Re: [boost] Handling testcases that are known to fail?

From: Jeremy Siek (jsiek_at_[hidden])
Date: 2002-01-23 23:45:56

You're making the assumption that having errors show up in the regression
tests will force a maintainer to make library X work on some compiler.
This assumption is incorrect. Boost is a volunteer organization and can't
force anyone to do anything. For example, I haven't ported the BGL to
Borland, and don't plan to unless 1) Borland fixes their compiler or 2)
someone pays me to do the port. So having an expected failure list is
merely informative to users, it lets them know the intentions of a
maintainer. For example, I am committed to supporting BGL on KCC, VC++,
g++, and a few others. When I receive a list of errors, it would be nice
to just have the unexpected errors from these compilers, uncluttered by
the expected ones from Borland.

Cheers,
Jeremy

P.S. I don't have a particular grudge against Borland. For example, the
iterator_adaptor library works on Borland. It is just Borland's a hard
problem for the BGL.

On Wed, 23 Jan 2002, Douglas Gregor wrote:
gregod>
gregod> I guess my fear of this approach is because I can foresee
gregod> this as the easy way out. It's tough to get library X
gregod> working on this compiler, so I'll just expect everything
gregod> to fail on this compiler and ignore it. While this seems
gregod> fair to the developer (in an ideal world, we shouldn't
gregod> have to deal with these broken compilers), it's horrible
gregod> for our users.
gregod>
gregod> 	Doug

----------------------------------------------------------------------
 Jeremy Siek                          http://php.indiana.edu/~jsiek/
 Ph.D. Student, Indiana Univ. B'ton   email: jsiek_at_[hidden]
 C++ Booster (http://www.boost.org)   office phone: [phone]
----------------------------------------------------------------------