On 12/01/2026 3:56 AM, Mike Parker wrote:
Bastiaan recalled an initiative for nonallocating exceptions and wondered what the story was. Walter said there was a compiler switch that enabled them and it did work, but there were some issues with it he never fully understood. I said Razvan could probably explain it, and that I recalled it had to do with stack trace generation. Walter said that sounded like that was probably the issue.
Non-allocating would not be DIP1008, it would've been my value typed exceptions. DIP1008 allocated with malloc and then RC'd.
Sadly they wouldn't have worked, due to the architecture of the compiler wrt. inferred attributes.
You would've had to manually annotate them and it would be too frequent to be usable.
As far as I know, the design was still good, but... still DOA currently. So I'm in the process of trying to find a new solution to exceptions. Without the need for platform/target specific stack unwinding code.
