ENIGMA Forums
General fluff => Announcements => Topic started by: Josh @ Dreamland on December 02, 2011, 11:25:03 am
-
By my calculations, this will be the third time we've changed EDL's featureset considerably. In the beginning, I was just going to support GML as laid out in 6.1. Over time, I decided, "fuck that, we need real constructs and language features." Since then, a lot has been changing, even in the world of computer science as a whole. For instance, C++'s feature set has grown sharply with the publishing of the most recent ISO. So basically, I guess we need to start discussing EDL tuneups before I go ahead and write up the new parser.
Biggest changes proposed:
- Use of C++ struct keyword in plain EDL. Since I started that JS port, which I thought TGMG was running with but I now guess is still my problem, a lot of turbulence erupted in maintaining a useful, cross-compatible language. While the Definitions resource provides excellent extensibility to the C++ version, some elements missing from GML are too basic to require to be defined in their respective languages. What I'm saying is, JavaScript makes it a bitch to define classes unless you understand what exactly a JavaScript class is, and this is likely the case for other languages we may support. The bottom line is, it would be nice if ENIGMA provided a syntactical wrapper for useful basics like structures.
- Rusky is getting his Lambdas. Now that C++ supports Lambdas, I don't see why not to support them in EDL.
- Inline array declaration. EDL is in dire need of some way to put numbers into arrays adequately.
- A more elegant preprocessor system should be available to the users. C has preprocessors that change how your code looks before it compiles. If you've used much EDL at all, you've used them, probably without noticing it. I would like to share that system with users; the issue is, C's are really ugly, and really bulky.
- EDL's understanding of types is getting an overhaul. We will now ensure that complicated expressions don't trip up the compile because ENIGMA can't adequately cast them. With this comes support for ENIGMA's varargs class.
Now, the point of this thread is to present what we want added, get more suggestions for what to add, and also, to deliberate on how these features will interact when they are added.
For example, how do we want Lambdas to look? How do we want arrays to look? How do we want preprocessors to look?
Originally, my proposal for preprocessors was [[ifdef BUILD_WINDOWS]]do_something_with_the_registry_or_something()[[endif]].
And, well, most languages use [...] for arrays. [1, 2, 3, 4, 5].
And, C++ uses [] to indicate lambda. [](int a) { return a + 1; }
So, with a good enough parser, no worries, right? Not really; some of these get ambiguous.
What if I want a singleton array of a singleton array of a variable called pragma? [[pragma]]. Oops. Even if ENIGMA got it right, the highlighter wouldn't, so it'd be ugly as hell. So, do I ask users to simply put spaces between their brackets? I think not. The best option is to use something else for one of them. For example, [%ifdef BUILD_WINDOWS%]. Since % is a binary operator, any mention of it after a [ or before a ] is invalid. So now, how to distinguish [] as an empty array and [] as a lambda? I think that'll mostly take care of itself, since in a lambda the next token must be ().
With that, here are our current proposals:
C++ structs
Can be used in any code that uses the structure individually, declared for the whole object with local struct, or declared for the whole game with global struct. The latter will require reparse of the entire game, similar to placing the structure in definitions. As such, it may be dropped or discouraged.
Lambdas
Rusky's proposal for lambdas is to mirror C#'s syntax. mylambda = (x,y) => { x + y }. For simpler expressions, square = x => x*x
Inline Arrays
Basically, [] instantiates an array of variants as its own class, including a length member. There would be functions taking Array& as an argument. Var would be able to easily produce an Array& in O(1), and there would be a var::var(Array) constructor. What this means is, you'd be able to do this:
var a = [1, 2, 3, 4, 5];
a = reverse(a);
Array assign
One of MATLAB and other language's brighter ideas, this could be enabled with a special case in the parser:
var a, b, c;
[a, b, c] = get_abc();
Where get_abc(); returns Array.
Some things I'm presently pondering:
1. Can the new parser have the ass to choose between mean(double,double), mean(varargs&), and mean(Array)?
2. Can we scrap Array altogether and just use var, or would a separate array class be more efficient?
So anyway, if you don't like a proposal, if you have a better idea, or if you have something new you would like to see in the language, let us know.
Peace.
-
Most of the ideas are great, though making all of that for the parser will take freaking years.
Though about arrays, why does it change arrays into function right now? Like "if(ass[0])" is "if (ass(0))". I know that its probably because they can have dynamic size and they don't have a specific type (variant I guess is a type). Its just that I had problems with some array declaration before. For example, this errors:
int some_array[10];
some_array[5]=10;
Preprocessor_Environment_Editable/IDE_EDIT_objectfunctionality.h:37:17: error: 'some_array' cannot be used as a function
While this works fine:local int some_array[15];
Also, this errors:n=15;
local int some_array[n];
I guess you get the point. So basically, will this get fixed when you change the array system?
edit: Also, btw, this code freezes not only LGM but crashes the whole JAVA VM:
local int intensity_res=32,
intmap_w=int(room_width/intensity_res),
intmap_h=int(room_height/intensity_res);
local int intensity_map[][];
for (int i=0; i<=intmap_w; i++){
for (int c=0; c<=intmap_h; i++){
intensity_map[i][c]=0;
}
}
I found that a few weeks ago, but I didn't have the time to post, and I just remember about it.
-
It's so var can do 2-dimensional array subscripts at once. It was for array[1,2] originally, but erroneously gets replaced with () in more cases.
-
OOOOOOOOoooooooooooooooooooohhhhhhhhhhhhhhhhhhhh.
-
Don't forget support for type inference(auto), which is pretty much mandatory when it comes to lambdas.
-
Well, I've actually still not decided whether or not to rely fully on C++11. I figure this new parser is going to need to be able to do type inference for itself. But sure, I think an auto keyword in the spec would be a good idea.
-
Global structs probably the most useful kind, so if you're going to provide a cross-language implementation there should be some resource similar to definitions that lets you define them. Maybe with an (optional?) graphical editor. It might even make sense to extend EDL enough that you can entirely replace Definitions (modulo #include) and just use that.
If you're going with lambdas, you can either rely on C++11 or write your own implementation by generating and constructing a class with operator() for each lambda. If you want captured variables to work outside the enclosing scope, you'll have to figure out what they are anyway, so you may as well go with generating your own classes since that will work on old/non-C++11 compilers.
I like the idea of arrays as a separate type- it makes var-compatibility simpler. You can pass/return arrays but not vars, etc. Also destructuring assignment is cool.
-
For C++11 lambdas, I believe the actual type of the lambdas is undefined by the spec. The only requirement is that lambdas that do not capture variables must be cast-able to function pointers.
So, if Josh intends to have type inference on his parser, he'll *have* to provide his own lambda implementation.
That is, unless Josh can use some Clang magic powder to make that problem go away.
-
Clang's no longer in the picture. I figure if I'm to maintain ENIGMA's interface with a C++ parser, it should be my own.
As for Array being a separate type--I realized it's not just a good idea, it's necessary. If I want Arrays of Arrays of Arrays of variants, I need some way of representing that which is independent of var, which can only store 2D arrays.
-
luiscubal: the spec does, however, require that lambdas be capturable by e.g. std::function<int(int, int)>
-
In addition, std::function is polymorphic, meaning that std::function<int(int, int)> can be converted to std::function<int(double, double)>.
-
inb4 i make another idc-as-long-as-i-can-code-gml post
-
I'm using Java right now for a school project, and I find it somewhat quirky.
It has and extensive standard library, but that's its main probem when it comes to deprecating stuff in order to innovate.
The more functionalities you add, the more you'll have to support.
As for Enigma, some of us just wanted a quick and (slightly less) dirty GM clone.
Anyway, I know GM sucks because it takes too much effort to build a simple textbox, nevermind a whole GUI.
Than there is another big problem with slowness and the confusion between classes and instances.
But still, sometimes I find myself wanting to go back to that old and simple piece of crap.
-
It's kind of easy to maintain backwards compatibility with a languages with as few features as GML. The point's to add functionality for people who find +=1 to be futile and who like to use static typing. And real structures.
-
GM will have structures and will compile GML before ENIGMA has a parser.
:troll:
-
GM will have structures and will compile GML before ENIGMA has a parser.
:troll:
Its funny because its true.
:troll:
-
GM will have structures and will compile GML before ENIGMA has a parser.
:troll:
Its funny because its true.
:troll:
GM will cost a dollar for every week it takes us to finish ENIGMA.
:troll:
-
GM will have structures and will compile GML before ENIGMA has a parser.
:troll:
Its funny because its true.
:troll:
GM will cost a dollar for every week it takes us to finish ENIGMA.
:troll:
So, ENIGMA won't ever de done?
-
:troll:
-
^