View mode: basic / threaded / horizontal-split · Log in · Help
September 15, 2009
Compile-time AAs
Don has recently said that adding dynamic arrays at compile-time looks easy. I'd also like to have compile-time associative arrays. So you can fill them inside a CT function (the compiler may try to do the same for AAs created inside a static this(){}), and save some run time.
Even if such compile-time AAs have to be immutable at run-time they can be useful anyway.
A smarter implementation of such CT AAs may even use a perfet hashing, to make them quite fast.

Bye,
bearophile
September 15, 2009
Re: Compile-time AAs
Hear, hear, I second the motion .. especially if such feature can be used to implement reverse enums.

It's tedious writing reverse mapping tables (in C++) for converting the runtime value of an enum back to its symbolic (lexical) name for debug purposes.  Sounds like bearophile's CT AA might help to achieve this facility with ease in D.  Then again, the features of D that I'm still learning about continue to amaze me so perhaps one can do this (reverse enum) thing in D already .. just that it (the mystery feature) hasn't found me yet.

Cheers
Justin Johansson

<_/>


bearophile Wrote:

> Don has recently said that adding dynamic arrays at compile-time looks easy. I'd also like to have compile-time associative arrays. So you can fill them inside a CT function (the compiler may try to do the same for AAs created inside a static this(){}), and save some run time.
> Even if such compile-time AAs have to be immutable at run-time they can be useful anyway.
> A smarter implementation of such CT AAs may even use a perfet hashing, to make them quite fast.
> 
> Bye,
> bearophile
September 15, 2009
Re: Compile-time AAs
Justin Johansson wrote:
> Hear, hear, I second the motion .. especially if such feature can be used to implement reverse enums.
> 
> It's tedious writing reverse mapping tables (in C++) for converting the runtime value of an enum back to its symbolic (lexical) name for debug purposes.  Sounds like bearophile's CT AA might help to achieve this facility with ease in D.  Then again, the features of D that I'm still learning about continue to amaze me so perhaps one can do this (reverse enum) thing in D already .. just that it (the mystery feature) hasn't found me yet.
> 
> Cheers
> Justin Johansson
> 

Giving enums a tupleof property would make more sense, I think. I've
always wondered why they don't have it.
September 15, 2009
Re: Compile-time AAs
bearophile wrote:
> Don has recently said that adding dynamic arrays at compile-time looks easy. I'd also like to have compile-time associative arrays. So you can fill them inside a CT function (the compiler may try to do the same for AAs created inside a static this(){}), and save some run time.
> Even if such compile-time AAs have to be immutable at run-time they can be useful anyway.
> A smarter implementation of such CT AAs may even use a perfet hashing, to make them quite fast.

Indeed. I think perfect hashing is one of the primary appeals of a 
compile-time AA.

BTW, you can use AAs inside CTFE already. There's probably missing 
functionality, though -- create a Bugzilla test case for anything you 
find. The primary thing which is missing is that you can't use an AA 
literal to populate a runtime AA (this is a backend issue, not a CTFE 
limitation). You can get the AA keys and values as arrays, though, so 
you could populate the AA yourself.
September 15, 2009
Re: Compile-time AAs
Ellery Newcomer:
> Giving enums a tupleof property would make more sense, I think. I've
> always wondered why they don't have it.

While we wait some years for a saner solution, you may use this bad looking hack (D2):
http://www.digitalmars.com/d/2.0/phobos/std_typecons.html#defineEnum

Bye,
bearophile
September 15, 2009
Re: Compile-time AAs
Don Wrote:

> bearophile wrote:
> > Don has recently said that adding dynamic arrays at compile-time looks easy. I'd also like to have compile-time associative arrays. So you can fill them inside a CT function (the compiler may try to do the same for AAs created inside a static this(){}), and save some run time.
> > Even if such compile-time AAs have to be immutable at run-time they can be useful anyway.
> > A smarter implementation of such CT AAs may even use a perfet hashing, to make them quite fast.
> 
> Indeed. I think perfect hashing is one of the primary appeals of a 
> compile-time AA.
> 
> BTW, you can use AAs inside CTFE already. There's probably missing 
> functionality, though -- create a Bugzilla test case for anything you 
> find. The primary thing which is missing is that you can't use an AA 
> literal to populate a runtime AA (this is a backend issue, not a CTFE 
> limitation). You can get the AA keys and values as arrays, though, so 
> you could populate the AA yourself.

Just out of curiosity, how are compile time AAs implemented? The D runtime already handles the creation and lookups of these arrays, so let's suppose I changed my runtime to have a completely different AA implementation, what would happen when you mix both compile time and runtime AAs?
September 15, 2009
Re: Compile-time AAs
Jeremie Pelletier wrote:
> Don Wrote:
> 
>> bearophile wrote:
>>> Don has recently said that adding dynamic arrays at compile-time looks easy. I'd also like to have compile-time associative arrays. So you can fill them inside a CT function (the compiler may try to do the same for AAs created inside a static this(){}), and save some run time.
>>> Even if such compile-time AAs have to be immutable at run-time they can be useful anyway.
>>> A smarter implementation of such CT AAs may even use a perfet hashing, to make them quite fast.
>> Indeed. I think perfect hashing is one of the primary appeals of a 
>> compile-time AA.
>>
>> BTW, you can use AAs inside CTFE already. There's probably missing 
>> functionality, though -- create a Bugzilla test case for anything you 
>> find. The primary thing which is missing is that you can't use an AA 
>> literal to populate a runtime AA (this is a backend issue, not a CTFE 
>> limitation). You can get the AA keys and values as arrays, though, so 
>> you could populate the AA yourself.
> 
> Just out of curiosity, how are compile time AAs implemented? The D runtime already handles the creation and lookups of these arrays, so let's suppose I changed my runtime to have a completely different AA implementation, what would happen when you mix both compile time and runtime AAs?
The funny thing is, there are AA literals, but they can only be used at 
compile time. "An AssocArrayLiteral cannot be used to statically 
initialize anything." (expression.html in the spec).
A compile-time AA is just a pointer to an AA literal. It's really 
peculiar, because that's the only time they can be used. So there's 
pretty much zero interaction between compile-time and run-time AAs!
Top | Discussion index | About this forum | D home