[ITK-dev] [ITK-users] [ANN] ITK 4.9 Release Candidate 1 is ready for testing!

[Simon Alexander]

I've sent a patch set to gerrit, managed to screw it up somehow and
generated 4 reviews instead of 1, whups.  Matt, you're on them not sure who
else should be.

One of them is just a script modification so that my bash port likes your
utilities script, just a proposal.

The other three address each error.  Two are a loop fusion issue where
mixed type comparison is being generated incorrectly, changing the bound
type fixes this.

More concerning: the third change is an expansion issue with itk::Size<3>
 where it doesn't get the bound right for a .Fill call and never
initializes the last element.

Apologies  for the delay, i'm hesitant to blame things on the compiler
until I've convinced myself I know what is going on.  TODO generate
simplified versions to file with VS team.

cheers,
Simon

On Mon, 7 Dec 2015 at 13:53 Simon Alexander <skalexander at gmail.com> wrote:

> Hi Matt,
>
> Quick update:  I've found two compiler bugs so far - not great.   On
> hopefully better news, I've also found that update 1 has shipped.  So I
> will configure that and run it against your RC2 to see if these things are
> resolved.  If not I have workarounds, unsure of the wisdom of adding them.
>
>
> On Fri, 4 Dec 2015 at 17:34 Matt McCormick <matt.mccormick at kitware.com>
> wrote:
>
>> Hi Simon.
>>
>> Apologies I haven't been able to put much time into this, ran into some
>>> interesting challenges with VS2015 itself (at least in my configuration)
>>> hanging permanently when changing from debug to release mode, or vice
>>> versa.  It seems the generated project files are a bit of a challenge for
>>> it.
>>>
>>
>> Yes, Visual Studio has difficulty with large projects that have many
>> build targets. With the 4.9 release, we addressed the behavior a bit by
>> reducing the number of header tests and by turning BUILD_EXAMPLES OFF by
>> default.
>>
>> By setting ITK_BUILD_DEFAULT_MODULES to OFF, responsiveness can be
>> improved when focusing on development of a subset of modules.
>>
>>
>>
>>> At any rate, I have narrowed down part of the issue at least to being
>>> not in the interpolate filter itself, but rather in the test driver code
>>> ... somewhere in the nested loops incrementing the underlying array
>>> elements is behaving incorrectly in release mode.  I'll try and have  a
>>> closer look this weekend.
>>>
>>
>>  Thanks for taking a look at this!
>>
>> In the past, we ran into some issues with Visual Studio in Release mode
>> where adding const or using a reference for the returned type avoided a
>> faulty optimization. Bug reports to upstream Visual Studio have also helped
>> resolve the issue in some cases, too.
>>
>> Thanks,
>> Matt
>>
>>
>>
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-developers/attachments/20151208/5f05ee93/attachment.html>