEUtils, does a softloaded CLib auto-load the Module? I don’t remember it being a dependency.