On Monday, 2 January 2017 at 19:01:43 UTC, Stefan Koch wrote:
> On Monday, 2 January 2017 at 18:40:44 UTC, Stefan Koch wrote:
>> So guys.
>>
>> basic function-calls are supported.
>>
>> Hell YEAH!
>
> Meaning this will compile :
>
> uint caller(uint n)
> {
>   return callee(n, 2);
> }
>
> uint callee(uint a, uint b)
> {
>   return a*b;
> }
>
> static assert(caller(3) == 6);
> static assert(caller(24) == 48);
>
>
> uint rfn(uint n)
> {
>   if (n) return n + rfn(n-1);
>   return 1;
> }
>
> pragma(msg, rfn(2000));

There's something I've been wondering about. NewCTFE is basically just another backend, right? Does this mean that you'll have to implement various optimizations such as tail-call elision independently from what DMD with the default backend currently does?

On Monday, 2 January 2017 at 21:52:05 UTC, Meta wrote:
>
> There's something I've been wondering about. NewCTFE is basically just another backend, right? Does this mean that you'll have to implement various optimizations such as tail-call elision independently from what DMD with the default backend currently does?

Yes, I could borrow analysis passes from dmd however to check if the optimization is applicable.

However I do not intend to do so.
CTFE is mostly used for small one-shot things.
There is no point in eliding calls.
Executing the code as soon as possible is often faster then doing the optimization.

On Monday, 2 January 2017 at 18:40:44 UTC, Stefan Koch wrote:
> Hell YEAH!

Nice work.

Hi Guys,

I just enabled the handling for with(Type).
makes the following code work :

import core.time;
import core.sys.linux.time;

 auto _posixClock(ClockType clockType)
    {
            with(ClockType) final switch(clockType)
            {
            case bootTime: return CLOCK_BOOTTIME;
            case coarse: return CLOCK_MONOTONIC_COARSE;
            case normal: return CLOCK_MONOTONIC;
            case precise: return CLOCK_MONOTONIC;
            case processCPUTime: return CLOCK_PROCESS_CPUTIME_ID;
            case raw: return CLOCK_MONOTONIC_RAW;
            case threadCPUTime: return CLOCK_THREAD_CPUTIME_ID;
            case second: assert(0);
            }
}

static assert(_posixClock(ClockType.bootTime) == CLOCK_BOOTTIME);

On Monday, 2 January 2017 at 18:40:44 UTC, Stefan Koch wrote:
> So guys.
>
> basic function-calls are supported.
>
> Hell YEAH!

Disabled for now. Since it pushes some buffers/caches to their limit. A more flexible solution is needed.
However the good news is that the ABI seems to work better than expected.
Now it's time to deal with this-pointers and method calls.

After remaining issues with nested switches have to be fixed.

On Thursday, 5 January 2017 at 06:04:00 UTC, Stefan Koch wrote:
> On Monday, 2 January 2017 at 18:40:44 UTC, Stefan Koch wrote:
>> So guys.
>>
>> basic function-calls are supported.
>>
>> Hell YEAH!
>
> Disabled for now. Since it pushes some buffers/caches to their limit. A more flexible solution is needed.

This is now fixed by simply extending the limits.
Now more failures start to show and more manual bailouts are required again.

Although, I expected this to happen it bums me. :(

I have fixed a few bugs in the handling of casts.
And fixed an oversight which would cause Xors to bailout.
However Phobos-unittests are still regressed.

I am now working on Method-Calls or member-function-calls.

On Thursday, 5 January 2017 at 06:04:00 UTC, Stefan Koch wrote:
> On Monday, 2 January 2017 at 18:40:44 UTC, Stefan Koch wrote:
>> So guys.
>>
>> basic function-calls are supported.
>>
>> Hell YEAH!
>
> Disabled for now. Since it pushes some buffers/caches to their limit. A more flexible solution is needed.
> However the good news is that the ABI seems to work better than expected.
> Now it's time to deal with this-pointers and method calls.
>
> After remaining issues with nested switches have to be fixed.

Basic Function call support reenabled.
Phobos unittests pass.

On Monday, 16 January 2017 at 16:31:25 UTC, Stefan Koch wrote:
> Basic Function call support reenabled.
> Phobos unittests pass.

Nice!

On Monday, 16 January 2017 at 16:31:25 UTC, Stefan Koch wrote:
> On Thursday, 5 January 2017 at 06:04:00 UTC, Stefan Koch wrote:
>> On Monday, 2 January 2017 at 18:40:44 UTC, Stefan Koch wrote:
>>> So guys.
>>>
>>> basic function-calls are supported.
>>>
>>> Hell YEAH!
>>
>> Disabled for now. Since it pushes some buffers/caches to their limit. A more flexible solution is needed.
>> However the good news is that the ABI seems to work better than expected.
>> Now it's time to deal with this-pointers and method calls.
>>
>> After remaining issues with nested switches have to be fixed.
>
> Basic Function call support reenabled.
> Phobos unittests pass.

Just found a bug with recursive functions, which is now fixed.
So time for some perf:
-- t.d
 uint rfn(uint n)
 {
   if (n > 2) return n + rfn(n-1) + rfn(n-2);
   return 1;
 }

pragma(msg, rfn(30));
--

uplink@uplink-pc:~/d/dmd$ time src/dmd t.d -c
4674396u

real	0m0.396s
user	0m0.392s
sys	0m0.000s
uplink@uplink-pc:~/d/dmd$ time dmd t.d -c
4674396u

real	0m1.993s
user	0m1.872s
sys	0m0.120s