Any sunpro experts out there?
The sunpro type_traits tests are currently mostly failing because is_enum is broken, this in turn is likely to really be an is_convertible failure. There are a few possible preprocessor branches that could be taken for is_convertible, if some kind soul with access to sunpro would give these a try and see if any fix the errors that would be very welcome. Thanks, John.
The sunpro type_traits tests are currently mostly failing because is_enum is broken, this in turn is likely to really be an is_convertible failure.
There are a few possible preprocessor branches that could be taken for is_convertible, if some kind soul with access to sunpro would give these a try and see if any fix the errors that would be very welcome.
Which boost version? HEAD, 1.33.1, RC_1_34? Not a sunpro expert by any means but I have 6.2, 8.1, 10 and 11 for Sparc and a new test box with 11 for AMD, but I would say boost should ignore Sunpro prior to 11, really, since getting to build on 6.2 or 8.1 is close to impossible. Perhaps you could also write to Simon at Sun who is working on making Sunpro as boost compatible as possible, including any patches: http://blogs.sun.com/roller/page/sga. That may be an even better idea. Can you explain more about what you want done? A small test run or a patch? Best, Tom
Tomas Tomas Puverle wrote:
Which boost version? HEAD, 1.33.1, RC_1_34?
RC_1_34_0 I guess.
Not a sunpro expert by any means but I have 6.2, 8.1, 10 and 11 for Sparc and a new test box with 11 for AMD, but I would say boost should ignore Sunpro prior to 11, really, since getting to build on 6.2 or 8.1 is close to impossible.
Just a remark. When talking about sun compilers anything but the internal compiler versions makes my brain hurt. For RC_1_34_0 it's 5.8. Thomas -- Thomas Witt witt@acm.org
Tomas Puverle wrote:
The sunpro type_traits tests are currently mostly failing because is_enum is broken, this in turn is likely to really be an is_convertible failure.
There are a few possible preprocessor branches that could be taken for is_convertible, if some kind soul with access to sunpro would give these a try and see if any fix the errors that would be very welcome.
Which boost version? HEAD, 1.33.1, RC_1_34? Not a sunpro expert by any means but I have 6.2, 8.1, 10 and 11 for Sparc and a new test box with 11 for AMD, but I would say boost should ignore Sunpro prior to 11, really, since getting to build on 6.2 or 8.1 is close to impossible. Perhaps you could also write to Simon at Sun who is working on making Sunpro as boost compatible as possible, including any patches: http://blogs.sun.com/roller/page/sga. That may be an even better idea.
It does look like the way to go: actually he's claiming that all the type_traits tests pass in his blog, so either we're missing a patch, or our test compiler isn't up to date. I don't suppose you have an email address for the guy? I can't find one on the blog site. Thanks, John.
It does look like the way to go: actually he's claiming that all the type_traits tests pass in his blog, so either we're missing a patch, or our test compiler isn't up to date. I don't suppose you have an email address for the guy? I can't find one on the blog site.
No, I don't, but if I had to guess it would be sga@sun.com. However, he tends to be responsive to comments in his blog, probably worthwhile trying that.
On 6/8/06, Tomas Puverle <Tomas.Puverle@morganstanley.com> wrote:
It does look like the way to go: actually he's claiming that all the type_traits tests pass in his blog, so either we're missing a patch, or our test compiler isn't up to date. I don't suppose you have an email address for the guy? I can't find one on the blog site.
No, I don't, but if I had to guess it would be sga@sun.com.
See this posting for an e-mail address: http://article.gmane.org/gmane.comp.lib.mtl.devel/336
However, he tends to be responsive to comments in his blog, probably worthwhile trying that.
He's very responsive, if you're developing C++ on Solaris, it's worth visiting his blog from time to time to see what he's focussing on fixing.
John Maddock wrote:
The sunpro type_traits tests are currently mostly failing because is_enum is broken, this in turn is likely to really be an is_convertible failure.
There are a few possible preprocessor branches that could be taken for is_convertible, if some kind soul with access to sunpro would give these a try and see if any fix the errors that would be very welcome.
