round up of more disastrous 64 bit bugs fixed

http://ftp.digitalmars.com/dmd1beta.zip http://ftp.digitalmars.com/dmd2beta.zip

Great.  To everyone:  So that we have a successful release, I suggest putting this compiler though the paces on any code you can find today, before it's officially released.  That's my only secret to uncovering all these bugs:  I spent a whole bunch of time compiling dstats and std.parallelism and (to a lesser extent) plot2kill, isolating bugs and filing reports.  Anyone who has both the time and a codebase with good unit test coverage that they know works on 32-bit, please do the same.

On 2/16/2011 4:46 AM, Walter Bright wrote:
> round up of more disastrous 64 bit bugs fixed
>
> http://ftp.digitalmars.com/dmd1beta.zip
> http://ftp.digitalmars.com/dmd2beta.zip
> _______________________________________________
> dmd-beta mailing list
> dmd-beta at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-beta
>

An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/dmd-beta/attachments/20110216/d9d5500e/attachment.html>

BTW, --no-warn-search-mismatch fails on ancient versions of the linker from some crufty old Linux distros that my sysadmin runs.  I don't think it's a big deal, b/c we're probably the only people in the world left running these distros (the kernel is 2.6.9 from 2004), but maybe a note about this would be a good thing.

On Wed, Feb 16, 2011 at 9:17 AM, David Simcha <dsimcha at gmail.com> wrote:

>  Awesome.  I just retested on a total of ~30-40 kloc from various projects
> I've worked on and, as far as I can tell, *everything works*.  One
> question, though.  This is a how-to question, not a bug or anything.
> Sometimes when I don't need huge address space I may want to be able to
> build in both 32 mode and 64 mode, benchmark and use whichever's faster.
> (32 mode may be faster b/c it wastes less cache space, 64 may be faster b/c
> it has more registers).  I see for Phobos you have managed to get the linker
> to ignore 32-bit libs that it's passed when in 64-bit mode and vice-versa.
> However, when I pass other 64-bit libs in 32-bit mode and vice-versa,
> linking still fails.  Is there something I need to do besides have the
> --no-warn-search-mismatch flag in dmd.conf to get this to work?
>
>
> On 2/16/2011 4:46 AM, Walter Bright wrote:
>
> round up of more disastrous 64 bit bugs fixed
>
> http://ftp.digitalmars.com/dmd1beta.zip
> http://ftp.digitalmars.com/dmd2beta.zip
> _______________________________________________
> dmd-beta mailing list
> dmd-beta at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-beta
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/dmd-beta/attachments/20110216/6a15fd4d/attachment.html>

Never mind, I figured it out.  You just have to pass the linker the name of the lib, put it in a lib directory, etc.

Correct:

Have a library named libfoo.a in lib32 and one in lib64.

Pass -L-lfoo

Incorrect:
Have a library named foo32.a and one named foo64.a

Pass -L/somedirectory/foo32.a -L/somedirectory/foo64.a

On Wed, Feb 16, 2011 at 9:17 AM, David Simcha <dsimcha at gmail.com> wrote:

>  Awesome.  I just retested on a total of ~30-40 kloc from various projects
> I've worked on and, as far as I can tell, *everything works*.  One
> question, though.  This is a how-to question, not a bug or anything.
> Sometimes when I don't need huge address space I may want to be able to
> build in both 32 mode and 64 mode, benchmark and use whichever's faster.
> (32 mode may be faster b/c it wastes less cache space, 64 may be faster b/c
> it has more registers).  I see for Phobos you have managed to get the linker
> to ignore 32-bit libs that it's passed when in 64-bit mode and vice-versa.
> However, when I pass other 64-bit libs in 32-bit mode and vice-versa,
> linking still fails.  Is there something I need to do besides have the
> --no-warn-search-mismatch flag in dmd.conf to get this to work?
>
>
> On 2/16/2011 4:46 AM, Walter Bright wrote:
>
> round up of more disastrous 64 bit bugs fixed
>
> http://ftp.digitalmars.com/dmd1beta.zip
> http://ftp.digitalmars.com/dmd2beta.zip
> _______________________________________________
> dmd-beta mailing list
> dmd-beta at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-beta
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/dmd-beta/attachments/20110216/2b676423/attachment-0001.html>

