phobos commit, revision 1627


user: Don Clugston

msg:
Disabled unittests for CustomFloat, as an emergency action to allow Phobos unittests to compile. Something is horribly wrong with the new CustomFloat.

http://www.dsource.org/projects/phobos/changeset/1627

I have no idea how this happened, as they all compiled and passed for me.  Can you describe what the problem is?

On 6/11/2010 3:19 AM, dsource.org wrote:
> phobos commit, revision 1627
>
>
> user: Don Clugston
>
> msg:
> Disabled unittests for CustomFloat, as an emergency action to allow Phobos unittests to compile. Something is horribly wrong with the new CustomFloat.
>
> http://www.dsource.org/projects/phobos/changeset/1627
>
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>
> 

On 11 June 2010 15:40, David Simcha <dsimcha at gmail.com> wrote:
> I have no idea how this happened, as they all compiled and passed for me. ?Can you describe what the problem is?

Building phobos with -unittest generates
Error: Out of memory
while compiling that test.
Does it still work for you?

I had only tried to compile the individual std.numeric module and run the unittests.  This definitely works fine.  I had never tried to compile Phobos as a whole because compiling Phobos on Windows is somewhat of a PITA and I didn't anticipate such strange problems as these.

At any rate, I just tried it now and it works for me on Windows with 2.046. It looks like the oddities that had made building Phobos on Windows such a PITA (stuff related to minit.o) have been fixed recently, too.  On the beta, it fails with completely unrelated error messages related to CTFE of sqrt in std.math:

std\math.d(1611): Error: sqrt cannot be interpreted at compile time, because
it
has no available source code
std\math.d(1614): Error: static assert  (2L * (1L / (0.5L *
sqrt(0x1p-16382L)) /
 2L) * (1L / (0.5L * sqrt(0x1p-16382L)) / 2L) <=
0x1.fffffffffffffffep+16383L) i
s not evaluatable at compile time

What OS were you trying to build on?

On Fri, Jun 11, 2010 at 11:00 AM, Don Clugston <dclugston at googlemail.com>wrote:

> On 11 June 2010 15:40, David Simcha <dsimcha at gmail.com> wrote:
> > I have no idea how this happened, as they all compiled and passed for me.
> >  Can you describe what the problem is?
>
> Building phobos with -unittest generates
> Error: Out of memory
> while compiling that test.
> Does it still work for you?
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/phobos/attachments/20100611/45260b59/attachment.html>

On 11 June 2010 17:18, David Simcha <dsimcha at gmail.com> wrote:
> I had only tried to compile the individual std.numeric module and run the unittests.? This definitely works fine.? I had never tried to compile Phobos as a whole because compiling Phobos on Windows is somewhat of a PITA and I didn't anticipate such strange problems as these.
>
> At any rate, I just tried it now and it works for me on Windows with 2.046. It looks like the oddities that had made building Phobos on Windows such a PITA (stuff related to minit.o) have been fixed recently, too.? On the beta, it fails with completely unrelated error messages related to CTFE of sqrt in std.math:
>
> std\math.d(1611): Error: sqrt cannot be interpreted at compile time, because
> it
> has no available source code
> std\math.d(1614): Error: static assert? (2L * (1L / (0.5L *
> sqrt(0x1p-16382L)) /
> ?2L) * (1L / (0.5L * sqrt(0x1p-16382L)) / 2L) <=
> 0x1.fffffffffffffffep+16383L) i
> s not evaluatable at compile time
>
> What OS were you trying to build on?

Windows XP. Are you using the second beta? I've seen those error messages before, but they got fixed ages ago. Sounds to me like beta 1.

I assume the second beta, I downloaded it today from ftp.digitalmars.com and it was called 'dmd2beta.zip'.  From looking though the dmd-beta list, I can't find any other beta, though Walter did post an updated version a few days after the original version.

On Fri, Jun 11, 2010 at 12:15 PM, Don Clugston <dclugston at googlemail.com>wrote:

> On 11 June 2010 17:18, David Simcha <dsimcha at gmail.com> wrote:
> > I had only tried to compile the individual std.numeric module and run the unittests.  This definitely works fine.  I had never tried to compile
> Phobos
> > as a whole because compiling Phobos on Windows is somewhat of a PITA and
> I
> > didn't anticipate such strange problems as these.
> >
> > At any rate, I just tried it now and it works for me on Windows with
> 2.046.
> > It looks like the oddities that had made building Phobos on Windows such
> a
> > PITA (stuff related to minit.o) have been fixed recently, too.  On the
> beta,
> > it fails with completely unrelated error messages related to CTFE of sqrt
> in
> > std.math:
> >
> > std\math.d(1611): Error: sqrt cannot be interpreted at compile time,
> because
> > it
> > has no available source code
> > std\math.d(1614): Error: static assert  (2L * (1L / (0.5L *
> > sqrt(0x1p-16382L)) /
> >  2L) * (1L / (0.5L * sqrt(0x1p-16382L)) / 2L) <=
> > 0x1.fffffffffffffffep+16383L) i
> > s not evaluatable at compile time
> >
> > What OS were you trying to build on?
>
> Windows XP. Are you using the second beta? I've seen those error
> messages before, but they got fixed ages ago. Sounds to me like beta
> 1.
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/phobos/attachments/20100611/ae9ebcbc/attachment.html>

Indeed that error indicates an older dmd. Make sure running dmd shows version 2.047.

Andrei

David Simcha wrote:
> I assume the second beta, I downloaded it today from ftp.digitalmars.com <http://ftp.digitalmars.com> and it was called 'dmd2beta.zip'.  From looking though the dmd-beta list, I can't find any other beta, though Walter did post an updated version a few days after the original version.
> 
> On Fri, Jun 11, 2010 at 12:15 PM, Don Clugston <dclugston at googlemail.com <mailto:dclugston at googlemail.com>> wrote:
> 
>     On 11 June 2010 17:18, David Simcha <dsimcha at gmail.com
>     <mailto:dsimcha at gmail.com>> wrote:
>      > I had only tried to compile the individual std.numeric module and
>     run the
>      > unittests.  This definitely works fine.  I had never tried to
>     compile Phobos
>      > as a whole because compiling Phobos on Windows is somewhat of a
>     PITA and I
>      > didn't anticipate such strange problems as these.
>      >
>      > At any rate, I just tried it now and it works for me on Windows
>     with 2.046.
>      > It looks like the oddities that had made building Phobos on
>     Windows such a
>      > PITA (stuff related to minit.o) have been fixed recently, too.
>     On the beta,
>      > it fails with completely unrelated error messages related to CTFE
>     of sqrt in
>      > std.math:
>      >
>      > std\math.d(1611): Error: sqrt cannot be interpreted at compile
>     time, because
>      > it
>      > has no available source code
>      > std\math.d(1614): Error: static assert  (2L * (1L / (0.5L *
>      > sqrt(0x1p-16382L)) /
>      >  2L) * (1L / (0.5L * sqrt(0x1p-16382L)) / 2L) <=
>      > 0x1.fffffffffffffffep+16383L) i
>      > s not evaluatable at compile time
>      >
>      > What OS were you trying to build on?
> 
>     Windows XP. Are you using the second beta? I've seen those error
>     messages before, but they got fixed ages ago. Sounds to me like beta
>     1.
>     _______________________________________________
>     phobos mailing list
>     phobos at puremagic.com <mailto: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

Ok, I see why.  I had 2.046 and the beta installed concurrently, and I forgot to point my path variable at the beta before building Phobos.  Now I can reproduce this, both using the Phobos build scripts and compiling the file individually with unittests turned on from Codeblocks.  It still works fine on 2.046, though I did notice that DMD does take a while to (successfully) compile it and uses ~60 MB of memory.  It looks like a compiler regression, or else CustomFloat was relying on a bug that got fixed.

On Fri, Jun 11, 2010 at 12:29 PM, Andrei Alexandrescu <andrei at erdani.com>wrote:

> Indeed that error indicates an older dmd. Make sure running dmd shows version 2.047.
>
> Andrei
>
> David Simcha wrote:
>
>> I assume the second beta, I downloaded it today from ftp.digitalmars.com<
>> http://ftp.digitalmars.com> and it was called 'dmd2beta.zip'.  From
>> looking though the dmd-beta list, I can't find any other beta, though Walter
>> did post an updated version a few days after the original version.
>>
>>
>> On Fri, Jun 11, 2010 at 12:15 PM, Don Clugston <dclugston at googlemail.com<mailto: dclugston at googlemail.com>> wrote:
>>
>>    On 11 June 2010 17:18, David Simcha <dsimcha at gmail.com
>>    <mailto:dsimcha at gmail.com>> wrote:
>>     > I had only tried to compile the individual std.numeric module and
>>    run the
>>     > unittests.  This definitely works fine.  I had never tried to
>>    compile Phobos
>>     > as a whole because compiling Phobos on Windows is somewhat of a
>>    PITA and I
>>     > didn't anticipate such strange problems as these.
>>     >
>>     > At any rate, I just tried it now and it works for me on Windows
>>    with 2.046.
>>     > It looks like the oddities that had made building Phobos on
>>    Windows such a
>>     > PITA (stuff related to minit.o) have been fixed recently, too.
>> On the beta,
>>     > it fails with completely unrelated error messages related to CTFE
>>    of sqrt in
>>     > std.math:
>>     >
>>     > std\math.d(1611): Error: sqrt cannot be interpreted at compile
>>    time, because
>>     > it
>>     > has no available source code
>>     > std\math.d(1614): Error: static assert  (2L * (1L / (0.5L *
>>     > sqrt(0x1p-16382L)) /
>>     >  2L) * (1L / (0.5L * sqrt(0x1p-16382L)) / 2L) <=
>>     > 0x1.fffffffffffffffep+16383L) i
>>     > s not evaluatable at compile time
>>     >
>>     > What OS were you trying to build on?
>>
>>    Windows XP. Are you using the second beta? I've seen those error
>>    messages before, but they got fixed ages ago. Sounds to me like beta
>>    1.
>>    _______________________________________________
>>    phobos mailing list
>>    phobos at puremagic.com <mailto: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
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/phobos/attachments/20100611/07410930/attachment-0001.html>

On 11 June 2010 18:50, David Simcha <dsimcha at gmail.com> wrote:
> Ok, I see why.? I had 2.046 and the beta installed concurrently, and I forgot to point my path variable at the beta before building Phobos.? Now I can reproduce this, both using the Phobos build scripts and compiling the file individually with unittests turned on from Codeblocks.? It still works fine on 2.046, though I did notice that DMD does take a while to (successfully) compile it and uses ~60 MB of memory.? It looks like a compiler regression, or else CustomFloat was relying on a bug that got fixed.

It seems to be a recursive mixin. Every time it's instantiated, it
gets 4 bytes bigger.
Seems to be triggered by opOpAssign; maybe it's the change to
opOpAssign in this release: "+=" was passed to opOpAssign
now it's "+". There's code in opOpAssign which is clearly wrong - it's
stripping off the "=", which no longer exists.

If you don't think you can fix this really quickly, I think we should remove it completely from this release, and wait until the next one. It doesn't seem urgent to me, but getting a release out is urgent.

Ok, yeah, you're right.  Getting rid of the code to strip off the = solves this.  Thanks.  I've checked in the fix, after testing that it builds properly on the beta.  Everything should be fine now.

On Fri, Jun 11, 2010 at 4:17 PM, Don Clugston <dclugston at googlemail.com>wrote:

> On 11 June 2010 18:50, David Simcha <dsimcha at gmail.com> wrote:
> > Ok, I see why.  I had 2.046 and the beta installed concurrently, and I forgot to point my path variable at the beta before building Phobos.  Now
> I
> > can reproduce this, both using the Phobos build scripts and compiling the file individually with unittests turned on from Codeblocks.  It still
> works
> > fine on 2.046, though I did notice that DMD does take a while to (successfully) compile it and uses ~60 MB of memory.  It looks like a compiler regression, or else CustomFloat was relying on a bug that got fixed.
>
> It seems to be a recursive mixin. Every time it's instantiated, it
> gets 4 bytes bigger.
> Seems to be triggered by opOpAssign; maybe it's the change to
> opOpAssign in this release: "+=" was passed to opOpAssign
> now it's "+". There's code in opOpAssign which is clearly wrong - it's
> stripping off the "=", which no longer exists.
>
> If you don't think you can fix this really quickly, I think we should
> remove it completely from this release, and wait until the next one.
> It doesn't seem urgent to me, but getting a release out is urgent.
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/phobos/attachments/20100611/603bdae7/attachment.html>