Who wants Borland support?
In 1.33.1, Boost.Parameter supported Borland. In 1.34, we are running into compiler bugs that we don't know how to avoid. Somehow the compiler is mis-deducing the return type of operator[] on line 328 of boost/parameter/aux_/arg_list.hpp (see http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/developer/is...) If we can't find a solution to this problem very soon we'll have to mark these tests as expected failures and drop support of Borland for 1.34, and we're stumped. I'm hoping someone who cares about Borland will help us come up with an answer. Thanks, -- Dave Abrahams Boost Consulting www.boost-consulting.com
On Sat, 22 Jul 2006 08:52:33 -0400, David Abrahams <dave@boost-consulting.com> wrote:
If we can't find a solution to this problem very soon we'll have to mark these tests as expected failures and drop support of Borland for 1.34, and we're stumped
Dave, wouldn't "support for Borland compilers older than BCB2006" be dropped anyway after 1.34? -- [ Gennaro Prota, C++ developer for hire ]
Gennaro Prota <gennaro_prota@yahoo.com> writes:
On Sat, 22 Jul 2006 08:52:33 -0400, David Abrahams <dave@boost-consulting.com> wrote:
If we can't find a solution to this problem very soon we'll have to mark these tests as expected failures and drop support of Borland for 1.34, and we're stumped
Dave, wouldn't "support for Borland compilers older than BCB2006" be dropped anyway after 1.34?
I don't know, and I'm not sure it's relevant. -- Dave Abrahams Boost Consulting www.boost-consulting.com
On Sat, 22 Jul 2006 19:55:50 -0400, David Abrahams <dave@boost-consulting.com> wrote:
Dave, wouldn't "support for Borland compilers older than BCB2006" be dropped anyway after 1.34?
I don't know, and I'm not sure it's relevant.
Well, for sure anything below 0x581 is deprecated starting from 1.34.0. I had a look at the Boost.Parameter code in the hope that I could find a workaround but honestly I think there's no hope. For a while boost developers have done heroic efforts to support compilers that not even their vendors cared about. Some of these went ahead and improved. Some others didn't, and I don't believe it's our job to reawaken dead bodies (and this is from someone who liked Borland, years ago). A similar thing is happening with Digital Mars. I've recently found that when faced with something like: template <typename Integer> void f(Integer x); template <typename T> void f(T x) { ...} it thinks they are two different templates and if you attempt a call it yields a message along the lines of "ambiguous call: could be f(Integer) or f(T)". Probably it is internally treating the declaration as a string and comparing the supposed parameter-list substrings, who knows. The point is... it is a "young" compiler, though it exists for many years now, and it's noticeably improving, so we try supporting it. But if it doesn't "keep up" and meet some minimum quality requirements in one year or so, it is deemed. I expect boost code to exploit more and more sophisticated techniques in the next years. And make use of the new C++0x features. And then relying on TR2... Well, you got the point. -- [ Gennaro Prota, C++ developer for hire ]
Gennaro Prota <gennaro_prota@yahoo.com> writes:
On Sat, 22 Jul 2006 19:55:50 -0400, David Abrahams <dave@boost-consulting.com> wrote:
Dave, wouldn't "support for Borland compilers older than BCB2006" be dropped anyway after 1.34?
I don't know, and I'm not sure it's relevant.
Well, for sure anything below 0x581 is deprecated starting from 1.34.0.
That doesn't gibe with http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/developer/is..., which seems to indicate that Borland is supposed to be supported. -- Dave Abrahams Boost Consulting www.boost-consulting.com
On Sat, 22 Jul 2006 23:19:13 -0400, David Abrahams <dave@boost-consulting.com> wrote:
Well, for sure anything below 0x581 is deprecated starting from 1.34.0.
That doesn't gibe with http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/developer/is..., which seems to indicate that Borland is supposed to be supported.
Not sure what you mean. In any case you know much better than me that deprecated doesn't mean not supported, just possibly no more supported in the future. For the record, my claim comes from config/compiler/borland.hpp -- [ Gennaro Prota, C++ developer for hire ]
On Sat, 22 Jul 2006 23:19:13 -0400, David Abrahams <dave@boost-consulting.com> wrote:
That doesn't gibe with http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/developer/is..., which seems to indicate that Borland is supposed to be supported.
