Macro system done in theory
too afraid to begin debugging, resting for a moment
This commit is contained in:
@@ -6,13 +6,13 @@ Reference loops are resource leaks. There are two primary ways to avoid referenc
|
||||
- Constants reference their constituent Expressions
|
||||
- Expressions reference Atoms
|
||||
- During evaluation, Constants replace their unbound names with Constants
|
||||
- There is a reference cycle here, but it always goes through a Constant.
|
||||
> **todo** A potential fix may be to update all Constants to point to a dummy value before freeing Trees
|
||||
- There is a reference cycle here, but it always goes through a Constant.
|
||||
> **todo** A potential fix may be to update all Constants to point to a dummy value before freeing Trees
|
||||
- Atoms reference the Systems that implement them
|
||||
- Atoms may reference Expressions that are not younger than them
|
||||
- This link is managed by the System but tied to Atom and not System lifecycle
|
||||
- Atoms can technically be applied to themselves, but it's a copying apply so it probably isn't a risk factor
|
||||
- This link is managed by the System but tied to Atom and not System lifecycle
|
||||
- Atoms can technically be applied to themselves, but it's a copying apply so it probably isn't a risk factor
|
||||
- Systems reference the Extension that contains them
|
||||
- Extensions reference the Port that connects them
|
||||
- The Extension signals the remote peer to disconnect on drop
|
||||
- The port is also referenced in a loose receiver thread, which always eventually tries to find the Extension or polls for ingress so it always eventually exits after the Extension's drop handler is called
|
||||
- The Extension signals the remote peer to disconnect on drop
|
||||
- The port is also referenced in a loose receiver thread, which always eventually tries to find the Extension or polls for ingress so it always eventually exits after the Extension's drop handler is called
|
||||
|
||||
@@ -17,11 +17,11 @@ Priority numbers are written in hexadecimal normal form to avoid precision bugs,
|
||||
- **32-39**: Binary operators, in inverse priority order
|
||||
- **80-87**: Expression-like structures such as if/then/else
|
||||
- **128-135**: Anything that creates lambdas
|
||||
Programs triggered by a lower priority pattern than this can assume that all names are correctly bound
|
||||
Programs triggered by a lower priority pattern than this can assume that all names are correctly bound
|
||||
- **200**: Aliases extracted for readability
|
||||
The user-accessible entry points of all macro programs must be lower priority than this, so any arbitrary syntax can be extracted into an alias with no side effects
|
||||
The user-accessible entry points of all macro programs must be lower priority than this, so any arbitrary syntax can be extracted into an alias with no side effects
|
||||
- **224-231**: Integration; documented hooks exposed by a macro package to allow third party packages to extend its functionality
|
||||
The `statement` pattern produced by `do{}` blocks and matched by `let` and `cps` is a good example of this. When any of these are triggered, all macro programs are in a documented state.
|
||||
The `statement` pattern produced by `do{}` blocks and matched by `let` and `cps` is a good example of this. When any of these are triggered, all macro programs are in a documented state.
|
||||
- **248-255**: Transitional states within macro programs get the highest priority
|
||||
|
||||
The numbers are arbitrary and up for debate. These are just the ones I came up with when writing the examples.
|
||||
|
||||
@@ -42,9 +42,9 @@ Prioritised macro patterns must start and end with a vectorial placeholder. They
|
||||
Macros are checked from the outermost block inwards.
|
||||
|
||||
1. For every name token, test all named macros starting with that name
|
||||
1. If the tail is implicit, continue iterating
|
||||
1. If the tail is implicit, continue iterating
|
||||
2. Test all prioritized macros
|
||||
1. Take the first rule that matches in the highest prioritized block
|
||||
1. Take the first rule that matches in the highest prioritized block
|
||||
|
||||
Test all in a set of macros
|
||||
1. Take the first rule that matches in each block
|
||||
@@ -75,26 +75,26 @@ Recursion has to happen through the interpreter itself, so the macro system is d
|
||||
- line parser `macro` parses a macro with the existing logic
|
||||
- atom `MacRecurState` holds the recursion state
|
||||
- function `resolve_recur` finds all matches on a MacTree
|
||||
- type: `MacRecurState -> MacTree -> MacTree`
|
||||
- use all relevant macros to find all matches in the tree
|
||||
- since macros must contain a locally defined token, it can be assumed that at the point that a constant is evaluated and all imports in the parent module have been resolved, necessarily all relevant macro rules must have been loaded
|
||||
- for each match
|
||||
- check for recursion violations
|
||||
- wrap the body in iife-s corresponding to the named values in the match state
|
||||
- emit a recursive call to process and run the body, and pass the same recursive call as argument for the macro to use
|
||||
- type: `MacRecurState -> MacTree -> MacTree`
|
||||
- use all relevant macros to find all matches in the tree
|
||||
- since macros must contain a locally defined token, it can be assumed that at the point that a constant is evaluated and all imports in the parent module have been resolved, necessarily all relevant macro rules must have been loaded
|
||||
- for each match
|
||||
- check for recursion violations
|
||||
- wrap the body in iife-s corresponding to the named values in the match state
|
||||
- emit a recursive call to process and run the body, and pass the same recursive call as argument for the macro to use
|
||||
```
|
||||
(\recur. lower (recur $body) recur)
|
||||
(resolve_recur $mac_recur_state)
|
||||
```
|
||||
- emit a single call to `instantiate_tpl` which receives all of these
|
||||
- emit a single call to `instantiate_tpl` which receives all of these
|
||||
- function `instantiate_tpl` inserts `MacTree` values into a `MacTree(tpl)`
|
||||
- type: `MacTree(tpl) [-> MacTree] -> MacTree`
|
||||
- type: `MacTree(tpl) [-> MacTree] -> MacTree`
|
||||
_this function deduces the number of arguments from the first argument. This combines poorly with autocurry, but it's an easy way to avoid representing standalone tree lists_
|
||||
- walks the tree to find max template slot number, reads and type checks as many template values
|
||||
- returns the populated tree
|
||||
- walks the tree to find max template slot number, reads and type checks as many template values
|
||||
- returns the populated tree
|
||||
- function `resolve` is the main entry point of the code
|
||||
- type: `MacTree -> MacTree`
|
||||
- invokes `resolve_recur` with an empty `MacRecurState`
|
||||
- type: `MacTree -> MacTree`
|
||||
- invokes `resolve_recur` with an empty `MacRecurState`
|
||||
- function `lower` is the main exit point of the code
|
||||
- type: `MacTree -> any`
|
||||
- Lowers `MacTree` into the equivalent `Expr`.
|
||||
- type: `MacTree -> any`
|
||||
- Lowers `MacTree` into the equivalent `Expr`.
|
||||
Reference in New Issue
Block a user