Re: [boost] Boost regression notification (2005-06-30)

From: David Abrahams (dave_at_[hidden])
Date: 2005-07-05 09:21:51

Doug Gregor <dgregor_at_[hidden]> writes:

> On Jul 5, 2005, at 7:59 AM, David Abrahams wrote:
>
>> Douglas Gregor <doug.gregor_at_[hidden]> writes:
>>
>>> On Jul 3, 2005, at 5:31 AM, Paul Baxter wrote:
>>>> In terms of compiler status, if there really is an issue between 
>>>> debug
>>>> and
>>>> release (usually bad compiler optimisations) shouldn't we:
>>>> a) clearly note in the test report we're in debug or release build
>>>> (can also
>>>> be discovered by looking at the build directory name)
>>>>
>>>> b) Actually test a compiler using a reasonable set of optimisations
>>>> since
>>>> we're trying to report if a compiler can use Boost in actual builds,
>>>> not in
>>>> a no optimisation alternative.
>>>
>>> I hadn't realized you were running a release built. It is very
>>> important to do (for many reasons), but I think we should add
>>> "-release" to the end of the toolset name to indicate what we're 
>>> doing.
>>> Then we can more immediately see what problems are caused by compiler
>>> optimizations.
>>
>> It's also important to check through all of our tests to make sure
>> they're not using "assert" and/or to force NDEBUG to be undefined.
>> Otherwise many of our tests could be passing because they have no code
>> to execute.
>
> Well, we *should* be using BOOST_CHECK, because we want assert()s in 
> library code to disappear for release builds to be sure we haven't put 
> an important side effect in them.

Maybe we *should* but there are plenty of tests lying around that
predate it.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com