10-215
To: J3
From: Nick Maclaren
Subject: Interop TR: _Bool (5.2.5 Macros)
Date: 2010 October 06
Reference: 10-165r2.pdf


It is a serious mistake for the TR to either use or provide support
for C99's 'boolean' type, for a great many reasons.  The following
is a very condensed summary of the situation.

C99 has not been universally accepted by the IT community, and a great
many important interfaces and applications explicitly require C90.
Also, an equally large number use C++ and will want to use this
interface via the defined C subset -- the standards are such that this
can be argued to be defined behaviour.  It therefore makes sense not
to rely on or support anything that is not at least compatible between
C90, C99 and C++.

C99's '_Bool' and the (redefinable) macros 'bool', 'true' and 'false'
from <stdbool.h> were introduced against strong opposition within WG14
as a hack to emulate C++'s 'bool' type, and they are largely but not
entirely compatible - but ONLY at the source level.  The question for
the TR is whether they are compatible at the data representation level,
and that is not required to be the case.

C++0X (WG21 document n3092) makes no reference to _Bool (and therefore
does not include it in the language), and says the following about
header <cstdbool>:

    __bool_true_false_are_defined is defined. [table 26, p. 460]

    The header <cstdbool> and the header <stdbool.h> shall not define
    macros named bool, true and false.  [18.10 support.runtime, p. 460]

It is noteworthy that C99 makes no use of _Bool in any of the new
functions that it has introduced into the library, and continues to use
int for that purpose.  Fortran should not tread new ground in C
specifications!

Also, _Bool and C++ bool are fundamentally non-interoperable with
Fortran, because they have delights like being required to support
arithmetic operations but with non-arithmetic semantics.

There are more reasons not to touch _Bool, but that should be enough.

There is also an error in 10-165r2.pdf, because the type is _Bool and
not Bool.


Edits to 10-165r2:

11:0+  In the entry for CFI_type_Bool in table 5.2, the C type "Bool"
should be corrected to "_Bool".

12:14  "_Bool" should be replaced by "int".

12:15-16  "true" should be replaced by "1" and "false" by "0".