From: Giovanni Bajo (giovannibajo_at_[hidden])
Date: 2002-07-06 20:10:32
----- Original Message -----
From: "Greg Comeau" <comeau_at_[hidden]>
Sent: Saturday, July 06, 2002 7:21 PM
Subject: [boost] Comeau feedback
> Some feedback: I skimmed this newsgroup going back a few months, and
> saw a few Comeau C++ discussions I wasn't clear about, and don't know
> if they are still of significance. On the chance that they are, if I may:
> * --long_long ... looked like something may have been set up wrong.
> If this is still an issue, we can look at it anew so I can understand
> it better.
> * As of 4.3.0, __NO_LONG_LONG is set if --long_long is not requested
It's almost ok. Latest comeau.hpp still misses the following lines:
# if !defined(__EXCEPTIONS)
# define BOOST_NO_EXCEPTIONS
# if !defined(__NO_LONG_LONG)
# define BOOST_HAS_LONG_LONG
Maybe John Maddock can officially check in this little change.
> * __int64: I wasn't sure what the issue was here
It was related to the fact that Comeau in strict mode does not support
__int64, so we needed a way to correctly detect its use. It's now fixed,
with an explicit check against _MSC_VER.
> * Some discussion occurred about windows.h, but I'm not sure why that
> has any significance to Boost (after all, Boost runs on non-Windows
Because there are some libraries that need access to platform specific APIs.
For example, to access the files or the timer. The problem is still unsolved
here: Comeau in strict mode can't be seen as a fully-compliant Windows
platform (because e.g. windows.h doesn't compile in strict mode). On the
other hand, under Windows we don't have support for POSIX or similar. So
currently, for instance, there is no way Comeau (strict mode) can pass the
Boost.Regex regression test.
> * Most all versions of 4.3.0 beta were setting _MSC_VER to 1, in all
> Giovanni pointed this out, since previously this was only done in
> strict_mode. This was added as an interim crutch while adding the other
> new backends that are now supported, and setting _MSC_VER to 1 in
> non-strict mode has now been removed, after verifying that all out tests
> are able to run w/o it.
This is very important, since it's the only way Boost has to detect strict
mode. Thank you for changing it back to the previous semantic.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk