General fluff => Announcements => Topic started by: Josh @ Dreamland on July 08, 2010, 12:01:23 PM

Title: Progress
Post by: Josh @ Dreamland on July 08, 2010, 12:01:23 PM
I'm so happy with a couple changes I've made and am making.

First off, I'm proud to announce that standard templates are now completely available to ENIGMA users.
Code: (C) [Select]
map a; // is the same as...
map<> a; // is the same as...
map <variant> a; // is the same as...
map <variant,variant> a;

That's right: ENIGMA now correctly defaults all non-defaulted template parameters (as in, template parameters that are not already optional according to the C++ library) to variant. That means that once Ism and I are finished coordinating our new "Definitions" resource (previously "WhiteSpace"... we're still working on a catchy name) you will be able to #include <map>, use it, and use codes like this:

Code: (EDL) [Select]
map a;
a[100] = "one hundred";
a[15625682900000000.0] = "lol, big float";
a[3.141592653589793238] = "pie";
a["pi"] = 3.14;

That's right: with the implementation of var4, comparison operators are correctly implemented for all types and mirrored const. This makes them capable of complete interaction with the standard template library. GM's ds_* functions are officially toilet food in comparison.

Furthermore, map isn't exactly necessary for large numbers. Var4 also implements Lua's algorithm for fast, sparse arrays. This introduces a number of nice properties:

Code: (EDL) [Select]
var a = 0;
for (int i=0; i<100; i++) a[i] = i;
a[1000000000] = "one billion";

How can this be efficient? It's simple. a[0..127] operates in O(1), because that's how much space was allocated for the array. a[128] operates in either O(1) or O(n), depending on how your system resources are doing. a[256..INT_MAX] operates in O(log(N)), as it relies on a map for the lookup. (This will hopefully be replaced with a sparse hash map at some point).

Currently the system is functional, but not finished. Var4's lua_table<> needs to ensure that elements can hop data structure when the array part is resized. Templates need field tested, which will happen after my latest renovation: the dot operator.

Because C++ introduces structures and classes which I do not want to alienate, I have to take special care to ensure that the left-hand side of the dot is not an instance of a structure. This requires a type resolver, which I hope to complete inside 400 lines. We'll see. Basically, it will return the type of the expression that precedes the dot. I intend for the following to happen:

If it's an instance of a class, check for a member by name of the variable to the right hand side of the dot. If the member exists, ignore the dot. Possibly, if the class was actually a pointer, not a direct reference, replace the dot with ->. C++ should have done this from square one; I see no use for pointer.member. If no member was found, or if the type was scalar, replace the dot with an integer-based accessor function and record that the global was indeed accessed that way so optimizations can be done later. Also try to warn if it was a class without a member and without operator int() or some other scalar that can be cast easily. Perhaps it should be up to ENIGMA to find an appropriate cast path, such as int(var()).

Anyway, just something to think about.
Back to work, ciao for now.
Title: Re: Progress
Post by: freezway on July 08, 2010, 12:32:47 PM
Thats fucking awesome! uh, u mind commiting?
Title: Re: Progress
Post by: IsmAvatar on July 08, 2010, 01:31:29 PM
Yes, please commit something that somewhat works. I think the latest revisions have some sort of error about instances or something?
Title: Re: Progress
Post by: Josh @ Dreamland on July 08, 2010, 07:35:19 PM
Screw you. You can have it when I'm 100% happy with the type resolver. Or I cuss it out and abandon it, then commit this part in disgust.
Title: Re: Progress
Post by: freezway on July 08, 2010, 07:47:13 PM
also, i have started uysing to bugtracker to submit bugs