Re: Clarify rules for testing patches

[Hans-Peter Nilsson <hp at bitrange dot com>]

On Tue, 4 Dec 2001, Zack Weinberg wrote:
> +test with uses the code you changed.  If you have access to only a few
> +platforms, you should <a href="simtest-howto.html">use a simulator</a>
> +to increase your test coverage.</p>

This change in the simulator sentence mostly reflects opinion
AFAICS.  I suggest that the old wording be retained; recommend
simulator cross-testing without "if"s.

> -    comment changes; if there are <strong>any</strong> changes to the
> -    code itself you should re-bootstrap.)  It is recommended that you
> -    also <a href="simtest-howto.html">use a simulator</a> to increase
> -    your test coverage.</li>

I'd like to think that as much test coverage as possible is
good, not just "if one has only a few platforms".  What is "a
few" and what is "a platform"?  History has shown that people
sometimes cannot distinguish between architectural features
leading to quite different code paths (like cc0-target or not;
perhaps also 32- and 64-bit platforms).  To the extreme, one
might think that i?86-cygwin is a different platform than
i?86-linux-gnu and that two platforms are not too few.  On the
other hand, one could reasonably think that bootstrapping and
testing systems with sparc, powerpc, hppa, mips and i?86 means
everything is tested, front-end and back-endwise.  See the
recent cross-build breakage (still present on the release branch
AFAIK), which isn't reproducible when just bootstrapping.
Also, none of those ports is a cc0-port, so that wouldn't be
tested.  Both situations would be caught by testing mn10300-elf
according to the simtest instructions, or in general, testing
the set documented as working in simtest-howto.html.  BTW,
arm-elf and mips-elf are broken since weeks (maintainers
pinged); the rest works IIRC.

brgds, H-P