Enable the runtime to emit a warning message when a mapped variable does not have an appropriate device counterpart that can be used. The current patch uses the use_device_addr as an example but other cases may be possible. The patch has been tested with the warning as an error to make sure that it does not disturb any of the existing tests. The warning currently uses MESSAGE but can be changed to REPORT if reviewers think it would be more appropriate.
I am not sure if we should generate this message. The spec says "Each list item must have a corresponding list item in the device data environment or be accessible on the target device."
The pointer may be in the unified_shared_memory.
The message seems useful, accidentally putting x there rather than x[:0] like the user probably intended is likely to be a relatively common mistake unfortunately. Would it be feasible to, when the message would be printed for a pointer-type variable, check if it's target is accessible and suggest the appropriate fix, or at least say "maybe you meant to use an array section, the array section target is valid"? That would make it immensely easier for an lower experience user to take action on.
I would agree to that sentiment. If the runtime can easily detect the issue and report, it should do so. Just letting the code die with a SEGFAULT is certainly fun for folks who love to debug, but is not really productive.