Agree with brad plus: we want to establish a good level of confidence with this release. I'm not saying we should fix every bug out there, but if there's obvious stuff on the table it's better to get on it instead if releasing.

Sent by shouting through my showerhead.

On Feb 18, 2011, at 12:13 PM, Walter Bright <walter at digitalmars.com> wrote:

> 
> 
> Don Clugston wrote:
>> I've found something pretty bad. On Windows, the unittests for both
>> Phobos and druntime have been run with -release!!!
>> This would suggest that they haven't actually been run properly at all!
>> Removing -release causes failures in std.conv and std.json, but I
>> think both are the same problem:
>> to!float( "123e+2" ) is broken.
>> 
>> 
> 
> At this point, I think we could go on forever with fixing just one more problem. We need to do a release and then get to work fixing more of them. I figure there will be a slew of 64 bit problems.
> _______________________________________________
> dmd-beta mailing list
> dmd-beta at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-beta

On 18 February 2011 11:01, Brad Roberts <braddr at puremagic.com> wrote:
> That's pretty embarassing. ?I haven't dug yet, maybe you have already. ?How long has the makefile been broken like that? ?If not long, I'd vote for holding the release and fixing the makefile and the failing tests.
>
> Also, is the bigint bug a regression or an old bug just recently found? ?If it's a regression, I'd also vote for fixing it. ?I really don't like shipping with new regressions.

None of these things are regressions. I'm working on some of the BigInt issues.

Wait.. It has -release, but it also has -unittest which overrides -release when it comes to asserts.  That means that the only issue would be files that are linked in from the .lib file rather than being rebuilt as part of the unittest.exe build.  Or am I overlooking something again?

On Fri, 18 Feb 2011, Don Clugston wrote:

> I've found something pretty bad. On Windows, the unittests for both
> Phobos and druntime have been run with -release!!!
> This would suggest that they haven't actually been run properly at all!
> Removing -release causes failures in std.conv and std.json, but I
> think both are the same problem:
> to!float( "123e+2" ) is broken.
> 
> 
> On 18 February 2011 07:20, Walter Bright <walter at digitalmars.com> wrote:
> >
> > Thanks everyone for their hard work on this beta. I'm going to release it late tonight unless something terrible comes up.

On 19 February 2011 00:38, Brad Roberts <braddr at puremagic.com> wrote:
> Wait.. It has -release, but it also has -unittest which overrides -release when it comes to asserts. ?That means that the only issue would be files that are linked in from the .lib file rather than being rebuilt as part of the unittest.exe build. ?Or am I overlooking something again?

That's what I originally thought, too, but I found in practice that it
doesn't work - not properly, anyway. Removing -release makes tests
fail.
The makefile also needs a way to make a Phobos debug build. A couple
of horrific BigInt bugs slipped through because I had assumed that if
you do a -unittest build,'
you get asserts enabled. But you don't. My test cases should have been
triggering asserts.

> On Fri, 18 Feb 2011, Don Clugston wrote:
>
>> I've found something pretty bad. On Windows, the unittests for both
>> Phobos and druntime have been run with -release!!!
>> This would suggest that they haven't actually been run properly at all!
>> Removing -release causes failures in std.conv and std.json, but I
>> think both are the same problem:
>> to!float( "123e+2" ) is broken.
>>
>>
>> On 18 February 2011 07:20, Walter Bright <walter at digitalmars.com> wrote:
>> >
>> > Thanks everyone for their hard work on this beta. I'm going to release it late tonight unless something terrible comes up.
> _______________________________________________
> dmd-beta mailing list
> dmd-beta at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-beta
>

On 2/18/2011 10:07 PM, Don Clugston wrote:
> On 19 February 2011 00:38, Brad Roberts <braddr at puremagic.com> wrote:
>> Wait.. It has -release, but it also has -unittest which overrides -release when it comes to asserts.  That means that the only issue would be files that are linked in from the .lib file rather than being rebuilt as part of the unittest.exe build.  Or am I overlooking something again?
> 
> That's what I originally thought, too, but I found in practice that it
> doesn't work - not properly, anyway. Removing -release makes tests
> fail.
> The makefile also needs a way to make a Phobos debug build. A couple
> of horrific BigInt bugs slipped through because I had assumed that if
> you do a -unittest build,'
> you get asserts enabled. But you don't. My test cases should have been
> triggering asserts.
> 

Obviously.. so that's another bug to worry about.

Also, it also means that posix is doing something differently since it's not seeing the problems that windows does.

Don Clugston wrote:
> I've found something pretty bad. On Windows, the unittests for both
> Phobos and druntime have been run with -release!!!
> This would suggest that they haven't actually been run properly at all!
> Removing -release causes failures in std.conv and std.json, but I
> think both are the same problem:
> to!float( "123e+2" ) is broken.
>
> 

That one turns out to be an optimizer bug. I'm working on a fix.