... some small, some bigger. On my OSX machine all unittests run properly. Please let me know how things work for you!

Andrei

Still fails on Windows.

On 28 July 2010 10:10, Andrei Alexandrescu <andrei at erdani.com> wrote:
> ... some small, some bigger. On my OSX machine all unittests run properly. Please let me know how things work for you!
>
> Andrei
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>

Here's a regression, from Walter's test suite. Used to work, now vomits six obscure error messages.

import std.stdio;

void foo(creal a)
{
  writeln(a);
}


On 28 July 2010 10:10, Andrei Alexandrescu <andrei at erdani.com> wrote:
> ... some small, some bigger. On my OSX machine all unittests run properly. Please let me know how things work for you!
>
> Andrei
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>

Also:
\src\phobos\std\format.d(1282): Error: undefined identifier S, did you
mean variable f?

That was as far as I've got. I've reverted back to Phobos svn 1679. There is probably a later version which is OK, but I tried a few later versions, and they were all broken.

We should probably be a bit more ruthless about broken checkins. Once the unittests are failing, nobody can check any more and the errors start compounding.


On 28 July 2010 13:12, Don Clugston <dclugston at googlemail.com> wrote:
> Here's a regression, from Walter's test suite. Used to work, now vomits six obscure error messages.
>
> import std.stdio;
>
> void foo(creal a)
> {
> ?writeln(a);
> }
>
>
> On 28 July 2010 10:10, Andrei Alexandrescu <andrei at erdani.com> wrote:
>> ... some small, some bigger. On my OSX machine all unittests run properly. Please let me know how things work for you!
>>
>> Andrei
>> _______________________________________________
>> phobos mailing list
>> phobos at puremagic.com
>> http://lists.puremagic.com/mailman/listinfo/phobos
>>
>

Apologies for causing such frustration.

While working on std.format, I found fit to drop support for formatting built-in complex numbers. That's what caused the problems. From here, we could either add it back or remove the offending unittests. Please advise.


Andrei

Don Clugston wrote:
> Also:
> \src\phobos\std\format.d(1282): Error: undefined identifier S, did you
> mean variable f?
> 
> That was as far as I've got. I've reverted back to Phobos svn 1679. There is probably a later version which is OK, but I tried a few later versions, and they were all broken.
> 
> We should probably be a bit more ruthless about broken checkins. Once the unittests are failing, nobody can check any more and the errors start compounding.
> 
> 
> On 28 July 2010 13:12, Don Clugston <dclugston at googlemail.com> wrote:
>> Here's a regression, from Walter's test suite. Used to work, now vomits six obscure error messages.
>>
>> import std.stdio;
>>
>> void foo(creal a)
>> {
>>  writeln(a);
>> }
>>
>>
>> On 28 July 2010 10:10, Andrei Alexandrescu <andrei at erdani.com> wrote:
>>> ... some small, some bigger. On my OSX machine all unittests run properly. Please let me know how things work for you!
>>>
>>> Andrei
>>> _______________________________________________
>>> phobos mailing list
>>> phobos at puremagic.com
>>> http://lists.puremagic.com/mailman/listinfo/phobos
>>>
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos

Please add it back in. There needs to be more transition time for dropping complex.

Andrei Alexandrescu wrote:
> Apologies for causing such frustration.
>
> While working on std.format, I found fit to drop support for formatting built-in complex numbers. That's what caused the problems. From here, we could either add it back or remove the offending unittests. Please advise.
>
>
> Andrei
>
> Don Clugston wrote:
>> Also:
>> \src\phobos\std\format.d(1282): Error: undefined identifier S, did you
>> mean variable f?
>>
>> That was as far as I've got. I've reverted back to Phobos svn 1679. There is probably a later version which is OK, but I tried a few later versions, and they were all broken.
>>
>> We should probably be a bit more ruthless about broken checkins. Once the unittests are failing, nobody can check any more and the errors start compounding.
>>
>>
>> On 28 July 2010 13:12, Don Clugston <dclugston at googlemail.com> wrote:
>>> Here's a regression, from Walter's test suite. Used to work, now vomits six obscure error messages.
>>>
>>> import std.stdio;
>>>
>>> void foo(creal a)
>>> {
>>>  writeln(a);
>>> }
>>>
>>>
>>> On 28 July 2010 10:10, Andrei Alexandrescu <andrei at erdani.com> wrote:
>>>> ... some small, some bigger. On my OSX machine all unittests run
>>>> properly.
>>>> Please let me know how things work for you!
>>>>
>>>> Andrei
>>>> _______________________________________________
>>>> phobos mailing list
>>>> phobos at puremagic.com
>>>> http://lists.puremagic.com/mailman/listinfo/phobos
>>>>
>> _______________________________________________
>> phobos mailing list
>> phobos at puremagic.com
>> http://lists.puremagic.com/mailman/listinfo/phobos
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>
>

I say it's high time to ditch the built-in complex numbers.  But then again, I'm the author of std.complex and would of course like to see it used, so how can I say anything else?  ;)

But I think Walter has a good point:  Perhaps there should be one or two releases where creal & co. are deprecated, but still work, before support for them is removed from Phobos.