Unfortunately I just found two new regressions:

http://d.puremagic.com/issues/show_bug.cgi?id=5595 http://d.puremagic.com/issues/show_bug.cgi?id=5596

On Wed, Feb 16, 2011 at 10:15 AM, David Simcha <dsimcha at gmail.com> wrote:

> Never mind, I figured it out.  You just have to pass the linker the name of the lib, put it in a lib directory, etc.
>
> Correct:
>
> Have a library named libfoo.a in lib32 and one in lib64.
>
> Pass -L-lfoo
>
> Incorrect:
> Have a library named foo32.a and one named foo64.a
>
> Pass -L/somedirectory/foo32.a -L/somedirectory/foo64.a
>
> On Wed, Feb 16, 2011 at 9:17 AM, David Simcha <dsimcha at gmail.com> wrote:
>
>>  Awesome.  I just retested on a total of ~30-40 kloc from various projects
>> I've worked on and, as far as I can tell, *everything works*.  One
>> question, though.  This is a how-to question, not a bug or anything.
>> Sometimes when I don't need huge address space I may want to be able to
>> build in both 32 mode and 64 mode, benchmark and use whichever's faster.
>> (32 mode may be faster b/c it wastes less cache space, 64 may be faster b/c
>> it has more registers).  I see for Phobos you have managed to get the linker
>> to ignore 32-bit libs that it's passed when in 64-bit mode and vice-versa.
>> However, when I pass other 64-bit libs in 32-bit mode and vice-versa,
>> linking still fails.  Is there something I need to do besides have the
>> --no-warn-search-mismatch flag in dmd.conf to get this to work?
>>
>>
>> On 2/16/2011 4:46 AM, Walter Bright wrote:
>>
>> round up of more disastrous 64 bit bugs fixed
>>
>> http://ftp.digitalmars.com/dmd1beta.zip
>> http://ftp.digitalmars.com/dmd2beta.zip
>> _______________________________________________
>> dmd-beta mailing list
>> dmd-beta at puremagic.com
>> http://lists.puremagic.com/mailman/listinfo/dmd-beta
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puremagic.com/pipermail/dmd-beta/attachments/20110216/83e269d9/attachment.html>

Oh no, 5 messages from David, do I dare read them? <g>

David Simcha wrote:
> Great.  To everyone:  So that we have a successful release, I suggest putting this compiler though the paces on any code you can find today, before it's officially released.  That's my only secret to uncovering all these bugs:  I spent a whole bunch of time compiling dstats and std.parallelism and (to a lesser extent) plot2kill, isolating bugs and filing reports.  Anyone who has both the time and a codebase with good unit test coverage that they know works on 32-bit, please do the same.
>

I'd like to add that David has done an awesome job isolating and reporting bugs (building on all the work Brad has done). I know it's hard (and tedious) to isolate failures down from a large codebase. I appreciate it, and it has been a big help in getting the 64 bit compiler ready to rock.

I don't know of another way to make it work cleanly. At least those users can adjust their dmd.conf as necessary to accommodate older linkers.

David Simcha wrote:
> BTW, --no-warn-search-mismatch fails on ancient versions of the linker from some crufty old Linux distros that my sysadmin runs.  I don't think it's a big deal, b/c we're probably the only people in the world left running these distros (the kernel is 2.6.9 from 2004), but maybe a note about this would be a good thing.
>

David Simcha wrote:
> Unfortunately I just found two new regressions:
>
> http://d.puremagic.com/issues/show_bug.cgi?id=5595 http://d.puremagic.com/issues/show_bug.cgi?id=5596
>

Oh, damn. Just when I got to the last of your messages!