Discussion on the new Let block

[jens]

We can optimize performance for multiple declarations by caching the setter once and repeatedly calling it, instead of reifying a new setter for each declaration tuple:

This makes it somewhat faster...

[dardoro]

@jens

This makes it somewhat faster...

Yes.
It's interesting that "this(caller)" degrades when contexts are embedded.
This is a microbenchmark result for

1. "this( caller)"
2. "run()"
3. "run( this( caller))"
   
   

Reasonable result for case 3, as a straight composition of 1&3, should be 150-200 ms.
Of course it may be meanningles, as microbenchmark results fluctuate heavily.

Also, a full copy of the process at the start is very time-consuming. The same test with "Live coding..."

[qw23]

Jens’ new solution (let (2)) is, indeed, much faster than the original, e.g. for a let with 5 variables, as demonstrated below.

I found another quirk regarding run time, also demonstrated below: the library’s original let block takes more than linear extra runtime for each extra variable!

Edit
Initiallly I had the impression the new let was like Scheme's let*, i.e. something like would work. Later I found it doesn't. Or perhaps I’ve been doing something wrong?

[mark4sisb]

@qw23

Initiallly I had the impression the new let was like Scheme's let*, i.e. something like would work. Later I found it doesn't. Or perhaps I’ve been doing something wrong?

The issue you are having is that all inputs are evaluated before the block is run. This can be fixed by changing the second input type to "any (unevaluated)" and calling it in the block:

[qw23]

Great! Combining @mark4sisb’s brilliant (!) proposal with @jens’s post #21, and my own really simple version of let (working title: let-), we can now define the more efficient:

It may still be over 10 times slower than separate script variables + set but on the other hand it enables really compact, yet easily readable, code for declaring and initializing variables.

BTW definition of let-

Edit
… or is this actually even the equivalent of letrec*?

[loucheman]

Can you do this with only one block (just a let block not let* & let-) ?

[cymplecy]

I think so

Note @qwe has a preferred approach to make things look neat and tidy so they are easier to understand

[loucheman]

Thk, i will test it soon

[loucheman]

work perfectly, thk to both of you