On 17 September 2013 01:33, Brad Roberts <braddr@puremagic.com> wrote:
> It looks like the auto-tester for gdc has been failing for quite a while in the phobos math.d tests.  Anyone looking at those results?
>
> http;//d.puremagic.com/test-results/?projectid=2 http://d.puremagic.com/test-results/platform-history.ghtml?projectid=2&os=Linux_32
>
> The history report is truncated at about a month.  It did pass at one point.
>

I did look, it seems only to be a bug with that version of gcc you are using (can't reproduce with 4.9).  IIRC, storing the result to a local is a workaround for the gcc codegen bug.


-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';

On 9/17/13 12:41 AM, Iain Buclaw wrote:
> On 17 September 2013 01:33, Brad Roberts <braddr@puremagic.com> wrote:
>> It looks like the auto-tester for gdc has been failing for quite a while in
>> the phobos math.d tests.  Anyone looking at those results?
>>
>> http;//d.puremagic.com/test-results/?projectid=2
>> http://d.puremagic.com/test-results/platform-history.ghtml?projectid=2&os=Linux_32
>>
>> The history report is truncated at about a month.  It did pass at one point.
>>
>
> I did look, it seems only to be a bug with that version of gcc you are
> using (can't reproduce with 4.9).  IIRC, storing the result to a local
> is a workaround for the gcc codegen bug.

Is 4.8.1 not good enough?  That's the version that the auto-tester is building against.  I can grab 4.9 and give that a whirl.

On 1 October 2013 04:54, Brad Roberts <braddr@puremagic.com> wrote:
> On 9/17/13 12:41 AM, Iain Buclaw wrote:
>>
>> On 17 September 2013 01:33, Brad Roberts <braddr@puremagic.com> wrote:
>>>
>>> It looks like the auto-tester for gdc has been failing for quite a while
>>> in
>>> the phobos math.d tests.  Anyone looking at those results?
>>>
>>> http;//d.puremagic.com/test-results/?projectid=2
>>>
>>> http://d.puremagic.com/test-results/platform-history.ghtml?projectid=2&os=Linux_32
>>>
>>> The history report is truncated at about a month.  It did pass at one point.
>>>
>>
>> I did look, it seems only to be a bug with that version of gcc you are using (can't reproduce with 4.9).  IIRC, storing the result to a local is a workaround for the gcc codegen bug.
>
>
> Is 4.8.1 not good enough?  That's the version that the auto-tester is building against.  I can grab 4.9 and give that a whirl.
>

I can give it a whirl on 4.8.2 to see if that fixes the gcc bug.


-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';

On 9/30/13 8:59 PM, Iain Buclaw wrote:
> On 1 October 2013 04:54, Brad Roberts <braddr@puremagic.com> wrote:
>> On 9/17/13 12:41 AM, Iain Buclaw wrote:
>>>
>>> On 17 September 2013 01:33, Brad Roberts <braddr@puremagic.com> wrote:
>>>>
>>>> It looks like the auto-tester for gdc has been failing for quite a while
>>>> in
>>>> the phobos math.d tests.  Anyone looking at those results?
>>>>
>>>> http;//d.puremagic.com/test-results/?projectid=2
>>>>
>>>> http://d.puremagic.com/test-results/platform-history.ghtml?projectid=2&os=Linux_32
>>>>
>>>> The history report is truncated at about a month.  It did pass at one
>>>> point.
>>>>
>>>
>>> I did look, it seems only to be a bug with that version of gcc you are
>>> using (can't reproduce with 4.9).  IIRC, storing the result to a local
>>> is a workaround for the gcc codegen bug.
>>
>>
>> Is 4.8.1 not good enough?  That's the version that the auto-tester is
>> building against.  I can grab 4.9 and give that a whirl.
>>
>
> I can give it a whirl on 4.8.2 to see if that fixes the gcc bug.
>
>

Attempting with 4.9-20130929 failed:

    http://d.puremagic.com/test-results/test_data.ghtml?projectid=2&runid=51650&logid=12

In case something about the make process changed, here's that block of the script:

roughly:
  mkdir workdir
  cd workdir
  git clone <gdc master branch url>
  cd GDC

exactly (redirect to log file removed):
  #GCC_VER=4.8.1
  GCC_VER=4.9-20130929
  tar jxf ../../src/gcc-$GCC_VER.tar.bz2
  ./setup-gcc.sh gcc-$GCC_VER
  mkdir output-dir
  cd output-dir
  ../gcc-$GCC_VER/configure --disable-bootstrap --enable-languages=d --prefix=`pwd`/install-dir
  make -j3

