Boost logo

Boost Testing :

From: David Abrahams (dave_at_[hidden])
Date: 2005-10-26 07:02:24

"Robert Ramey" <ramey_at_[hidden]> writes:

> David Abrahams wrote:
>> "Alisdair Meredith" <alisdair.meredith_at_[hidden]> writes:
>>> Robert Ramey wrote:
>>>> I've been running tests on my local machine in release mode. It
>>>> turns out that all my tests fail with the message
>>>> Run output:
>>>> EXIT STATUS: -1073741819
>>>> This occurs with both borland 5.51 and 5.64 compilers.
>>>> I think i recall having this before and concluding
>>>> that it was an issue with Boost Test
>>>> Perhaps someone might have a look at this.
>>> Borland has serious problems with running destructors in release
>>> mode. With a new compiler finally due out by the end of the year, we
>>> hope to see some serious bug fixes, but until then I would recommend
>>> ALWAYS disabling inline function expansion, even in release mode.
>>> This has been standard advice on the Borland newsgroups for several
>>> years now.
>> I guess I'm going to do that in all our standard toolsets now. Please
>> squawk if there are any objections!
> all standard toolset or all borland standard toolsets?

Borland only, of course.

> In anycase, I'm not sure this is a good idea.

It's already been done, over a week ago.

> The whole idea of testing
> with release mode would be to flush out bugs before users do - and I don't
> think users would necessarily do this.

So, are you prepared to find workarounds for the compiler's codegen

> My original concern was that relase mode couldn't be tested without
> tracking down something hard to find in the test library. So I'm
> thinking that maybe we can just apply the "fix" to just the test
> library build and see what happens with the rest of the tests.

Okay, I guess that's a more targeted approach.

> Applying it to everything may hide the very things we are looking
> for. If it turns out that release mode for borland is totally
> unusable, we can then apply the fix globally.

I'll roll back the changes to the toolsets. As far as I'm concerned,
you can add <inlining>off to the requirements of the library in
question. I'd check with Gennadiy first, though.

Dave Abrahams
Boost Consulting

Boost-testing list run by mbergal at