ginger loops
This commit is contained in:
parent
9ef363410f
commit
b95218e9d4
223
src/_posts/2021-04-27-loops-in-ginger.md
Normal file
223
src/_posts/2021-04-27-loops-in-ginger.md
Normal file
@ -0,0 +1,223 @@
|
|||||||
|
---
|
||||||
|
title: >-
|
||||||
|
Loops in Ginger
|
||||||
|
description: >-
|
||||||
|
Bringing it back around.
|
||||||
|
series: ginger
|
||||||
|
tags: tech
|
||||||
|
---
|
||||||
|
|
||||||
|
In previous posts in this series I went over the general idea of the ginger
|
||||||
|
programming language, and some of its properties. To recap:
|
||||||
|
|
||||||
|
* Ginger is a programming language whose syntax defines a directed graph, in the
|
||||||
|
same way that a LISP language's syntax defines nested lists.
|
||||||
|
|
||||||
|
* Graph edges indicate an operation, while nodes indicate a value.
|
||||||
|
|
||||||
|
* The special values `in` and `out` are used when interpreting a graph as a
|
||||||
|
function.
|
||||||
|
|
||||||
|
* A special node type, the tuple, is defined as being a node whose value is an
|
||||||
|
ordered set of input edges.
|
||||||
|
|
||||||
|
* Another special node type, the fork, is the complement to the tuple. A fork is
|
||||||
|
defined as being a node whose value is an ordered set of output edges.
|
||||||
|
|
||||||
|
* The special `if` operation accepts a 2-tuple, the first value being some state
|
||||||
|
value and the second being a tuple. The `if` operation expects to be directed
|
||||||
|
towards a 2-fork. If the boolean is true then the top output edge of the fork
|
||||||
|
is taken, otherwise the bottom is taken. The state value is what's passed to
|
||||||
|
the taken edge.
|
||||||
|
|
||||||
|
There were some other detail rules but I don't remember them off the top of my
|
||||||
|
head.
|
||||||
|
|
||||||
|
## Loops
|
||||||
|
|
||||||
|
Today I'd like to go over my ideas for how loops would work in ginger. With
|
||||||
|
loops established ginger would officially be a Turing complete language and,
|
||||||
|
given time and energy, real work could actually begin on it.
|
||||||
|
|
||||||
|
As with conditionals I'll start by establishing a base example. Let's say we'd
|
||||||
|
like to define an operation which prints out numbers from 0 up to `n`, where `n`
|
||||||
|
is given as an argument. In go this would look like:
|
||||||
|
|
||||||
|
```go
|
||||||
|
func printRange(n int) int {
|
||||||
|
for i := 0; i < n; i++ {
|
||||||
|
fmt.Println(i)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
```
|
||||||
|
|
||||||
|
With that established, let's start looking at different patterns.
|
||||||
|
|
||||||
|
## Goto
|
||||||
|
|
||||||
|
In the olden days the primary looping construct was `goto`, which essentially
|
||||||
|
teleports the program counter (aka instruction pointer) to another place in the
|
||||||
|
execution stack. Pretty much any other looping construct can be derived from
|
||||||
|
`goto` and some kind of conditional, so it's a good starting place when
|
||||||
|
considering loops in ginger.
|
||||||
|
|
||||||
|
```
|
||||||
|
(in -println-> } -incr-> out) -> println-incr
|
||||||
|
|
||||||
|
0 -> } -> } -if-> { -> out
|
||||||
|
in -> } -eq-> } { -> } -upd-> } -+
|
||||||
|
^ 0 -> } |
|
||||||
|
| println-incr -> } |
|
||||||
|
| |
|
||||||
|
+--------------------------------+
|
||||||
|
```
|
||||||
|
|
||||||
|
(Note: the `upd` operation is used here for convenience. It takes in three
|
||||||
|
arguments: A tuple, an index, and an operation. It applies the operation to the
|
||||||
|
tuple element at the given index, and returns a new tuple with that index set to
|
||||||
|
the value returned.)
|
||||||
|
|
||||||
|
Here `goto` is performed using a literal arrow going from the right to left.
|
||||||
|
it's ugly and hard to write, and would only be moreso the more possible gotos an
|
||||||
|
operation has.
|
||||||
|
|
||||||
|
It also complicates our graphs in a significant way: up till now ginger graphs
|
||||||
|
have have always been directed _acyclic_ graphs (DAGs), but by introducing this
|
||||||
|
construct we allow that graphs might be cyclic. It's not immediately clear to me
|
||||||
|
what the consequences of this will be, but I'm sure they will be great. If
|
||||||
|
nothign else it will make the compiler much more complex, as each value can no
|
||||||
|
longer be defined in terms of its input edge, as that edge might resolve back to
|
||||||
|
the value itself.
|
||||||
|
|
||||||
|
While conceptually sound, I think this strategy fails the practability test. We
|
||||||
|
can do better.
|
||||||
|
|
||||||
|
## While
|
||||||
|
|
||||||
|
The `while` construct is the basic looping primitive of iterative languages
|
||||||
|
(some call it `for`, but they're just lying to themselves).
|
||||||
|
|
||||||
|
Try as I might, I can't come up with a way to make `while` work with ginger.
|
||||||
|
`while` ultimately relies on scoped variables being updated in place to
|
||||||
|
function, while ginger is based on the concept of pipelining a set of values
|
||||||
|
through a series of operations. From the point of view of the programmer these
|
||||||
|
operations are essentially immutable, so the requirement of a variable which can
|
||||||
|
be updated in place cannot be met.
|
||||||
|
|
||||||
|
## Recur
|
||||||
|
|
||||||
|
This pattern is based on how many functional languages, for example erlang,
|
||||||
|
handle looping. Rather than introducing new primitives around looping, these
|
||||||
|
language instead ensure that tail calls are properly optimized and uses those
|
||||||
|
instead. So loops are implemented as recursive function calls.
|
||||||
|
|
||||||
|
For ginger to do this it would make sense to introduce a new special value,
|
||||||
|
`recur`, which could be used alongside `in` and `out` within operations. When
|
||||||
|
the execution path hits a `recur` then it gets teleported back to the `in`
|
||||||
|
value, with the input to `recur` now being the output from `in`. Usage of it
|
||||||
|
would look like:
|
||||||
|
|
||||||
|
```
|
||||||
|
(
|
||||||
|
|
||||||
|
(in -println-> } -incr-> out) -> println-incr
|
||||||
|
|
||||||
|
in -> } -if-> { -> out
|
||||||
|
in -eq-> } { -> } -upd-> } -> recur
|
||||||
|
0 -> }
|
||||||
|
println-incr -> }
|
||||||
|
|
||||||
|
) -> inner-op
|
||||||
|
|
||||||
|
0 -> } -inner-op-> out
|
||||||
|
in -> }
|
||||||
|
```
|
||||||
|
|
||||||
|
This looks pretty similar to the `goto` example overall, but with the major
|
||||||
|
difference that the looping body had to be wrapped into an inner operation. The
|
||||||
|
reason for this is that the outer operation only takes in one argument, `n`, but
|
||||||
|
the loop actually needs two pieces of state to function: `n` and the current
|
||||||
|
value. So the inner operation loops over these two pieces of state, and the
|
||||||
|
outer operation supplies `n` and an initial iteration value (`0`) to that inner
|
||||||
|
operation.
|
||||||
|
|
||||||
|
This seems cumbersome on the surface, but what other languages do (such as
|
||||||
|
erlang, which is the one I'm most familiar with) is to provide built-in macros
|
||||||
|
on top of this primitive which make it more pleasant to use. These include
|
||||||
|
function polymorphism and a more familiar `for` construct. With a decent macro
|
||||||
|
capability ginger could do the same.