BTW, those pages are driving me crazy. One day I see Comeau, another day I don't. I've fixed 50% of the Digital Mars failures and, now that I wanted to attempt a complete support, it's no more there. Am I missing something or the set of supported compiler/lib/platform combinations is a bit erratic? Furthermore, may I ask what does "STLport 5.1.0" without further qualifications mean? I have RC1 and usually I'm not even able to reproduce the error messages. Hope we are not testing against their CVS. -- [ Gennaro Prota, C++ developer for hire ]
Gennaro Prota wrote:
Furthermore, may I ask what does "STLport 5.1.0" without further qualifications mean? I have RC1 and usually I'm not even able to reproduce the error messages. Hope we are not testing against their CVS.
I don't know what it means for others, but for the tests I run "RSI Droid2K V2" they are indeed from the STLport CVS. I'll put a note to that effect on my info page :-) -- -- Grafik - Don't Assume Anything -- Redshift Software, Inc. - http://redshift-software.com -- rrivera/acm.org - grafik/redshift-software.com -- 102708583/icq - grafikrobot/aim - grafikrobot/yahoo
On Sun, 23 Jul 2006 10:25:50 -0500, Rene Rivera <grafikrobot@gmail.com> wrote:
Gennaro Prota wrote:
Furthermore, may I ask what does "STLport 5.1.0" without further qualifications mean? [...]
I don't know what it means for others, but for the tests I run "RSI Droid2K V2" they are indeed from the STLport CVS. I'll put a note to that effect on my info page :-)
Yes, thanks. But you only test the config library with that, don't you? PS: anyone else beginning to thing that we need a (simple) database for all the regression testing stuff? -- [ Gennaro Prota, C++ developer for hire ]
Gennaro Prota wrote:
On Sun, 23 Jul 2006 10:25:50 -0500, Rene Rivera <grafikrobot@gmail.com> wrote:
Furthermore, may I ask what does "STLport 5.1.0" without further qualifications mean? [...] I don't know what it means for others, but for the tests I run "RSI Droid2K V2" they are indeed from the STLport CVS. I'll put a note to
Gennaro Prota wrote: that effect on my info page :-)
Yes, thanks. But you only test the config library with that, don't you?
Nope. Those are full runs.
PS: anyone else beginning to thing that we need a (simple) database for all the regression testing stuff?
Yes. -- -- Grafik - Don't Assume Anything -- Redshift Software, Inc. - http://redshift-software.com -- rrivera/acm.org - grafik/redshift-software.com - 102708583/icq
Gennaro Prota writes:
On Sat, 22 Jul 2006 23:19:13 -0400, David Abrahams <dave@boost-consulting.com> wrote:
That doesn't gibe with http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/developer/is..., which seems to indicate that Borland is supposed to be supported.
BTW, those pages are driving me crazy. One day I see Comeau, another day I don't. I've fixed 50% of the Digital Mars failures and, now that I wanted to attempt a complete support, it's no more there.
The presence of a particular compiler on the issues page is determined by whether the corresponding toolset is declared as "required" (http://lists.boost.org/MailArchives/boost-testing/msg00333.php). AFAIK, Digital Mars was ever on the list of required toolsets; I _think_ its temporary presence on the page was cause by a recent breakage of the explicit markup file.
Am I missing something or the set of supported compiler/lib/platform combinations is a bit erratic?
It's not. It's definitely not being changed without a notice to the list.
Furthermore, may I ask what does "STLport 5.1.0" without further qualifications mean?
In "dmc-8_49b-stlport-5_1_0", it's STLport 5.1.0 RC1.
I have RC1 and usually I'm not even able to reproduce the error messages.
Hmm, let us know if we can help to track these down. -- Aleksey Gurtovoy MetaCommunications Engineering
David Abrahams <dave@boost-consulting.com> writes:
In 1.33.1, Boost.Parameter supported Borland. In 1.34, we are running into compiler bugs that we don't know how to avoid. Somehow the compiler is mis-deducing the return type of operator[] on line 328 of boost/parameter/aux_/arg_list.hpp (see http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/developer/is...)
If we can't find a solution to this problem very soon we'll have to mark these tests as expected failures and drop support of Borland for 1.34, and we're stumped. I'm hoping someone who cares about Borland will help us come up with an answer.
Now fixed. -- Dave Abrahams Boost Consulting www.boost-consulting.com
participants (4)
-
Aleksey Gurtovoy -
David Abrahams -
Gennaro Prota -
Rene Rivera