[Haskell-cafe] "back doors" into the IO monad

[Claude Heiland-Allen]

On 23/10/10 23:28, Manlio Perillo wrote:
> Hi.
>
> What are the available methods to execute IO actions from pure code?
>
> I know only unsafePerformIO and foreign import (to call a non pure
> foreign function).
>
>
> Assuming I want to execute external untrusted code using plugins (via
> the `plugins` package), is it possible to completely forbid execution of
> IO actions from user code?

Even pure code can cause denial of service through huge memory 
consumption or infinite loop.  Perhaps it would be wise to look at how 
'lambdabot'/'mueval' work, using OS-level security limits to restrict 
amount of memory and time that can be used by untrusted code (which 
already has had its IO priviledges revoked, see their source codes for 
details).

> Thanks   Manlio


Claude