|
||||||
|
|
||||||
|
The benefits here are that the graphs remain acyclic, and the syntax has not
|
||||||
|
been made more cumbersome. It follows conventions established by other
|
||||||
|
languages, and ensures the language will be capable of tail-recursion.
|
||||||
|
|
||||||
|
## Map/Reduce
|
||||||
|
|
||||||
|
Another functional strategy which is useful is that of the map/reduce power
|
||||||
|
couple. The `map` operation takes a sequence of values and an operation, and
|
||||||
|
returns a sequence of the same length where the operation has been applied to
|
||||||
|
each value in the original sequence individually. The `reduce` operation is more
|
||||||
|
complicated (and not necessary for out example), but it's essentially a
|
||||||
|
mechanism to turn a sequence of values into a single value.
|
||||||
|
|
||||||
|
For our example we only need `map`, plus one more helper operation: `range`.
|
||||||
|
`range` takes a number `n` and returns a sequence of numbers starting at `0` and
|
||||||
|
ending at `n-1`. Our print example now looks like:
|
||||||
|
|
||||||
|
```
|
||||||
|
in -range-> } -map-> out
|
||||||
|
println -> }
|
||||||
|
```
|
||||||
|
|
||||||
|
Very simple! Map/reduce is a well established pattern and is probably the
|
||||||
|
best way to construct functional programs. However, the question remains whether
|
||||||
|
these are the best _primitives_ for looping, and I don't believe they are. Both
|
||||||
|
`map` and `reduce` can be derived from conditional and looping primitives like
|
||||||
|
`if` and `recur`, and they can't do some things that those primitives can. While
|
||||||
|
|
||||||
|
|
||||||
|
I expect one of the first things which will be done in ginger is to define `map`
|
||||||
|
and `reduce` in terms of `if` and a looping primitive, and use them generously
|
||||||
|
throughout the code, I think the fact that they can be defined in terms of
|
||||||
|
lower-level primitives indicates that they aren't the right looping primitives
|
||||||
|
for ginger.
|
||||||
|
|
||||||
|
## Conclusion
|
||||||
|
|
||||||
|
Unlike with the conditionals posts, where I started out not really knowing what
|
||||||
|
I wanted to do with conditionals, I more or less knew where this post was going
|
||||||
|
from the beginning. `recur` is, in my mind, the best primitive for looping in
|
||||||
|
ginger. It provides the flexibility to be extended to any use-case, while not
|
||||||
|
complicating the structure of the language. While possibly cumbersome to
|
||||||
|
implement directly, `recur` can be used as a primitive to construct more
|
||||||
|
convenient looping operations like `map` and `reduce`.
|
||||||
|
|
||||||
|
As a final treat (lucky you!), here's `map` defined using `if` and `recur`:
|
||||||
|
|
||||||
|
```
|
||||||
|
(
|
||||||
|
in -0-> mapped-seq
|
||||||
|
in -1-> orig-seq
|
||||||
|
in -2-> op
|
||||||
|
|
||||||
|
mapped-seq -len-> i
|
||||||
|
|
||||||
|
mapped-seq -> } -if-> { -> out
|
||||||
|
orig-seq -len-> } -eq-> } { -> } -append-> } -> recur
|
||||||
|
i -> } } }
|
||||||
|
} }
|
||||||
|
orig-seq -i-> } -op-> } }
|
||||||
|
}
|
||||||
|
orig-seq -> }
|
||||||
|
op -> }
|
||||||
|
) -> inner-map
|
||||||
|
|
||||||
|
() -> } -inner-map-> out
|
||||||
|
in -0-> }
|
||||||
|
in -1-> }
|
||||||
|
```
|
||||||
|
|
||||||
|
The next step for ginger is going to be writing an actual implementation of the
|
||||||
|
graph structure in some other language (let's be honest, it'll be in go). After
|
||||||
|
that we'll need a syntax definition which can be used to encode/decode that
|
||||||
|
structure, and from there we can start actually implementing the language!
|
Loading…
Reference in New Issue
Block a user