Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
> The thing is that unlike TCL we have one interpreter for both trusted
> and untrusted cases.
> My thinking is to factor out all the code that only applies to trusted
> cases from the interpreter init code, and only call it if we try to
> compile a trusted function and it hasn't been run yet. Does that seem
> reasonable?
That would work. You'd need two state flags instead of just one, but
that doesn't seem bad.
Recovering when you get an error is probably the trickiest part of this.
regards, tom lane