[PATCH] Add a 'no-asserts' requirement option to LIT.

[David Blaikie]

On Mon, Mar 30, 2015 at 9:37 AM, Owen Anderson <resistor at mac.com> wrote:

>
> > On Mar 29, 2015, at 4:45 PM, Chandler Carruth <chandlerc at gmail.com>
> wrote:
> >
> > I don't understand the requested change...
> >
> > We already have the ability to say 'REQUIRES: asserts' when checking
> things only enabled in asserts builds.
> >
> > It sounds like you want to check for things only enabled in *no-*asserts
> builds? I'm surprised any such things exist. My inclination is to say "we
> shouldn't have no-asserts emit data that we don't emit in asserts".
>
> Our on-line compiler is designed to be both fast and fault-tolerant.  We
> place pre-conditions on the input that we expect our client to meet in
> order to maintain acceptable compile time guarantees, and we want to assert
> on those pre-conditions in a +asserts build.  However, violations of these
> preconditions are generally recoverable, at a compile time cost, and we
> want the production on-line compiler (-asserts) never to crash on an
> end-user’s device.
>

In any case we came to the conclusion that these should be warnings/errors
that can be disabled for the embedded use case & that seemed to make
everyone happy, so we still don't need any behavior that can only be
checked in a non-asserts build. \o/ Also, this means both the positive and
negative behavior can be tested in any build of LLVM rather than parts only
in asserts and other parts only in non-asserts builds. (Chandler was just a
bit late to the party & seems to have had a bit of trouble finding the
context/tail of the thread)

- David
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150330/cf989d75/attachment.html>