On Friday, 10 August 2012 at 12:32:28 UTC, bearophile wrote:
This second D version uses the same class definitions, but
allocates the class instances on the stack. The code is bug
prone and ugly. The other disadvantages are unchanged:
void main() {
import std.stdio;
import std.conv: emplace;
import core.stdc.stdlib: alloca;
enum size_t size_Val = __traits(classInstanceSize, Val);
enum size_t size_Plus = __traits(classInstanceSize, Plus);
enum size_t size_Minus = __traits(classInstanceSize, Minus);
Val e1 = emplace!Val(alloca(size_Val)[0 .. size_Val], 5);
Val e2 = emplace!Val(alloca(size_Val)[0 .. size_Val], 3);
Val e3 = emplace!Val(alloca(size_Val)[0 .. size_Val], 1);
Plus e4 = emplace!Plus(alloca(size_Plus)[0 .. size_Plus],
e2, e3);
Minus ex2 = emplace!Minus(alloca(size_Minus)[0 ..
size_Minus], e1, e4);
writeln("Val: ", eval(ex2));
}
Probably there are ways to improve my D versions, or to write
better versions.
Bye,
bearophile
I think version 2 would be the easiest one to improve, by
including a combined emplace/alloca convenience function in
Phobos for this common use-case.
See the technique used in:
http://www.digitalmars.com/d/archives/digitalmars/D/run-time_stack-based_allocation_166305.html
"auto Create(void* buf=alloca(frame_size))"
