Re: [boost] Regression tests on IRIX and Compaq Tru64 Unix

From: Beman Dawes (beman_at_[hidden])
Date: 2000-12-12 20:51:34

At 03:56 PM 12/12/2000 -0700, Mark D Rintoul wrote:

 >Jens Maurer wrote:
 >>
 >> Thanks to Ralf W. Grosse-Kunstleve, I have been able to run the
 >> boost regression tests on IRIX and Compaq Tru64 Unix.  I've
 >> added the results to the boost compiler status page.
 >>
 >> The main problem on both platforms is the lack of new-style
 >> <cXXX> headers for the C standard library, which make the overall
 >> results look slightly disappointing (only two tests pass).
 >>
 >> Should we try to hack around missing <cXXX> headers?
 >> (I don't think so.)
 >>
 >> Jens Maurer
 >
 >	Hmmm...okay, I'll bite.  Why not?  I agree that the source
 >is already #defined to death, but there is really not a huge number
 >of places to change this.  I went through and counted them the other
 >day and I think there was less than 20.  A simple BOOST_STDC_HEADERS
 >define and we long suffering IRIX users are on our way.  As it is
 >right now, I download the new version of boost and hack on the files
 >I'm currently using to work on my SGI.

I'd like to make sure I understand the issue.  Please correct me if I'm 
wrong.

The implementation doesn't contain <cwhatever> headers.  Boost uses 
<cwhatever> headers.

The problem can be hacked around by #include <whatever.h> and then doing 
the usual namespace std { using ::somename; ... } trick.

There are two places to apply this hack:

   1) Do it in the boost headers at the point of the original #include, 
controlled by #ifdef's.  Downside: messes up the boost headers.  Has to be 
applied in every new boost header, as they are added.  Although because of 
the Microsoft almost-C++ compiler, the namespace part of the trick has to 
be done anyhow.

   2) Provide a set of dummy <cwhatever> headers, containing the 
hack.  Users would presumably include these headers in their search path 
only if their standard library implementation didn't supply the <cwhatever> 
headers.  Downside: it might appear boost was blessing these broken 
implementations.

Is this a correct analysis?

--Beman

 >
 >	How about a different argument?  It should be noted that
 >the boost code otherwise works *great* on IRIX.  We SGI owners haven't
 >complained a bit yet about how things need to be changed to support us.
 >Can't you throw us a bone?  :)
 >
 >			Danny
 >--
 >------------------
 >Mark D. (Danny) Rintoul, Dept. 9235
 >Computational Biology & Materials Technology
 >Sandia National Laboratories
 >Phone: [phone], Fax: [phone]
 >email: rintoul_at_[hidden]
 >------------------