On Oct 1, 2013 8:08 AM, "Brad Roberts" <braddr@puremagic.com> wrote:
>
> On 9/30/13 8:59 PM, Iain Buclaw wrote:
>>
>> On 1 October 2013 04:54, Brad Roberts <braddr@puremagic.com> wrote:
>>>
>>> On 9/17/13 12:41 AM, Iain Buclaw wrote:
>>>>
>>>>
>>>> On 17 September 2013 01:33, Brad Roberts <braddr@puremagic.com> wrote:
>>>>>
>>>>>
>>>>> It looks like the auto-tester for gdc has been failing for quite a
while
>>>>> in
>>>>> the phobos math.d tests.  Anyone looking at those results?
>>>>>
>>>>> http;//d.puremagic.com/test-results/?projectid=2
>>>>>
>>>>>
http://d.puremagic.com/test-results/platform-history.ghtml?projectid=2&os=Linux_32
>>>>>
>>>>> The history report is truncated at about a month.  It did pass at one point.
>>>>>
>>>>
>>>> I did look, it seems only to be a bug with that version of gcc you are using (can't reproduce with 4.9).  IIRC, storing the result to a local is a workaround for the gcc codegen bug.
>>>
>>>
>>>
>>> Is 4.8.1 not good enough?  That's the version that the auto-tester is building against.  I can grab 4.9 and give that a whirl.
>>>
>>
>> I can give it a whirl on 4.8.2 to see if that fixes the gcc bug.
>>
>>
>
> Attempting with 4.9-20130929 failed:
>
>
http://d.puremagic.com/test-results/test_data.ghtml?projectid=2&runid=51650&logid=12
>
> In case something about the make process changed, here's that block of
the script:
>
> roughly:
>   mkdir workdir
>   cd workdir
>   git clone <gdc master branch url>
>   cd GDC
>
> exactly (redirect to log file removed):
>   #GCC_VER=4.8.1
>   GCC_VER=4.9-20130929
>   tar jxf ../../src/gcc-$GCC_VER.tar.bz2
>   ./setup-gcc.sh gcc-$GCC_VER
>   mkdir output-dir
>   cd output-dir
>   ../gcc-$GCC_VER/configure --disable-bootstrap --enable-languages=d
--prefix=`pwd`/install-dir
>   make -j3
>
>

All that I can think of is that flex isn't installed on the autotester box?

Regards
-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';

On 10/1/13 3:00 AM, Iain Buclaw wrote:
>
> All that I can think of is that flex isn't installed on the autotester box?
>
> Regards
> --
> Iain Buclaw
>
> *(p < e ? p++ : p) = (c & 0x0f) + '0';

Indeed it wasn't.  However, shouldn't:

a) the 4.8.1 build have failed for the same reason?
b) a configure check have found that before getting that deep in the build process?

Anyway, I installed flex and have triggered a fresh attempt at the 4.9 snapshot build.  We'll see what else fails in a little while.

On 10/1/13 9:43 AM, Brad Roberts wrote:
> On 10/1/13 3:00 AM, Iain Buclaw wrote:
>>
>> All that I can think of is that flex isn't installed on the autotester box?
>>
>> Regards
>> --
>> Iain Buclaw
>>
>> *(p < e ? p++ : p) = (c & 0x0f) + '0';
>
> Indeed it wasn't.  However, shouldn't:
>
> a) the 4.8.1 build have failed for the same reason?
> b) a configure check have found that before getting that deep in the build process?
>
> Anyway, I installed flex and have triggered a fresh attempt at the 4.9 snapshot build.  We'll see
> what else fails in a little while.

The fresh attempt with 4.9-2013-0929 failed in exactly the same unit test as 4.8.1 failed.

core.exception.AssertError@../../../../gcc-4.9-20130929/libphobos/src/std/math.d(1947): unittest failure

http://d.puremagic.com/test-results/test_data.ghtml?projectid=2&runid=51662&logid=13

On Oct 2, 2013 1:50 AM, "Brad Roberts" <braddr@puremagic.com> wrote:
>
> On 10/1/13 9:43 AM, Brad Roberts wrote:
>>
>> On 10/1/13 3:00 AM, Iain Buclaw wrote:
>>>
>>>
>>> All that I can think of is that flex isn't installed on the autotester
box?
>>>
>>> Regards
>>> --
>>> Iain Buclaw
>>>
>>> *(p < e ? p++ : p) = (c & 0x0f) + '0';
>>
>>
>> Indeed it wasn't.  However, shouldn't:
>>
>> a) the 4.8.1 build have failed for the same reason?
>> b) a configure check have found that before getting that deep in the
build process?
>>
>> Anyway, I installed flex and have triggered a fresh attempt at the 4.9
snapshot build.  We'll see
>> what else fails in a little while.
>
>
> The fresh attempt with 4.9-2013-0929 failed in exactly the same unit test
as 4.8.1 failed.
>
> core.exception.AssertError@../../../../gcc-4.9-20130929/libphobos/src/std/math.d(1947):
unittest failure
>
>
http://d.puremagic.com/test-results/test_data.ghtml?projectid=2&runid=51662&logid=13
>
>

Now why doesn't that fail on either of my laptops (one x86, other x86_64)
or my dev server (x86_64 vm)...

Regards
-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';