ark-lang / ark Goto Github PK

Instead of writing:

if (x == null) {

}

It would be cleaner to use a question mark, similar to other languages:

if (?x) {

}

The question mark would check if the right most object is null or not, if it's null the condition will be true and will execute.

Parse expressions like:

int x = (5 + 5) - (a / b);

No idea what I was thinking when last night when I implemented the shunting yard algorithm and a stack for Expressions!? 👎

Implementing pointers, yay or nay? If I were to implement pointers, I'd want them to be more distinguishable from the traditional syntax. Also, I don't want the asterisks since that would cause conflicts with multiplication.

int y = 50;
int ~x = &y;
x -> 12341234;
// or
~x = 123

Do it bich

I've written a VM, it still needs work and theres probably a lot of pointer misuse in there too (lol) but we'll compile to bytecode for the VM. I'm not sure if we should handle expressions to generate code to evaluate the expression or if we should evaluate the expression and generate code for that?
@CaptainChloride You might wanna check out the VM, see if you can fix any errors. Also no clue how to handle decimals, strings should be a bit easier if we go for the C style.
Also I want == to compare strings, not have to do the C style !strcmp shit.

got college work to do, cant do loads of work on the project for a few days.
@CaptainChloride @IanMurray

I have llvm 3.5 installed on Archlinux and I get error like: clang: error: linker command failed with exit code 1 (use -v to see invocation)

Here is the complete error message at http://pastebin.com/iGaVK3j7

It works when adding "-ldl -pthread" on line 22 in the Makefile. With -ltinfo, it doesn't work though.

This will make it easier to improve error messages

Need to fix that security-ish issue, add dynamic allocation for stack(?), finish adding instructions, etc, etc. @CaptainChloride something you can work on when you feel like it ;) 😉

Need to make it such that the file name can be specified before the arguments.

./j4 -o output.out tests/for.j4

Strings aren't being tokenized correctly, some characters will cause undefined behaviour.

fn just_a_test(string s): string {
    ret "this is a test";
}

The lexer will tokenize this correctly, and "this is a test" will be tokenized to Token("this is a test", STRING); However, this will not tokenize correctly:

fn just_a_test(string s): string {
    ret "this is a test !£ wow %!";
}
``

Need these before I can parse while, if, do, and other logical statements.

@CaptainChloride

we need to discuss this feature

i.e, given the following arguments...

 j4 tests/simple.j4 -o folder/test.j4e

The folder will not be created, and the file won't be created.

Just tried building the compiler on my setup (OS X 10.10), ran into some issues:

1. Why are you using C99 instead of C11?
   1b. Why are you doing some weird enum bool thing? <stdbool.h> exists for this purpose
2. My compiler Apple LLVM version 6.0 (clang-600.0.56) (based on LLVM 3.5svn) does not automagically link either ncurses or zlib -- thus adding -lncurses and -lz was necessary in order to get it to build.

2b. What is -ltinfo? is it terminfo? Apparently doesn't exist here.

Perhaps (2) is just differences in llvm-config output... Anyway I managed to get it to build!

Pretty certain the lexer doesn't keep proper track of current character and line numbers. This should be stored for each token, which means error handling will be a lot easier.

Not as consistent as I'd like them to be

Without Arguments:

user@box:~/jayfor$ ./j4
error: no input files

With -d:

user@box:~/jayfor$ ./j4 -d
fopen: could not read file: Bad address
file: (null)

With -r:

user@box:~/jayfor$ ./j4 -r
fopen: could not read file: Bad address
file: (null)

With -o:

user@box:~/jayfor$ ./j4 -o
fopen: could not read file: Bad address
file: (null)

This isn't very important, but once the code actually starts running, this will need to be improved.

Need to decide if the language should be interpreted, write my own bytecode compiler, or use LLVM.

Characters can be defined with only one single quote, like so:

char x = 'a;

As opposed to having a closing single quote, saving a single character!!11111

We could still hold support for this syntax:

char x = 'a';

I'll fix this in a bit.

I've done a brief look-through of the code, and I must say it's neat and well-commented. Unfortunately, I seem to have spotted something... weird.

void destroy_break_ast_node(break_ast_node *bn) {
    if (!bn) {
        free(bn);
        bn = NULL;
    }
}

There's a bunch of functions like that. if(!bn)... means that bn == 0, meaning you're passing NULL to free... then setting it to NULL again?

Something fishy is up here. Probably just an oversight/result of late-night coding, these happen.

I like the jayfor syntax but the for loop could be like: for int x (0, 10), instead of for int x:(0, 10). I prefer for statement without colon because the parantheses already emphasize what is inside it.

Currently there is a function that parsers integers, this should be pretty easy to change so it parses other data types.

structures like this:

struct my_struct {
    int x
    int y
    float x
}

and structures with default values:

struct my_struct {
    float x = 0
    float y = 0
}

Not too important at the moment, fix later.

Probably wont fix this, but I am aware.

Since adding the new expression parsing, I've broken the default argument values:

fn this(int x = 5): void {

}

The x = 5 will not parse correctly as an expression.

Simply changing return statements to actually say "return" instead of "ret" will make it a lot more readable and pleasant to look at!

jk

Line 1042 of parser.c looks like this:

data_type *data_type = malloc(sizeof(data_type));

In this case, sizeof(data_type) returns the size of the pointer declared on the same line, not the size of the data_type data type.

When I run

./j4 tests/for.j4

I get a segmentation fault. Not sure if its supposed to happen (if you're aware of this). Informing anyway