-Lars


On Wed, 2010-07-28 at 08:20 -0700, Andrei Alexandrescu wrote:
> Apologies for causing such frustration.
> 
> While working on std.format, I found fit to drop support for formatting built-in complex numbers. That's what caused the problems. From here, we could either add it back or remove the offending unittests. Please advise.
> 
> 
> Andrei
> 
> Don Clugston wrote:
> > Also:
> > \src\phobos\std\format.d(1282): Error: undefined identifier S, did you
> > mean variable f?
> > 
> > That was as far as I've got. I've reverted back to Phobos svn 1679. There is probably a later version which is OK, but I tried a few later versions, and they were all broken.
> > 
> > We should probably be a bit more ruthless about broken checkins. Once the unittests are failing, nobody can check any more and the errors start compounding.
> > 
> > 
> > On 28 July 2010 13:12, Don Clugston <dclugston at googlemail.com> wrote:
> >> Here's a regression, from Walter's test suite. Used to work, now vomits six obscure error messages.
> >>
> >> import std.stdio;
> >>
> >> void foo(creal a)
> >> {
> >>  writeln(a);
> >> }
> >>
> >>
> >> On 28 July 2010 10:10, Andrei Alexandrescu <andrei at erdani.com> wrote:
> >>> ... some small, some bigger. On my OSX machine all unittests run properly. Please let me know how things work for you!
> >>>
> >>> Andrei
> >>> _______________________________________________
> >>> phobos mailing list
> >>> phobos at puremagic.com
> >>> http://lists.puremagic.com/mailman/listinfo/phobos
> >>>
> > _______________________________________________
> > phobos mailing list
> > phobos at puremagic.com
> > http://lists.puremagic.com/mailman/listinfo/phobos
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos

On 28 July 2010 20:39, Lars Tandle Kyllingstad <lars at kyllingen.net> wrote:
> I say it's high time to ditch the built-in complex numbers. ?But then again, I'm the author of std.complex and would of course like to see it used, so how can I say anything else? ?;)
>
> But I think Walter has a good point: ?Perhaps there should be one or two releases where creal & co. are deprecated, but still work, before support for them is removed from Phobos.
>
> -Lars

std.complex needs be to firmly established as a complete replacement
for creal et. al, before they can be declared deprecated. We're not
there yet (writefln of complex numbers didn't work in the last
release).
Even when it happens, functions like writeln should behave a bit more
gracefully. Does it do the same thing with other unusual types?

It's also worth noting that moving associative arrays to a library type has been very painful. Nearly ten releases later, we still have some horrible regressions. Doing the same thing with complex needs to be done very carefully.



>
>
> On Wed, 2010-07-28 at 08:20 -0700, Andrei Alexandrescu wrote:
>> Apologies for causing such frustration.
>>
>> While working on std.format, I found fit to drop support for formatting built-in complex numbers. That's what caused the problems. From here, we could either add it back or remove the offending unittests. Please advise.
>>
>>
>> Andrei
>>
>> Don Clugston wrote:
>> > Also:
>> > \src\phobos\std\format.d(1282): Error: undefined identifier S, did you
>> > mean variable f?
>> >
>> > That was as far as I've got. I've reverted back to Phobos svn 1679. There is probably a later version which is OK, but I tried a few later versions, and they were all broken.
>> >
>> > We should probably be a bit more ruthless about broken checkins. Once the unittests are failing, nobody can check any more and the errors start compounding.
>> >
>> >
>> > On 28 July 2010 13:12, Don Clugston <dclugston at googlemail.com> wrote:
>> >> Here's a regression, from Walter's test suite. Used to work, now vomits six obscure error messages.
>> >>
>> >> import std.stdio;
>> >>
>> >> void foo(creal a)
>> >> {
>> >> ?writeln(a);
>> >> }
>> >>
>> >>
>> >> On 28 July 2010 10:10, Andrei Alexandrescu <andrei at erdani.com> wrote:
>> >>> ... some small, some bigger. On my OSX machine all unittests run properly. Please let me know how things work for you!
>> >>>
>> >>> Andrei
>> >>> _______________________________________________
>> >>> phobos mailing list
>> >>> phobos at puremagic.com
>> >>> http://lists.puremagic.com/mailman/listinfo/phobos
>> >>>
>> > _______________________________________________
>> > phobos mailing list
>> > phobos at puremagic.com
>> > http://lists.puremagic.com/mailman/listinfo/phobos
>> _______________________________________________
>> phobos mailing list
>> phobos at puremagic.com
>> http://lists.puremagic.com/mailman/listinfo/phobos
>
>
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>

Don Clugston wrote:
> std.complex needs be to firmly established as a complete replacement for creal et. al, before they can be declared deprecated. We're not there yet (writefln of complex numbers didn't work in the last release).

Agreed.

> Even when it happens, functions like writeln should behave a bit more gracefully. Does it do the same thing with other unusual types?

Well it's funny. formattedWrite does behave nicely with unusual structs, but I emphasize - structs. The complex types are not structs so they aren't handled "nicely".

> It's also worth noting that moving associative arrays to a library type has been very painful. Nearly ten releases later, we still have some horrible regressions. Doing the same thing with complex needs to be done very carefully.

Agreed. I think and hope that the pain is well worth it.


Andrei

On Wed, 28 Jul 2010, Andrei Alexandrescu wrote:

> > Even when it happens, functions like writeln should behave a bit more gracefully. Does it do the same thing with other unusual types?
> 
> Well it's funny. formattedWrite does behave nicely with unusual structs, but I emphasize - structs. The complex types are not structs so they aren't handled "nicely".

Which opens the obvious question.. why aren't they structs?  The builtin types are value types, not reference types, right?