[Boost-build] V2/1.34: next steps

From: Vladimir Prus (ghost_at_[hidden])
Date: 2006-02-10 09:26:03

Hello everobody!

Some time ago we've talked what's necessary to switch C++ Boost to V2. I've 
posted an email with proposed steps (attached for convenience), and here's 
what we have now.

I'm almost done with the step 1 (making V1 and V2 produce the same results on 
gcc/Linux). In fact, I have just one test yes to handle, compared to a huge 
number two weeks ago. 

So, we can move to steps 2 and 3.

Step 2 is making sure I'm not the only one with "lucky" configuration and that 
gcc/Linux is indeed healthy. The instructions for running V1 and V2 
regressions and comparing them are at:

   http://zigzag.cs.msu.su/boost.build/wiki/TestingOnBoost

Can somebody volunteer to do such testing? 
If compilation time or space is a concern, then running just V2 regressions 
and sending the results for me will be sufficient -- I'll compare those with 
my local V1 results.

Step 3 is making sure that msvc/Windows is healthy as well. The instructions 
are again at:

   http://zigzag.cs.msu.su/boost.build/wiki/TestingOnBoost

The tests can be run by anybody with msvc/Windows, but issues in Boost.Build 
are likely to appear. Dave, any chance you can take this part?

The next step on horizon is trying running with regression.py. I have 
necessary modification already made, and we can proceed to this as soon as 
steps (2) and (3) are done.

- Volodya


attached mail follows:

David Abrahams wrote:

> 
> Happy New Year, everyone!

Happy New Year!

> 
> It's a brand new year; Boost should switch to a "brand new" build
> system, ASAP.  How close are we to being able to make that switch?

Good question. The absolutely fair answer is that we don't know until we
try. I can get the ball rolling, but we really need to decide on the steps
to be made, and work on them together to push this to completion. Here are
the steps that I'd propose:

1. I set up a script that will run all Boost regressions with V1 and V2 and
compare results, and adjust V2 Jamfiles to get zero differences on
Linux/gcc.

2. Somebody else with Linux/gcc run the same script to verify the results
are the same for him too.

3. Somebody with Windows/msvc compares V1 and V2 results too, coordinating
with me to get them equal.

4. We adjust the regression.py script to have "--v2" switch and ask some of
the current regression runners to test it.

5. We post to Boost mailing list, explaining that we're technically ready to
move to V2, and if anybody has any arguments against.

6. Regression runners switch to V2 and "bjam" in Boost starts to use V2 by
default, with V1 invoked by "--v1" option.

7. Any issues are ironed out.

There are two issues that are likely to arise along the way:
1. I still haven't tried to port serialization tests to V2, and they are
pretty contrived.

2. The auto-configuration of toolsets works with msvc only, so we need to
add auto-configuration to gcc, at least.

- Volodya

 

_______________________________________________
Boost-build mailing list
Boost-build_at_[hidden]
http://lists.boost.org/mailman/listinfo.cgi/boost-build