2007-01-02 Douglas Gregor <doug.gregor@gmail.com>
* c-common.c(c_common_nodes_and_builtins): Since variants of
void_type_node get built before it is given a name, we need to
give those variants the name, too.
(complete_array_type): We need to work with the canonical main
type of the array, from which we will build the qualified version.
* params.def (PARAM_VERIFY_CANONICAL_TYPES): New.
* print-tree.c (print_node): Display canonical type information
for each type.
* stor-layout.c (layout_type): When we don't know the
alignment of a type for which we're building an array, we end up
guessing wrong, so make the type require structural equality.
* tree.c (make_node_stat): When we build a new type, it is its
own canonical type.
(build_type_attribute_qual_variant): When building an attribute
variant, its canonical type is the non-attribute variant. However,
if the attributes are target-dependent and they differ, we need to
use structural equality checks for this type.
(build_qualified_type): A qualified type is not equivalent to its
unqualified variant; set the canonical type appropriately.
(build_distinct_type_copy): When building a distinct type from
another type, the new type is its own canonical type.
(build_variant_type_copy): When building a new type variant, we
assume that it is equivalent to the original type.
(build_pointer_type_for_mode): When building a pointer type, also
build a canonical type pointer.
(build_reference_type_for_mode): When building a reference type,
also build a canonical type reference.
(build_index_type): When we can't hash an index type (e.g.,
because its maximum value is negative), the index type requires
structural equality tests.
(build_array_type): Build the canonical form of an array type.
(build_function_type): Function types require structural equality,
because they contain default arguments, attributes, etc.
(build_method_type_directly): Ditto for method types.
(build_offset_type): Build the canonical offset type.
(build_complex_type): Build the canonical vector type.
(make_vector_type): Build the canonical vector type.
* tree.h (TYPE_CANONICAL): New.
(TYPE_STRUCTURAL_EQUALITY_P): New.
(SET_TYPE_STRUCTURAL_EQUALITY): New.
(struct tree_type): Added "canonical" field.
* params.h (VERIFY_CANONICAL_TYPES): New.
* doc/c-tree.texi (TYPE_CANONICAL): Document.
(TYPE_STRUCTURAL_EQUALITY_P): Document.
(SET_TYPE_STRUCTURAL_EQUALITY): Document.
* doc/invoke.texi (verify-canonical-types): Document --param
parameter for verifying canonical types.
2007-01-02 Douglas Gregor <doug.gregor@gmail.com>
* typeck.c (structural_comptypes): Renamed from "comptypes".
(comptypes): Use canonical type information to perform fast type
comparison. When VERIFY_CANONICAL_TYPES, verify that the
canonical type comparison returns the same results as we would see
from the current, structural check. Support COMPARE_STRUCTURAL
when we need structural checks.
* decl.c (typename_compare): Fix comment.
(build_typename_type): TYPENAME_TYPE nodes require structural
equality checks, because they resolve different based on the
current class type.
(make_unbound_class_template): UNBOUND_CLASS_TEMPLATE nodes
require structural equality checks (for now).
(build_ptrmemfunc_type): Build the canonical pointer to member
function type.
(compute_array_index_type): Whenever we build a new index type
to represent the size of an array in a template, we need to mark
this index type as requiring structural equality. This goes for
arrays with value-dependent sizes with the current ABI, or all
arrays with ABI-1.
* tree.c (cplus_array_hash): New.
(struct cplus_array_info): New.
(cplus_array_compare): New.
(cplus_array_htab): New.
(build_cplus_array_type_1): Use a hash table to cache the array
types we build. Build the canonical array type for each array
type.
(cp_build_qualified_type_real): When building a cv-qualified array
type, use the hash table of array types and build canonical array
types as necessary.
(bind_template_template_parm): BOUND_TEMPLATE_TEMPLATE_PARM nodes
use structural equality (for now).
* cp-tree.h (COMPARE_STRUCTURAL): New.
* pt.c (canonical_template_parms): New.
(canonical_type_parameter): New.
(process_template_parm): Find the canonical type parameter.
(lookup_template_class): When we have named the primary template
type, set the canonical type for our template class to the primary
template type. If any of the template arguments need structural
equality checks, the template class needs structural equality
checks.
(tsubst): When reducing the level of a template template
parameter, we require structural equality tests for the resulting
parameter because its template parameters have not had their types
canonicalized. When reducing a template type parameter, find the
canonical reduced type parameter.
(any_template_arguments_need_structural_equality_p): New.
2007-01-02 Douglas Gregor <doug.gregor@gmail.com>
* objc-act.c (objc_build_volatilized_type): Keep track of
canonical types.
(objc_get_protocol_qualified_type): Ditto.
From-SVN: r120341
Copyright (C) 2000, 2003 Free Software Foundation, Inc.
This file is intended to contain a few notes about writing C code
within GCC so that it compiles without error on the full range of
compilers GCC needs to be able to compile on.
