bitfields VS pure nothrow

[Alex Rønne Petersen]

On 13-05-2012 21:51, Guillaume Chatelet wrote:
> Sure enough bitfields is a strange beast and I ran into some subtleties
> with purity and nothrow.
>
> Consider the following code :
> ------------------------------
> struct POD {
>     int a;
> }
>
> int getA(POD o) pure nothrow {
>     return o.a;
> }
>
> POD setA(POD o, int value) pure nothrow {
>     o.a = value;
>     return o;
> }
> ------------------------------
>
> It compiles fine. But now with a bitfield :
> struct POD {
> 	mixin(std.bitmanip.bitfields!(
>          int, "a",10,
>          int, "", 22));
> }
>
> The compiler complains about missing pure and nothrow attributes
> Error: pure function 'getA' cannot call impure function 'a'
> Error: o.a is not nothrow
> Error: function test.getA 'getA' is nothrow yet may throw
> Error: pure function 'setA' cannot call impure function 'a'
> Error: o.a is not nothrow
> Error: function test.setA 'setA' is nothrow yet may throw
>
> Which makes sense as the mixin outputs :
> ------------------------------
> @property uint a() const {
>     auto result = (_a_&  1023U)>>0U;
>     return cast(uint) result;
> }
> @property void a(uint v){
>     assert(v>= a_min);
>     assert(v<= a_max);
>     _a_ = cast(typeof(_a_))
>           ((_a_&  ~1023U) | ((cast(typeof(_a_)) v<<  0U)&  1023U));
> }
> enum uint a_min = cast(uint)0U;
> enum uint a_max = cast(uint)1023U;
> private uint _a_;
> ------------------------------
>
> IMHO getters and setters could be nothrow but what about purity ? Looks
> like it's a bit far-fetched to qualify member methods as pure right ?
> Yet it makes sense regarding the semantic. What's your take on that ?

I've sent a pull request that fixes this: 
https://github.com/D-Programming-Language/phobos/pull/583

-- 
- Alex