chrisrink10 / mscript Goto Github PK

Allow programmers to safely reference potentially null values as in some.name?.maybenull() which would return null if some.name does not have member maybenull rather than throwing an error for attempting to access a null value.

Right now, the VM and Parser have separate error types which are pretty much the same. They both report whether an error occurred and then some further information. Regrettably the bytecode generator doesn't even have any error type since it started as a very simple series of routines that eventually evolved to have a lot of error conditions that are now ignored. I'd like to come up with a single error object and result type that can be used throughout the application to report errors.

Right now, globals are referenced in the VM by pushing expression values onto the data stack and passing an argument with the opcode indicating the number of data elements which are global nodes. I want to create a single data element as it's own VM value, but since global nodes can be any arbitrary expression, I'm not sure how that can be done elegantly.

This style would be like in Python, where return is equivalent to return None. In mscript, return would equal return null.

Still not sure if I want this, but maybe...

In Python, functions can be defined with variadic arguments using the collect into list operator (*ident) and keyword arguments using the collect into dict operator (**ident). Similarly, parameters in function definitions may specify default values if left unspecified. mscript should support a similar paradigm:

func Foo(a, *b, c=null, **d) { }

...could be called as:

Foo("a string", "another string", "a third string", c=1, some_dict_val=null)
//  ^a          ^b[0]             ^b[1]             ^c   ^**d

In iterator style for statements, the OPC_NEW_NAME opcode is pushed to the bytecode stack before pushing the iterator expression. This means that if the iterator name is the same as the for loop variable name, the loop will fail:

i = [1, 2, 3];
for var i in i {
    // we will literally never get here because
    // `var i` occurs before calling __next__ on
    // the iterator i
}

This works fine in Python:

i = [1, 2, 3]
for i in i:
    print(i)

prints

1
2
3

Related to issue #37

Allow programmers to return multiple values like return err, val

Allow programmers to create conditional expressions like (condition) ? (valIfTrue) : (valIfFalse) as in many other C-like languages.

The OPC_NEW_NAME opcode is pushed to the bytecode stack before pushing any existing values onto the data stack. This means that if a declared variable is the same name as any value in its initial expression, the value in the initial expression will be set to null:

var index := 1;
func test() {
    var index := index + 1; // index => null + 1
    return index;
}

In Go, programmers can write defer stmt to defer cleanup steps like closing files or database connections until exiting the current scope. This makes cleanup easier than the traditional methods like C's goto or lots of nested if statements.

Lua's interpreter seems to use a special opcode OP_TAILCALL to make a tail call.

Allow programmers to write assignment statements like a, b := b, a to swap a and b without a temporary variable

The code name = causes the interpreter to hang indefinitely and apparently consume an enormous amount of memory (memory consumption after hanging for a few mins):

==86835== HEAP SUMMARY:
==86835==     in use at exit: 1,047,424,811 bytes in 32,731,207 blocks
==86835==   total heap usage: 32,749,227 allocs, 18,020 frees, 1,049,981,757 bytes allocated```

Python does not use an incrementing for statement like mscript's for var i := begin : end : step { }. Instead, they provide a generic range function, which I think I prefer since it means we no longer have to maintain a separate syntactic form.

Only the lexer and parser know the line and column number of each symbol in code since they are not codified in the AST. This makes certain types of errors pretty vague in terms of error reporting:

> index;
mscript: reference to undefined variable `index`

As the title says, a var declaration inside of the for loop statement seems to trigger an error about builtin identifiers:

for var i := 1 : 10 {
    var x := 10;
}

The only composite data structure in mscript is the object for now (well... at least that was the plan). However, it'd be nice to have an array type too.

Definitely avoid the quagmire of this confusion that is JavaScript. Probably use an explicit self first parameter as in Lua and Python.

The grammar in parser.c actually suggests that this syntax should be allowed:

var first := "John", middle := "H", last := "Smith"

However, the actual parser quits after a single variable declaration. So it should allow that

The parser generates an AST for any input text which matches the grammar, but there are no checks for correct language semantics such as:

• break and continue statements may only appear in for statement blocks
• return statements may only appear in function blocks
• shadowing variables in containing lexical blocks (warning)
• redeclaring a variable name in the same lexical block

The lexer returns a string token containing the string a string as "a string" if that string appeared in the source with a " quotation or 'a string' if it appeared in the source with a ' character. This, of course, means that later consumers need to account for these quotes, which is just annoying and counterproductive.

For now, mscript has +=, -=, *=, /=, \=, and %= operators. There should be equivalent compound assignment operators for the standard bitwise ops too: &=, |=, ^=. Maybe <<= and >>= too?

I pulled the merge statement verbatim from M, but I'm not sure if it makes sense in mscript. It could still work, but I think there may be other better idioms.

Object literals can appear in code much the same way as dicts in Python or Objects in Javascript or tables in Lua:

{ key1: value1, key2: value2, ... }

Python uses the with ... as ...: statement to manage context for unmanaged resources and C# uses using (x = ...) { ... } for the same. mscript should include a context management protocol (probably the same way Python does with __enter__ and __exit__ methods). This probably fulfills the same role as the suggested defer statement referenced in issue #18.

The declarations (or assignments) used in for loops semantically occur within the loop block, but the current checks are performed as if they were outside of that block. This means we probably won't correctly catch shadowing and will incorrectly flag redeclarations within the same scope.

Similar to Go's go statement, I would like a statement (call it spawn) for now that creates a new thread and executes a function.

If a break or continue statement were to appear inside of a function definition inside of a for loop, they would have undefined semantics when the break or continue is actually hit in the bytecode. The verifier should check for this.

var fn;

for x in coll {
    fn := func() {
        continue;
    };
}

fn(); // => what happens here?

Variables are declared in their current lexical scope with the var keyword right now. For values known not to change, there should be a val keyword.