Is an ANSI C-compliant implementation allowed to include additional types and functions in its standard library, beyond those enumerated by the standard? (An ideal answer would reference the relevant part of the ANSI standard.)
I ask particularly because Mac OS 10.7 declares the getline
function in stdio.h, even when compiling with gcc or clang using the -ansi
flag. This breaks several older programs that define their own getline
function. Is this a fault of Mac OS 10.7? (The man page for getline
on Mac OS 10.7 says that getline
conforms to the POSIX.1 standard, which came in 2008.)
Edit: To clarify, I find it odd that including stdio.h in an ANSI C89 program on Mac OS 10.7 also pulls in the declaration for the getline
function, since getline
is not one of the functions enumerated in the K&R (and presumably ANSI) description of stdio.h. In particular, attempting to compile noweb:
gcc -ansi -pedantic -c -o notangle.o notangle.c
In file included from notangle.nw:28:
getline.h:4: error: conflicting types for ‘getline’
/usr/include/stdio.h:449: error: previous declaration of ‘getline’ was here
Is it a bug in Mac OS 10.7 includes the declaration of getline
in stdio.h even when compiling for the ANSI C89 standard?
Yes, a compliant implementation is allowed to define additional identifiers, including functions, as long as they are one of the reserved identifiers in the standard. For example:
All identifiers that begin with an underscore and either an uppercase letter or another underscore are always reserved for any use;
All identifiers that begin with an underscore are always reserved for use as identifiers with file scope in both the ordinary and tag name spaces;
All external names that begin with
is
,to
,str
,mem
orwcs
followed by a lowercase letter;In addition there are names that are reserved only if you include certain headers; for example, if you include
<errno.h>
then it can define any macro starting withE
followed by a digit or uppercase letter.However,
getline()
is not such a reserved name, and a compliant implementation must make it available for the programmer's own use.From section 7.1.3 paragraph 2 of n1570 (which is a draft of C1x):
This is the part that means
getline
shouldn't be defined by the<stdio.h>
, since it's not a reserved identifier according to the spec. So if your library definesgetline
in<stdio.h>
, it's not technically compliant with the C standard...However, you should be able to use the feature test macros to cause
getline
to be undefined in<stdio.h>
.This will give you only the definitions from the older POSIX standards. This won't work on some GNU C++ implementations, which is ExTrEmeLY fruSTRaTiNG for some folks.
The relevant section of the manpage is (taken from a glibc manpage, sorry...)
This part of the manpage tells you which macros need to be defined to which values in order to get the definition. My bet is that
_POSIX_C_SOURCE
is already defined by your compiler to200809L
.The idea of feature test macros is that if you define your macros, like
_POSIX_C_SOURCE
,_BSD_SOURCE
,_XOPEN_SOURCE
, etc. to the values you want, you won't need to worry about new library functions clashing with your existing functions. There is also_GNU_SOURCE
, which turns everything on if you use glibc, but I suggest giving that macro a wide berth.