Well, I'm not an sunpro "expert" but I use the compilers at work and am sad enough to have a Sun workstation running Sol10 and the latest compiler here at home. I've been playing with boost on that for a while so I should be able to help. I'm just about to update the compiler on that box to the latest patch level so I'll give it a try against 1.34 RC on the weekend. I've seen the is_enum breakage on 1.33 and was planning to investigate shortly before all my spare time disappeared, but the code made my head hurt a tad... If you've got a static IP address at your end I may be able to give you a login on this box if you want to have a play with the compiler yourself. Regards, Timo
Timo Geusch wrote:
Well, I'm not an sunpro "expert" but I use the compilers at work and am sad enough to have a Sun workstation running Sol10 and the latest compiler here at home. I've been playing with boost on that for a while so I should be able to help.
I'm just about to update the compiler on that box to the latest patch level so I'll give it a try against 1.34 RC on the weekend. I've seen the is_enum breakage on 1.33 and was planning to investigate shortly before all my spare time disappeared, but the code made my head hurt a tad...
OK, if you could try the type_traits tests with the very latest sunpro patches applied that would be great: Sun claim to have fixed this, so if the latest patches do indeed make the problem go away, then that's an easy fix :-) Many thanks, John.
If you've got a static IP address at your end I may be able to give you a login on this box if you want to have a play with the compiler yourself.
Regards, Timo _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
John Maddock wrote:
Timo Geusch wrote:
Well, I'm not an sunpro "expert" but I use the compilers at work and am sad enough to have a Sun workstation running Sol10 and the latest compiler here at home. I've been playing with boost on that for a while so I should be able to help.
I'm just about to update the compiler on that box to the latest patch level so I'll give it a try against 1.34 RC on the weekend. I've seen the is_enum breakage on 1.33 and was planning to investigate shortly before all my spare time disappeared, but the code made my head hurt a tad...
OK, if you could try the type_traits tests with the very latest sunpro patches applied that would be great: Sun claim to have fixed this, so if the latest patches do indeed make the problem go away, then that's an easy fix :-)
Finally managed an attempt at building 1.34 RC with the latest patch set for SUN Studio 11 and run the tests against type_traits. For both build & test I used bjam v1, not v2 as I'm not yet familiar with v2 and the sunpro-tools.jam file I'm using is a tad different from the one in the distribution. My jam file based on a bunch of patches that Simon Atanasyan published in his blog for 1.32 and generally achieves better build and test results than the one that comes with Boost as that one is missing a couple of important flags. I'll prepare a patch tonight and would hope that it'll make it into 1.34... Anyway, the test results don't look that good - 11 failures and 15 targets skipped because it failed to build a couple of tests (test_extent and rank_test failed to build, a quick trawl through the test output suggests that several of the is_<something>_test case fail as well, namely is_object_test, is_member_func_test and is_reference_test. I'll try to provide full output in the next couple of days. Regards, Timo
John Maddock wrote:
Timo Geusch wrote:
Well, I'm not an sunpro "expert" but I use the compilers at work and am sad enough to have a Sun workstation running Sol10 and the latest compiler here at home. I've been playing with boost on that for a while so I should be able to help.
I'm just about to update the compiler on that box to the latest patch level so I'll give it a try against 1.34 RC on the weekend. I've seen the is_enum breakage on 1.33 and was planning to investigate shortly before all my spare time disappeared, but the code made my head hurt a tad...
OK, if you could try the type_traits tests with the very latest sunpro patches applied that would be great: Sun claim to have fixed this, so if the latest patches do indeed make the problem go away, then that's an easy fix :-)
Unfortunately it doesn't so it's not quite that easy a fix. Several tests don't compile and a couple of those that do, fail. I've put up the output of a test run at http://www.unixconsult.co.uk/programming/type_traits_test_sunpro.out. Please note that these currently are just the build failures as I thought that it may be better to fix those first and then address the test failures seperately. Regards, Timo
Timo Geusch wrote:
Unfortunately it doesn't so it's not quite that easy a fix. Several tests don't compile and a couple of those that do, fail.
I've put up the output of a test run at http://www.unixconsult.co.uk/programming/type_traits_test_sunpro.out. Please note that these currently are just the build failures as I thought that it may be better to fix those first and then address the test failures seperately.
Actually I'm less worried about those that don't compile, than those that compile and do the wrong thing: for example is_empty<function-type>::value being *true*, that could be really really bad, and I'm frankly at a complete loss to understand how that could possibly happen. Other failures like the is_abstract ones are expected if BOOST_NO_IS_ABSTRACT is set: which it is for compiler versions <= 0x580, so it looks like maybe the compiler version hasn't been updated from previous releases? The remaining failures are all function type related: looks like sunpro doesn't like partial specialisations on pointers to functions, or maybe something has gone wrong before the metacode gets that far it's impossible to say. Anyway I'm going to contact Simon Atanasyan direct and see if he has any suggestions. John.
On Sat, 2006-06-17 at 01:20 +0100, John Maddock wrote:
Timo Geusch wrote:
Unfortunately it doesn't so it's not quite that easy a fix. Several tests don't compile and a couple of those that do, fail.
I've put up the output of a test run at http://www.unixconsult.co.uk/programming/type_traits_test_sunpro.out. Please note that these currently are just the build failures as I thought that it may be better to fix those first and then address the test failures seperately.
Actually I'm less worried about those that don't compile, than those that compile and do the wrong thing: for example is_empty<function-type>::value being *true*, that could be really really bad, and I'm frankly at a complete loss to understand how that could possibly happen. Other failures like the is_abstract ones are expected if BOOST_NO_IS_ABSTRACT is set: which it is for compiler versions <= 0x580, so it looks like maybe the compiler version hasn't been updated from previous releases?
I think this is partially my fault - I believe that Simon Atanasyan also patched the compiler configuration header file and I didn't apply those changes. That said, the difference between 1.33.1 with his source patches applied and 1.34 RC with no additional patches is one more failure against 1.34. I'll rerun the tests with his changes applied and will report the results.
From reading Simon's blog I'm also under the impression that the compiler which succeeds with in type_traits tests is their repository version of the compiler and not the -02 patch version, but I may be wrong on that count.
The remaining failures are all function type related: looks like sunpro doesn't like partial specialisations on pointers to functions, or maybe something has gone wrong before the metacode gets that far it's impossible to say.
There was a comment in his blog somewhere that it still has problems with 'complex' template expressions. Maybe this one is complex enough.
Anyway I'm going to contact Simon Atanasyan direct and see if he has any suggestions.
I've seen your email and will add a bit more info for Simon once I'm done enjoying the decidedly unBritish weather. Regards, Timo
Timo Geusch wrote:
Anyway I'm going to contact Simon Atanasyan direct and see if he has any suggestions.
I've seen your email and will add a bit more info for Simon once I'm done enjoying the decidedly unBritish weather.
Thanks Timo, suggest you enjoy the weather while you can as it's supposed to brake today I believe :-( John.
Timo Geusch wrote:
Well, I'm not an sunpro "expert" but I use the compilers at work and am sad enough to have a Sun workstation running Sol10 and the latest compiler here at home. I've been playing with boost on that for a while so I should be able to help.
I'm just about to update the compiler on that box to the latest patch level so I'll give it a try against 1.34 RC on the weekend. I've seen the is_enum breakage on 1.33 and was planning to investigate shortly before all my spare time disappeared, but the code made my head hurt a tad...
OK, if you could try the type_traits tests with the very latest sunpro patches applied that would be great: Sun claim to have fixed this, so if the latest patches do indeed make the problem go away, then that's an easy fix :-) Many thanks, John.
If you've got a static IP address at your end I may be able to give you a login on this box if you want to have a play with the compiler yourself.
Regards, Timo _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
John Maddock wrote:
Timo Geusch wrote:
Well, I'm not an sunpro "expert" but I use the compilers at work and am sad enough to have a Sun workstation running Sol10 and the latest compiler here at home. I've been playing with boost on that for a while so I should be able to help.
I'm just about to update the compiler on that box to the latest patch level so I'll give it a try against 1.34 RC on the weekend. I've seen the is_enum breakage on 1.33 and was planning to investigate shortly before all my spare time disappeared, but the code made my head hurt a tad...
OK, if you could try the type_traits tests with the very latest sunpro patches applied that would be great: Sun claim to have fixed this, so if the latest patches do indeed make the problem go away, then that's an easy fix :-)
Will do, as soon as I have the workstation and this monitor here talking to each other.
Timo Geusch wrote:
John Maddock wrote:
OK, if you could try the type_traits tests with the very latest sunpro patches applied that would be great: Sun claim to have fixed this, so if the latest patches do indeed make the problem go away, then that's an easy fix :-)
Will do, as soon as I have the workstation and this monitor here talking to each other.
Well, they're talking now and after copying over my changes to the boost v1 jamfile for sunpro I've got the libraries building. I should have some results as to build failures etc tonight and hopefully I'll be able to run the type_traits tests tonight as well. Regards, Timo
participants (5)
-
John Maddock -
Michael van der Westhuizen -
Thomas Witt -
Timo Geusch -
Tomas Puverle