From: Aleksey Gurtovoy (alexy_at_[hidden])
Date: 2003-05-06 17:39:21
David Abrahams wrote:
> Aleksey Gurtovoy <agurtovoy_at_[hidden]> writes:
> > I am afraid the original question is still open, though - besides
> > fixing the issue itself, can we make sure that when something like
> > this happens again in the future, bjam doesn't just quit the whole
> > regression process? It's only one library that is broken, and
> > terminating the whole thing on an error like "unknown dependent
> > target" just doesn't seem right to me (I would understand if it
> > crashed or something, but deliberatively quitting definitely seems
> > wrong :).
> Sure. You can grep for the error message in tools/build/*.jam and
> ECHO warning:
Won't that break the standalone builds? I mean, EXIT is probably quite
appropriate when you are just building the library itself; it's just not
appropriate in the context of the batch regression testing.
Also, what about a half-hundred of other EXIT's in tools/build/*.jam files?
Some of them definitely seem to fall into the same category as the one we
happened to bump into. We are definitely not in favor of the whole thing
breaking again when a month later somebody checks in a target with
an incompatible property or something like it.
How about if instead of EXIT-ing or ECHO-ing we do something like
report_error unknown dependent target $(main-target) ;
and have 'report_error' decide whether it's appropriate to exit or not,
depending on the context, e.g. by defining it differently for the regression
Does it make sense?
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk