Re: scanf non-conformance

["Joseph S. Myers" <jsm28 at cam dot ac dot uk>]

On 15 Aug 2000, Ulrich Drepper wrote:

> If you come up with some words I"ll certainly add them.

Here's a summary of the issues I know of (with the web archives pointed to
for my lists of issues with the headers rather than including those again
below).  It's a bit long but hopefully thorough; the header issues list
could be included as well if you want (and then updated as and when issues
are fixed in glibc).

-- 
Joseph S. Myers
jsm28@cam.ac.uk

Technical C standards conformance issues in glibc
=================================================

If you compile programs against glibc with __STRICT_ANSI__ defined
(as, for example, by gcc -ansi, gcc -std=c89, gcc -std=iso1990:199409
or gcc -std=c99), and use only the headers specified by the version of
the C standard chosen, glibc will attempt to conform to that version
of the C standard (as indicated by __STDC_VERSION__):

GCC options		Standard version
-ansi			ISO/IEC 9899:1990
-std=c89		ISO/IEC 9899:1990
-std=iso9899:199409	ISO/IEC 9899:1990 as amended by Amd.1:1995
-std=c99		ISO/IEC 9899:1999

(Note that -std=c99 is not available in GCC 2.95.2, and that no
version of GCC presently existing implements the full C99 standard.)

You may then define additional feature test macros to enable the
features from other standards, and use the headers defined in those
standards (for example, defining _POSIX_C_SOURCE to be 199506L to
enable features from ISO/IEC 9945-1:1996).

There are some technical ways in which glibc is known not to conform
to the supported versions of the C standard, as detailed below.  Some
of these relate to defects in the standard that are expected to be
fixed, or to compiler limitations.

Defects in the C99 standard
===========================

The definition of macros such as INT8_C in <stdint.h> and <inttypes.h>
is not implementable (Defect Report #209); this is expected to be
fixed in a Technical Corrigendum to make the macros yield a constant
expression of the promoted type (for example, int rather than char)
rather than needing to be able to represent constants of type char.
glibc follows this corrected version.

Several of the <fenv.h> functions are specified to return void, but
Defect Report #202 points out that under some circumstances they may
need to return an error status.  They are expected to be corrected to
return int; glibc follows this corrected specification.

Implementation of library functions
===================================

The implementation of some library functions does not fully follow the
standard specification:

C99 added additional forms of floating point constants (hexadecimal
constants, NaNs and infinities) to be recognised by strtod() and
scanf().  The effect is to change the behavior of some strictly
conforming C90 programs; glibc implements the C99 versions only
irrespective of the standard version selected.

C99 added %a as another scanf format specifier for floating point
values.  This conflicts with the glibc extension where %as, %a[ and
%aS mean to allocate the string for the data read.  A strictly
conforming C99 program using %as, %a[ or %aS in a scanf format string
will misbehave under glibc.

Compiler limitations
====================

The macros __STDC_IEC_559__, __STDC_IEC_559_COMPLEX__ and
__STDC_ISO_10646__ are properly supposed to be defined by the
compiler, and to be constant throughout the translation unit (before
and after any library headers are included).  However, they mainly
relate to library features, and the necessary magic has yet to be
implemented for GCC to predefine them to the correct values for the
library in use, so glibc defines them in <features.h>.  Programs that
test them before including any standard headers may misbehave.

GCC doesn't support the optional imaginary types.  Nor does it
understand the keyword _Complex.  This has the corresponding impact on
the relevant headers.

glibc's use of extern inline conflicts with C99: in C99, extern inline
means that an external definition is generated as well as possibly an
inline definition, but in GCC it means that no external definition is
generated.  When GCC's C99 mode implements C99 inline semantics, this
will break the uses of extern inline in glibc's headers.  (Actually,
glibc uses `extern __inline', which is beyond the scope of the
standard, but it would clearly be very confusing for `__inline' and
plain `inline' to have different meanings in C99 mode.)

glibc's <tgmath.h> implementation is arcane but thought to work
correctly; a clean and comprehensible version requires compiler
builtins.

For most of the headers required of freestanding implementations,
glibc relies on GCC to provide correct versions.  (At present, glibc
provides <stdint.h>, and GCC doesn't.)  GCC's <float.h> is missing
FLT_EVAL_METHOD and DECIMAL_DIG; glibc instead provides them in
<math.h>, which is not what the standard specifies.  GCC's <stdbool.h>
is broken: GCC lacks support for the _Bool type.

Implementing MATH_ERRNO, MATH_ERREXCEPT and math_errhandling in
<math.h> needs compiler support: see

http://sources.redhat.com/ml/libc-hacker/2000-06/msg00008.html
http://sources.redhat.com/ml/libc-hacker/2000-06/msg00014.html
http://sources.redhat.com/ml/libc-hacker/2000-06/msg00015.html

Issues with headers
===================

There are various technical issues with the definitions contained in
glibc's headers.  See

http://sources.redhat.com/ml/libc-alpha/2000-07/msg00259.html
http://sources.redhat.com/ml/libc-alpha/2000-07/msg00279.html