Do we have a revised timeline for 1.40 yet? I'm asking because the regex lib is ready for a substantial update (provided trunk tests cycle OK obviously!). Thanks, John.
John Maddock wrote:
Do we have a revised timeline for 1.40 yet?
I'm asking because the regex lib is ready for a substantial update (provided trunk tests cycle OK obviously!).
The release managers haven't discussed a timeline, but the release branch is still open for merges from trunk. If you feel good about it, then go for it (after the tests cycle). -- Eric Niebler BoostPro Computing http://www.boostpro.com
On 30 Jul 2009, at 10:43, Eric Niebler wrote:
John Maddock wrote:
Do we have a revised timeline for 1.40 yet? I'm asking because the regex lib is ready for a substantial update (provided trunk tests cycle OK obviously!).
The release managers haven't discussed a timeline, but the release branch is still open for merges from trunk. If you feel good about it, then go for it (after the tests cycle).
It was just reported to me that changed to the serialization library on the release branch completely break Boost.MPI there. Can I go ahead and fix this to make Boost.MPI on the release branch useable again? Matthias
Matthias Troyer wrote:
On 30 Jul 2009, at 10:43, Eric Niebler wrote:
John Maddock wrote:
Do we have a revised timeline for 1.40 yet? I'm asking because the regex lib is ready for a substantial update (provided trunk tests cycle OK obviously!).
The release managers haven't discussed a timeline, but the release branch is still open for merges from trunk. If you feel good about it, then go for it (after the tests cycle).
It was just reported to me that changed to the serialization library on the release branch completely break Boost.MPI there. Can I go ahead and fix this to make Boost.MPI on the release branch useable again?
Sure. Just follow the usual drill: fix it first on trunk, wait for the tests to cycle, then merge to release. Thanks. But why did changes to Serialization break MPI? Was MPI making use of undocumented parts of s11n, or has the s11n interface changed in a non-backward-compatible way? If the latter, I'm a bit concerned that such a breaking change would go in so close to release. -- Eric Niebler BoostPro Computing http://www.boostpro.com
On Thu, Aug 06, 2009 at 05:19:37PM -0400, Eric Niebler wrote:
Matthias Troyer wrote:
It was just reported to me that changed to the serialization library on the release branch completely break Boost.MPI there. Can I go ahead and fix this to make Boost.MPI on the release branch useable again?
Sure. Just follow the usual drill: fix it first on trunk, wait for the tests to cycle, then merge to release. Thanks.
Be sure to get someone to fix the test reporting. Currently, all testing with MPI is dropped to the floor. We have been speculating on boost-testing about the cause, but really need someone who knows the results web server to take a look. See http://lists.boost.org/boost-testing/2009/08/6355.php Thanks, -Steve
On 6 Aug 2009, at 17:46, Steve M. Robbins wrote:
On Thu, Aug 06, 2009 at 05:19:37PM -0400, Eric Niebler wrote:
Matthias Troyer wrote:
It was just reported to me that changed to the serialization library on the release branch completely break Boost.MPI there. Can I go ahead and fix this to make Boost.MPI on the release branch useable again?
Sure. Just follow the usual drill: fix it first on trunk, wait for the tests to cycle, then merge to release. Thanks.
It was already fixed on trunk a week ago and those who reported it there said that all works now. I have also confirmed it on my machine. Matthias
On 6 Aug 2009, at 15:19, Eric Niebler wrote:
Sure. Just follow the usual drill: fix it first on trunk, wait for the tests to cycle, then merge to release. Thanks.
But why did changes to Serialization break MPI? Was MPI making use of undocumented parts of s11n, or has the s11n interface changed in a non-backward-compatible way? If the latter, I'm a bit concerned that such a breaking change would go in so close to release.
Robert changed the implementation of pointer serialization. Now different headers need to be included and different templates instantiated in the source files for any archive to be used by Boost.Serialization. This is a breaking change and none of the archives created for releases before 1.40 will work out of the box with Boost 1.40. They all will have to be rewritten. This was reported to me on the trunk about 2 weeks ago and I fixed it there but I was not aware that he had moved it also to the release branch. I think he did not expect anyone to write their own archives, but Boost.MPI creates quite a number of different archive types. Matthias
2009/8/7 Matthias Troyer <troyer@phys.ethz.ch>:
Robert changed the implementation of pointer serialization. Now different headers need to be included and different templates instantiated in the source files for any archive to be used by Boost.Serialization. This is a breaking change and none of the archives created for releases before 1.40 will work out of the box with Boost 1.40. They all will have to be rewritten.
