Re:Re: [boost] Regression Tests, Boost.Serialization
Jeff Garland wrote:
Houston, we have a problem.
I had a few GB of disk space left a few days ago. Now, I'm unable to run all the tests. I could drop one of the gcc-3.4 versions. However, I'd still be short by ~10GB after that.
It's a fact - running the bjam serialization test is going to consume 20 GB. If the test machine doesn't have it the best is just to remove the serialization test from the status jamfile for that machine. Actually I think it would be unusual for a machine used for testing boost not to have this much free space available. So I would hope it only affects martin's tests.
It would be nice if we could drop serialization on compilers that just aren't going to work.
Having looked at the test results on all the compilers the only ones that I really wonder about are HP, and for now Comeau. All tests run with MSVC 6.0 and borland will fail without spirit 1.61 being installed. This is because spirit 1.61 is needed to build the libraries. So if the library build fails then all the tests are skipped. They are marked "fail" which is a little misleading - "skipped" might be better.
This is where we were hoping to be able to allow regression levels like 'basic' and 'torture' that would provide standard ways of controlling the number of tests. But none of this is currently available.
All in all I think the testing program is working well and in reasonable time. I need this to shake out issues with each platform. It just takes the resources that it takes. So far I've only been through two test cycles so it's premature to panic.
As for what we should do now, it would be nice to have an answer to the question I posed this morning: is serialization going into 1.32 or are we cutting it out in the final release:
I want to keep working to finish regardless of which release it becomes a part of.
If we are cutting it out the we can just remove it from status bjam now....
But that would mean that progress would be stopped on getting this done. Robert Ramey
Robert Ramey wrote:
Jeff Garland wrote:
Houston, we have a problem.
I had a few GB of disk space left a few days ago. Now, I'm unable to run all the tests. I could drop one of the gcc-3.4 versions. However, I'd still be short by ~10GB after that.
It's a fact - running the bjam serialization test is going to consume 20 GB.
The disk requirements can be changed. Stripping the test binaries and removing the .o files already has been mentioned. Another chance lies in merging similar (in terms of instantiated templates) tests into one binary.
If the test machine doesn't have it the best is just to remove the serialization test from the status jamfile for that machine.
That'd be quite unfortunate an solution. Actually I
think it would be unusual for a machine used for testing boost not to have this much free space available. So I would hope it only affects martin's tests.
Well, I don't think I have to expect the disk requirements to be tripled just because one library is being added to Boost and to save disk space just for that case. Not having some large company-owned box around for testing, I need to plan the resources. I wouldn't be surprised if other volunteers were in a similar situtation.
It would be nice if we could drop serialization on compilers that just aren't going to work.
Having looked at the test results on all the compilers the only ones that I really wonder about are HP, and for now Comeau. All tests run with MSVC 6.0 and borland will fail without spirit 1.61 being installed. This is because spirit 1.61 is needed to build the libraries. So if the library build fails then all the tests are skipped. They are marked "fail" which is a little misleading - "skipped" might be better.
Ah, good! This explains why gcc-2.95 doesn't need that much space for the tests :) [...]
As for what we should do now, it would be nice to have an answer to the question I posed this morning: is serialization going into 1.32 or are we cutting it out in the final release:
I want to keep working to finish regardless of which release it becomes a part of.
If we are cutting it out the we can just remove it from status bjam now....
But that would mean that progress would be stopped on getting this done.
Right. I do not ask to remove Boost.Serialization from the tests. That'd be the sadest thing we could do. Regards, m
On Fri, 16 Jul 2004 09:14:25 +0200, Martin Wille wrote
Robert Ramey wrote:
I want to keep working to finish regardless of which release it becomes a part of.
I understand, but....
If we are cutting it out the we can just remove it from status bjam now....
But that would mean that progress would be stopped on getting this done.
Right. I do not ask to remove Boost.Serialization from the tests. That'd be the sadest thing we could do.
....I believe the release needs to take priority over serialization at this point. It's purely a timing issue. Because of the diskspace problem, the full Linux regression tests aren't available so other library authors are unable to clean up the mainline. So while I would love to see Robert clean things up and get serialization in the release it seems unlikely this is going to be feasible unless someone comes up with a way to allow the Linux regression tests to continue (maybe someone wants to fork up $150 bucks and Fedex Martin a new disk drive ;-) In the meantime, Robert can look at my single compiler Mandrake Linux regression test and work off gcc3.3 serialization regression issues (as of yesterday everything was failing with what appears to be a Jamfile issue -- can't find library stdc++). I'm going to kick off a new version in a few minutes. Jeff
"Jeff Garland" <jeff@crystalclearsoftware.com> writes:
On Fri, 16 Jul 2004 09:14:25 +0200, Martin Wille wrote
Robert Ramey wrote:
I want to keep working to finish regardless of which release it becomes a part of.
I understand, but....
If we are cutting it out the we can just remove it from status bjam now....
But that would mean that progress would be stopped on getting this done.
Right. I do not ask to remove Boost.Serialization from the tests. That'd be the sadest thing we could do.
....I believe the release needs to take priority over serialization at this point. It's purely a timing issue. Because of the diskspace problem, the full Linux regression tests aren't available so other library authors are unable to clean up the mainline. So while I would love to see Robert clean things up and get serialization in the release it seems unlikely this is going to be feasible unless someone comes up with a way to allow the Linux regression tests to continue (maybe someone wants to fork up $150 bucks and Fedex Martin a new disk drive ;-)
I have a Linux machine with 16311080 of available disk where we can run the tests if someone will help me get the tests set up. -- Dave Abrahams Boost Consulting http://www.boost-consulting.com
David Abrahams writes:
"Jeff Garland" <jeff@crystalclearsoftware.com> writes:
On Fri, 16 Jul 2004 09:14:25 +0200, Martin Wille wrote
Robert Ramey wrote:
I want to keep working to finish regardless of which release it becomes a part of.
I understand, but....
If we are cutting it out the we can just remove it from status bjam now....
But that would mean that progress would be stopped on getting this done.
Right. I do not ask to remove Boost.Serialization from the tests. That'd be the sadest thing we could do.
....I believe the release needs to take priority over serialization at this point. It's purely a timing issue. Because of the diskspace problem, the full Linux regression tests aren't available so other library authors are unable to clean up the mainline. So while I would love to see Robert clean things up and get serialization in the release it seems unlikely this is going to be feasible unless someone comes up with a way to allow the Linux regression tests to continue (maybe someone wants to fork up $150 bucks and Fedex Martin a new disk drive ;-)
I have a Linux machine with 16311080 of available disk where we can run the tests if someone will help me get the tests set up.
It's piece of cake now -- http://tinyurl.com/4f2zp. -- Aleksey Gurtovoy MetaCommunications Engineering
On Fri, Jul 16, 2004 at 06:21:38AM -0700, Jeff Garland wrote:
On Fri, 16 Jul 2004 09:14:25 +0200, Martin Wille wrote
Right. I do not ask to remove Boost.Serialization from the tests. That'd be the sadest thing we could do.
....I believe the release needs to take priority over serialization at this point. It's purely a timing issue. Because of the diskspace problem, the full Linux regression tests aren't available so other library authors are unable to clean up the mainline. So while I would love to see Robert clean things up and get serialization in the release it seems unlikely this is going to be feasible unless someone comes up with a way to allow the Linux regression tests to continue (maybe someone wants to fork up $150 bucks and Fedex Martin a new disk drive ;-)
I am already using the serialization library whence I'd really like to see it in the upcoming release. If it helps to make sure it is included than I can offer to run the test suite on a linux box with gcc 3.4.1, at least until the release is out and the problems with the excessive disk space usage is dealt with. This way Martin Wille could exclude this particular platform from his tests and regain some of his disk space. Please contact me if there is any interest. Regards Christoph -- http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/cludwig.html LiDIA: http://www.informatik.tu-darmstadt.de/TI/LiDIA/Welcome.html
On Fri, 16 Jul 2004 09:14:25 +0200, Martin Wille wrote
Robert Ramey wrote:
I want to keep working to finish regardless of which release it becomes a part of.
I understand, but....
If we are cutting it out the we can just remove it from status bjam now....
But that would mean that progress would be stopped on getting this done.
Right. I do not ask to remove Boost.Serialization from the tests. That'd be the sadest thing we could do.
....I believe the release needs to take priority over serialization at
"Jeff Garland" <jeff@crystalclearsoftware.com> wrote in message news:20040716132138.M34852@crystalclearsoftware.com... this
point. It's purely a timing issue. Because of the diskspace problem, the full Linux regression tests aren't available so other library authors are unable to clean up the mainline. So while I would love to see Robert clean things up and get serialization in the release it seems unlikely this is going to be feasible unless someone comes up with a way to allow the Linux regression tests to continue (maybe someone wants to fork up $150 bucks and Fedex Martin a new disk drive ;-)
If it'd help get serialization into 1.32 I'd be glad to. What are the requirements? I assume Martin's in Austria, any suggestions on suppliers that will drop ship to Austria? Jeff Flinn
Jeff Flinn wrote:
"Jeff Garland" <jeff@crystalclearsoftware.com> wrote in message news:20040716132138.M34852@crystalclearsoftware.com...
...
Fedex Martin a new disk drive ;-)
If it'd help get serialization into 1.32 I'd be glad to. What are the requirements? I assume Martin's in Austria, any suggestions on suppliers that will drop ship to Austria?
That's a most generous offer! Thank you! However, I don't think this is necessary anymore. I found enough stuff to delete and removed the testing for gcc-3.4.0 (keeping 3.4.1) and it looks like I'm having enough free space now. If in the long run the build system is changed in order to shrink or to delete unneeded files then the space requirements for testing should shrink dramatically. Best regards, m PS: I'm in Munich, Germany, actually.
On Fri, 16 Jul 2004 06:21:38 -0700, Jeff Garland wrote
On Fri, 16 Jul 2004 09:14:25 +0200, Martin Wille wrote
Robert Ramey wrote:
I want to keep working to finish regardless of which release it becomes a part of.
BTW, I had to remove the line <runtime-link>static from the serialization/test/Jamfile to get the serializiation regression tests to link on my Linux machine. With g++ this was adding a -static which was apparently making it not find libstd++. Of course after I did this, I too ran out of diskspace (I have plenty I just need to clean up) so I'm it will be awhile before I repost. Jeff
participants (12)
-
Aleksey Gurtovoy -
Christoph Ludwig -
David Abrahams -
Jeff Flinn -
Jeff Garland -
Martin Wille -
Robert Ramey