ref: ad9537bd2daa3be47f01d68f25c034aaab83ed15
dir: /README/
Deviations from standard C
===========================
This compiler aims to be fully compatible with the C99 standard, but
it will have some differences:
- Type qualifiers are accepted but ignored.
-----------------------------------------
Type qualifiers make the type system ugly, and their uselessness adds
unnecessary complexity to the compiler (and increased compilation time):
- const: The definition of const is not clear in the standard.
If a const value is modified the behaviour is undefined
behaviour. It seems it was defined in order to be able to
allocate variables in ROM rather than error detection. This
implememtation will not warn about these modifications and
the compiler will treat them like normal variables (the standard
specifies that a diagnostic message must be printed).
- volatile: This qualifier was added to the standard
to be able to deal with longjmp (local variables that are not
volatile have undefined state) and for memory mapped registers
or variables whose values are modified asynchronously. This can
be achieved with special pragma values though.
In the first case, it generates a lot of problems with modern
processors and multithreading, when not holding the value in a
register is not good enough (an explicit memory barrier is needed).
In the second case, this is non-portable code by definition
(depending on the register mapped), so it is better to deal with
it using another solution (compiler extensions or direct
assembly). Since it is needed for the correct behaviour
of portable programs that use longjmp, this specifier eventually
will be implemented.
- restrict: This qualifier can only be applied to pointers to
mark that the pointed object has no other alias. This qualifier
was introduced to be able to fix some performance problems in
numerical algorithms, where FORTRAN could achieve a better
performance (and in fact even with this specifier FORTRAN has a
better performance in this field). Ignoring it doesn't make the
compiler non-standard and in almost all applications the performance
will be the same.
- Function type names
-------------------
C99 allows you to define type names of function types and write something
like:
int f(int (int));
Accepting function types in type names (or abstract declarators) makes the
grammar ambiguous because it is impossible to differentiate between:
(int (f)) -> function returning int with one parameter of type f
(int (f)) -> integer variable f
If you don't believe me try this code:
int
f(int g())
{
return g();
}
Function type names are stupid, because they are used as an alias
of the function pointer types, but it is stupid that something
like sizeof(int (int)) is not allowed (because here it should be
understood as the size of a function), but f(int (int)) is allowed
because it is understood as a parameter of function pointer type.
This complexity is not needed at all as function pointers fix all these
problems without this complexity (and they are the more usual
way of writing such code).
- Definition of variables with incomplete type
---------------------------------------------
C89 allows the definition of variables with incomplete type that
have external linkage and file scope. The type of the variable
is the composition of all the definitions find in the file. The exact
rules are a bit complex (3.7.2) so SCC ignores them at this moment
and it does not allow any definition of variables with incomplete type.
If you don't believe me try this code:
struct foo x;
struct foo {
int i;
};
- Variadic function alike macros
------------------------------
The standard (C99 6.10.3 c 4) forces to pass more parameters than
the number of parameters present in the variadic argument list
(excluding ...). Scc accepts a parameter list with the same number
of arguments.
#define P(a, ...) a
P(1)