R devel - Mar 2021 - An argument promise containing bytecode?

Dear all, 

I was playing around with some C code that inspects argument promises and I
noticed that forwarded arguments sometimes contain bytecode in the PRCODE
component. E.g. if I have a nested call like this:

  f <- function(x) x
  g <- function(x) f(x)

 g(1+1)

and inspect the value of ?x? in g?s frame from within f?s frame (basically
calling findVar(parent.frame(), x)), I would see a LANGSXP the first time g() 
is invoked, but BCODESXP on every subsequent time.  Using PREXPR gets me the
original LANGSXP in any case.

Now, this is curious behavior and I had a quick glance into R source code
related to closure evaluation and argument matching, but I could not see
anything that would generate bytecode for promises. Could someone with in-depth
knowledge of R?s machinery explain what is going on?

Thanks, 

Taras

Taras,

I don't think this has anything to do with promises, rather it is called
Just-In-Time (JIT) compilation - see ?enablleJIT in R, it is enabled by default,
so the function will be compiled on second use.

Cheers,
Simon


> On Mar 7, 2021, at 11:12 PM, Taras Zakharko <taras.zakharko at
uzh.ch> wrote:
> 
> Dear all, 
> 
> I was playing around with some C code that inspects argument promises and I
noticed that forwarded arguments sometimes contain bytecode in the PRCODE
component. E.g. if I have a nested call like this:
> 
>  f <- function(x) x
>  g <- function(x) f(x)
> 
> g(1+1)
> 
> and inspect the value of ?x? in g?s frame from within f?s frame (basically
calling findVar(parent.frame(), x)), I would see a LANGSXP the first time g() 
is invoked, but BCODESXP on every subsequent time.  Using PREXPR gets me the
original LANGSXP in any case.
> 
> Now, this is curious behavior and I had a quick glance into R source code
related to closure evaluation and argument matching, but I could not see
anything that would generate bytecode for promises. Could someone with in-depth
knowledge of R?s machinery explain what is going on?
> 
> Thanks, 
> 
> Taras
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>