Re: Contents of <confdefs.h> for C++

[Walter E. Brown]

Thanks, Paul, for the courteous and quick response.

Suppose my program defines entities whose names clash with names
declared in <stdlib.h>.  While it is certainly arguable whether this
constitutes good practice, it seems equally certain that such programs
comply with the relevant language standard(s).

(As an example, suppose I want my own rand() function.  Although
valid, such a program will fail to compile in C when a header
introduces a name clash.)

This situation is potentially much worse in C++, because C++ permits
overloaded functions.  This means that I can write a
correctly-behaving, fully-compliant program whose behavior may
nonetheless be different when <stdlib.h> is part of the program.

(Permit me to generalize for a moment:  Suppose I have some function
f(), taking a double parameter.  If my program invokes f(3), my
function is called because 3 will be converted to 3.0.  However, if an
included header defines a function f() that takes an int parameter,
then f(3) will call the header's f(int), not my f(double)!  Thus, my
program will still compile, yet exhibit potentially different
behavior.)

Because of such issues, it is important in my context that I have full
control over which headers are included.  I believe this is not
unreasonable.

However, there may be compelling reasons for autoconf's current
behavior; I will reserve judgement until I learn more about its
underlying rationale in this regard.  I had begun to look at the
source code, as suggested, and do understand the mechanics of the
macro you mention, but not yet the reasoning behind them; there seems
to be special concern about the exit() function, but that's all I
really know at the moment.

Thanks again,

    - WEB


----- Original Message -----
From: "Paul Eggert" <address@hidden>
To: <address@hidden>
Cc: <address@hidden>
Sent: Friday, July 05, 2002 12:21 AM
Subject: Re: Contents of <confdefs.h> for C++


> > From: "Walter E. Brown" <address@hidden>
> > Date: Thu, 04 Jul 2002 16:01:53 -0500
> >
> > This unwanted and seemingly undocumented "feature" is deadly to
many
> > of my C++ test programs, as it forcibly introduces extra symbols.
It
> > messes up my testing in other ways, too.
>
> Details, please?  Why can't you write your test programs so that
they
> are robust in the presence of "#include <stdlib.h>"?
>
> > At a minimum, can someone kindly provide the rationale behind this
> > "feature" so that, perhaps, other possibilities can be explored?
>
> You might look at the source code for autoconf, e.g. at
> _AC_PROG_CXX_EXIT_DECLARATION.  (I haven't a clue as to why
<stdlib.h>
> is being included, but then again I avoid C++ so I have an excuse
for
> being clueless.  :-)