Re: [Boost-testing] V2 testing: static or shared

From: Martin Wille (mw8329_at_[hidden])
Date: 2005-04-15 03:57:22

Vladimir Prus wrote:
> Hello!
> 
> In Boost.Build V2, the distiction between static and shared libraries is
> less dramatic that in V1. Each 'lib' target can be built as shared library
> or static library, with shared linking being the default.
> 
> This raises two questions about regression testing with V2:
> 
> 1. Should we use static linking by default, or shared linking. Shared
> linking might decrease link times, and binary sizes, but probably not all
> libraries are prepared for shared linking (especially on Windows)

I'd prefer dynamic linking wherever possible. Only libraries which need 
static linking on some systems should be tested statically on those 
systems. However, we need some guinea pig library for testing whether 
the build system works well for static libs. I suggest we use some small 
library for that.

> 2. Some libraries have separate tests for static and shared linking, and
> some don't. Should we keep this during migration to V2, or consistently run
> all tests for all compiled libraries in both variants, or run only shared
> variants?

Consistently testing in both variants will cost too many resources. 
Boost.Serialization seems to exercise every combination of 
release/debug, shared/static, test and archive type. This causes it to 
consume 30-50% of the whole time taken for all the regression tests. If 
we add that feature for all the other libraries then a test run will 
take a week or longer here (and also use a lot of additional HD space).

Regards,
m