The problem is that many ISO-standard constructs are not accepted by
either old or buggy compilers, and we keep getting bitten by them.
This knowledge until know has been sparsely spread around, so I
thought I'd collect it in one useful place. Please add and correct
any problems as you come across them.
I'm going to start from a base of the ISO C90 standard, since that is
probably what most people code to naturally. Obviously using
constructs introduced after that is not a good idea.
For the complete coding style conventions used in GCC, please read
http://gcc.gnu.org/codingconventions.html
String literals
---------------
Irix6 "cc -n32" and OSF4 "cc" have problems with constant string
initializers with parens around it, e.g.
const char string[] = ("A string");
This is unfortunate since this is what the GNU gettext macro N_
produces. You need to find a different way to code it.
Some compilers like MSVC++ have fairly low limits on the maximum
length of a string literal; 509 is the lowest we've come across. You
may need to break up a long printf statement into many smaller ones.
Empty macro arguments
---------------------
ISO C (6.8.3 in the 1990 standard) specifies the following:
If (before argument substitution) any argument consists of no
preprocessing tokens, the behavior is undefined.
This was relaxed by ISO C99, but some older compilers emit an error,
so code like
#define foo(x, y) x y
foo (bar, )
needs to be coded in some other way.
free and realloc
----------------
Some implementations crash upon attempts to free or realloc the null
pointer. Thus if mem might be null, you need to write
if (mem)
free (mem);
Trigraphs
---------
You weren't going to use them anyway, but some otherwise ISO C
compliant compilers do not accept trigraphs.
Suffixes on Integer Constants
-----------------------------
You should never use a 'l' suffix on integer constants ('L' is fine),
since it can easily be confused with the number '1'.
Common Coding Pitfalls
======================
errno
-----
errno might be declared as a macro.
Implicit int
------------
In C, the 'int' keyword can often be omitted from type declarations.
For instance, you can write
unsigned variable;
as shorthand for
unsigned int variable;
There are several places where this can cause trouble. First, suppose
'variable' is a long; then you might think
(unsigned) variable
would convert it to unsigned long. It does not. It converts to
unsigned int. This mostly causes problems on 64-bit platforms, where
long and int are not the same size.
Second, if you write a function definition with no return type at
all:
operate (int a, int b)
{
...
}
that function is expected to return int, *not* void. GCC will warn
about this.
Implicit function declarations always have return type int. So if you
correct the above definition to
void
operate (int a, int b)
...
but operate() is called above its definition, you will get an error
about a "type mismatch with previous implicit declaration". The cure
is to prototype all functions at the top of the file, or in an
appropriate header.
Char vs unsigned char vs int
----------------------------
In C, unqualified 'char' may be either signed or unsigned; it is the
implementation's choice. When you are processing 7-bit ASCII, it does
not matter. But when your program must handle arbitrary binary data,
or fully 8-bit character sets, you have a problem. The most obvious
issue is if you have a look-up table indexed by characters.
For instance, the character '\341' in ISO Latin 1 is SMALL LETTER A
WITH ACUTE ACCENT. In the proper locale, isalpha('\341') will be
true. But if you read '\341' from a file and store it in a plain
char, isalpha(c) may look up character 225, or it may look up
character -31. And the ctype table has no entry at offset -31, so
your program will crash. (If you're lucky.)
It is wise to use unsigned char everywhere you possibly can. This
avoids all these problems. Unfortunately, the routines in <string.h>
take plain char arguments, so you have to remember to cast them back
and forth - or avoid the use of strxxx() functions, which is probably
a good idea anyway.
Another common mistake is to use either char or unsigned char to
receive the result of getc() or related stdio functions. They may
return EOF, which is outside the range of values representable by
char. If you use char, some legal character value may be confused
with EOF, such as '\377' (SMALL LETTER Y WITH UMLAUT, in Latin-1).
The correct choice is int.
A more subtle version of the same mistake might look like this:
unsigned char pushback[NPUSHBACK];
int pbidx;
#define unget(c) (assert(pbidx < NPUSHBACK), pushback[pbidx++] = (c))
#define get(c) (pbidx ? pushback[--pbidx] : getchar())
...
unget(EOF);
which will mysteriously turn a pushed-back EOF into a SMALL LETTER Y
WITH UMLAUT.
Other common pitfalls
---------------------
o Expecting 'plain' char to be either sign or unsigned extending.
o Shifting an item by a negative amount or by greater than or equal to
the number of bits in a type (expecting shifts by 32 to be sensible
has caused quite a number of bugs at least in the early days).
o Expecting ints shifted right to be sign extended.
o Modifying the same value twice within one sequence point.
o Host vs. target floating point representation, including emitting NaNs
and Infinities in a form that the assembler handles.
o qsort being an unstable sort function (unstable in the sense that
multiple items that sort the same may be sorted in different orders
by different qsort functions).
o Passing incorrect types to fprintf and friends.
o Adding a function declaration for a module declared in another file to
a .c file instead of to a .h file.