Moreover how is it possible FreeBSD32 and OSX64 fail on master, but they pass with some pull requests, including mine even though it doesn't change any semantics?
_______________________________________________
dmd-internals mailing list
dmd-internals@puremagic.com
http://lists.puremagic.com/mailman/listinfo/dmd-internals

On May 11, 2012, at 4:06 AM, mrmocool@gmx.de wrote:

> Moreover how is it possible FreeBSD32 and OSX64 fail on master, but they pass with some pull requests, including mine even though it doesn't change any semantics?

Some bugs are related to the alignment of code segments and the like. Simply inserting a line of code or rearranging declarations may change the behavior. Since this fix was related to code generation, it's entirely possible something like this is happening.
_______________________________________________
dmd-internals mailing list
dmd-internals@puremagic.com
http://lists.puremagic.com/mailman/listinfo/dmd-internals

In the case of the freebsd failure, it was dmd itself crashing, not the generated code.  Walter's most recent checkin seems to have made it go away, though I'm not yet convinced the changes he made actually fix the problem.  Since I can reproduce it fairly easily, I'm going to continue to coax out a minimal test case of the problem to make sure.

For the OSX failure, it's one I've seen before, but it went away -- I don't remember if there was a direct fix at the time or not.  But it's definitly repeatable and needs attention.  It's not clear which layer is at fault, though quite possibly dmd since it's where there's been recent changes.

Later,
Brad

On Fri, 11 May 2012, Sean Kelly wrote:

> Date: Fri, 11 May 2012 14:15:38 -0700
> From: Sean Kelly <sean@invisibleduck.org>
> Reply-To: Discuss the internals of DMD <dmd-internals@puremagic.com>
> To: Discuss the internals of DMD <dmd-internals@puremagic.com>
> Subject: Re: [dmd-internals] master branch brokenness
> 
> On May 11, 2012, at 4:06 AM, mrmocool@gmx.de wrote:
> 
> > Moreover how is it possible FreeBSD32 and OSX64 fail on master, but they pass with some pull requests, including mine even though it doesn't change any semantics?
> 
> Some bugs are related to the alignment of code segments and the like. Simply inserting a line of code or rearranging declarations may change the behavior. Since this fix was related to code generation, it's entirely possible something like this is happening.
> _______________________________________________
> dmd-internals mailing list
> dmd-internals@puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-internals
> 
_______________________________________________
dmd-internals mailing list
dmd-internals@puremagic.com
http://lists.puremagic.com/mailman/listinfo/dmd-internals

On 5/11/2012 2:33 PM, Brad Roberts wrote:

> For the OSX failure, it's one I've seen before, but it went away -- I don't remember if there was a direct fix at the time or not. But it's definitly repeatable and needs attention. It's not clear which layer is at fault, though quite possibly dmd since it's where there's been recent changes.


It does work on an older version of the Phobos source. I'm going to upgrade my copy of Phobos and see.
_______________________________________________
dmd-internals mailing list
dmd-internals@puremagic.com
http://lists.puremagic.com/mailman/listinfo/dmd-internals