Boost logo

Boost :

From: williamkempf_at_[hidden]
Date: 2001-09-08 10:55:58

--- In boost_at_y..., John Maddock <John_Maddock_at_c...> wrote:
> Bill,
> Some more comments I missed out in my last post:
> Building:
> ~~~~~
> It would be nice if the docs contained a breif section on building
> library - in particular mentioning that threadmon.cpp is for win32
> and must live inside a dll.

I expect things to be built using the Boost.Build system, so most of
the above is irrelevant since it's specified in the Jam files.
Thoughts on this?

> Also the use of DllMain inside threadmon.cpp is MS specific:
> compiler requires DllEntryPoint as the entry point function name, I
> #ifdef __BORLANDC__
> #define DllMain DllEntryPoint
> #endif
> to the top of the file to fix that.

I was unaware of this. I'll add this fix. Thanks.
> It looks to me as though the current code cannot be exported from a
dll on
> Win32, personally I would very much prefer it if it could be - you
> need to define and add the usual BOOST_THREAD_DECL macros to all
the class
> declarations (unless am I missing something?).

I went down this route and VC++ couldn't do it (errors with linkage
due to template members). I don't consider the use of this library
as a DLL to be pressing. We can work this out later if it's truly

> Test programs:
> ~~~~~~~~~~
> Your test program exposes weaknesses in the current regression test
> (especially on Win32 as part has to built as a separate dll), I
don't know
> what we need to do about that.

I'm not familiar enough with the system to comment. Testing via the
Boost.Build system works, though I've asked David Abrahams for some
changes to that as well. I believe there's talk of interfacing the
regression system with the build system?

> Is the test program sufficient? Particularly for condition
> (which never get signalled as far as I can see)?

Conditions are signalled in both the test_condition_notify_one() and
test_condition_notify_all() tests, or am I missing something in what
your saying here?

I can't be 100% sure that the test is all conclusive. During
development I did find I had to add a test or two that was missed.
However, I covered as much as I could. Testing threads isn't an easy
task and I expect we'll add further tests as we go.

Bill Kempf

Boost list run by bdawes at, gregod at, cpdaniel at, john at