Please can someone write something about this for the release notes? Daniel
On 6 Aug 2009, at 15:19, Eric Niebler wrote:
Was MPI making use of undocumented parts of s11n, or has the s11n interface changed in a non-backward-compatible way? If the latter, I'm a bit concerned that such a breaking change would go in so close to release.
I'm not too concerned since nothing in Boost but Boost.MPI should be affected by that breaking change. One will have to make sure that users of Boost know that they'll have to update their archive sources though. Matthias
Matthias Troyer wrote:
On 6 Aug 2009, at 15:19, Eric Niebler wrote:
or has the s11n interface changed in a non-backward-compatible way?
no
Was MPI making use of undocumented parts of s11n,
MPI was leveraging on the implementation of the archive classes included with the package. Looks like good practice to me. But then this problem can (and apparently did) occur.
If the latter, I'm a bit concerned that such a breaking change would go in so close to release.
I didn't hear about this until today - that's (5 days after I merged into release and maybe 10 days since I loaded into the trunk. I notice that I can't see MPI tests on the test matrix which might be a problem. Perhaps if these were visible, I red (literally) flag would have gone up when I did the merge.
I'm not too concerned since nothing in Boost but Boost.MPI should be affected by that breaking change. One will have to make sure that users of Boost know that they'll have to update their archive sources though.
The changes are very easy. I'll add a not ot the release information. Robert Ramey
Matthias
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
On Thu, Aug 06, 2009 at 07:50:38PM -0800, Robert Ramey wrote:
If the latter, I'm a bit concerned that such a breaking change would go in so close to release.
I didn't hear about this until today - that's (5 days after I merged into release and maybe 10 days since I loaded into the trunk. I notice that I can't see MPI tests on the test matrix which might be a problem.
Yes, it is a problem: currently, a test report with MPI enabled is dropped on the floor. We have been speculating on boost-testing about the cause, but really need someone who knows the results web server to take a look. See http://lists.boost.org/boost-testing/2009/08/6355.php -Steve P.S. A secondary problem is that as far as I can tell, there is only one machine running tests with MPI enabled: mine.
Robert Ramey wrote:
Matthias Troyer wrote:
I'm not too concerned since nothing in Boost but Boost.MPI should be affected by that breaking change. One will have to make sure that users of Boost know that they'll have to update their archive sources though.
The changes are very easy. I'll add a not ot the release information.
From what I gather, this change breaks all user-defined archive types. If that's the case, then be sure to mention that when you update the release notes. Thanks. -- Eric Niebler BoostPro Computing http://www.boostpro.com
On Aug 6, 2009, at 9:50 PM, Robert Ramey wrote:
If the latter, I'm a bit concerned that such a breaking change would go in so close to release.
I didn't hear about this until today - that's (5 days after I merged into release and maybe 10 days since I loaded into the trunk. I notice that I can't see MPI tests on the test matrix which might be a problem. Perhaps if these were visible, I red (literally) flag would have gone up when I did the merge.
Actually it was reported on the trunk almost immediately after you committed the change and I fixed it on the trunk, but was not aware that you had copied the change to the release branch. Matthias
Matthias Troyer wrote:
On Aug 6, 2009, at 9:50 PM, Robert Ramey wrote:
If the latter, I'm a bit concerned that such a breaking change would go in so close to release.
I didn't hear about this until today - that's (5 days after I merged into release and maybe 10 days since I loaded into the trunk. I notice that I can't see MPI tests on the test matrix which might be a problem. Perhaps if these were visible, I red (literally) flag would have gone up when I did the merge.
Actually it was reported on the trunk almost immediately after you committed the change
Hmmm - looking at http://www.boost.org/development/tests/trunk/developer/summary.html I still don't see any test for MPI. But truth is, I normally wouldn't look at the tests for other libraries. So I wouldn't have seen it if it is there. Anyway, sorry for the inconvenience.
and I fixed it on the trunk, but was not aware that you had copied the change to the release branch.
Matthias
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
participants (6)
-
Daniel James -
Eric Niebler -
John Maddock -
Matthias Troyer -
Robert Ramey -
Steve M. Robbins