* Stefano Lattarini wrote on Thu, Jun 30, 2011 at 10:45:02PM CEST: > On Thursday 30 June 2011, Ralf Wildenhues wrote: > > I'm fairly sure that's only with historic shells and not an > > issue in practice any more (Sven Mascheck's pages will have the details > > for the curious).
> Anyway, I do think that Eric's original proposal of extending the > parallel-tests > harness to allow arbitrary and portable fd redirections is the way to go > eventually; Link? > here are my reasons: > - it would be pretty easy to implement and document; > - the testsuite harness is being revamped anyway; > - that change would also benefit all the future custom and built-in test > drivers; > - the current hack with "TESTS_ENVIRONMENTS = ...; 9>&2", while continuing to > work, will prevent the use of AM_TESTS_ENVIRONMENT :-( > > > > In each case, the user's namespace is invaded, and we go against our own > > > advice (in [1] we violate the "always terminate AM_TESTS_ENVIRONMENT with > > > a semicolon" rule, in [2] we violate "don't define TESTS_ENVIRONMENT in > > > the Makefile.am, it's user-reserved" rule). > > > > > > Also, the example above belongs IMHO more in a FAQ rather than in a > > > reference manual. > > > > Are you volunteering to update the FAQ? ;-) > > > Oh no ;-) Not this summer at least. That's a pretty strong statement, and when accompanied by a documentation regression (if I may exaggerate a bit), I'm not too happy when I read this. Please do not consider documentation work as second class citicen. Actually, writing (at least a sketch of) the documentation before implementing a feature is a very good way to get more organized and better design, IMVHO. Thanks, Ralf