[e-lang] Safe compile-time evaluation

[Mark S. Miller]

Kevin Reid wrote:
> Another issue with constant folding:
> 
> Is it allowable for an E compiler to hang due to code like:
> 
>    if (false_but_not_statically_known) {
>      __loop(thunk implements DeepFrozen { true })
>    }

No.


> (Exempting __loop is insufficient since user code might behave  
> similarly.)

Agreed.


> If not, how shall we avoid it? What I've thought of so far:
> 
> 1. Have a rubber-stamp auditor to mark objects[*] as allowing  
> constant-folding for calls to them. [...]
> 
> 2. Evaluate constant calls with an interpreter with execution time/ 
> step-count limits. (This smaller-than-vat-granularity [...]

Why smaller than vat granularity? The compiler could spawn a vat to do 
constant folding. Any time a constant fold takes too long, the compiler kills 
that vat, fails to fold that constant, possibly logs a warning[1], spawns a 
new vat, and proceeds with the next attempted constant fold.


> [*] Given Selfless auditors, this can be made specific to a certain  
> subset of messages to an object.

Could you explain your footnote? I don't understand it. Thanks.

[1] The compiler should log this warning only to its platform's write-only log 
file, in order to avoid providing its clients a visible source of non-determinism.

-- 
Text by me above is hereby placed in the public domain

     Cheers,
     --MarkM