X-Git-Url: http://erislabs.net/gitweb/?a=blobdiff_plain;f=lib%2Fverify.h;h=d603b1737943811be0dd623dfc224e771cf35668;hb=595fe7f463f7fb72ad6acc58d885585590f8f285;hp=328980fa362917971489b059236f60349bd09c3f;hpb=dddf08d8738ed46c097c018b89385205066c96bf;p=gnulib.git

diff --git a/lib/verify.h b/lib/verify.h
index 328980fa3..d603b1737 100644
--- a/lib/verify.h
+++ b/lib/verify.h
@@ -1,6 +1,6 @@
 /* Compile-time assert-like macros.
 
-   Copyright (C) 2005 Free Software Foundation, Inc.
+   Copyright (C) 2005, 2006 Free Software Foundation, Inc.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -16,25 +16,111 @@
    along with this program; if not, write to the Free Software Foundation,
    Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.  */
 
-/* Written by Paul Eggert and Jim Meyering.  */
+/* Written by Paul Eggert, Bruno Haible, and Jim Meyering.  */
 
 #ifndef VERIFY_H
 # define VERIFY_H 1
 
-/* Each of these macros verifies that its argument R is a nonzero
-   constant expression.  To be portable, R's type must be integer (or
-   boolean).  Unlike assert, there is no run-time overhead.
+/* Each of these macros verifies that its argument R is nonzero.  To
+   be portable, R should be an integer constant expression.  Unlike
+   assert (R), there is no run-time overhead.
 
    There are two macros, since no single macro can be used in all
-   contexts in C.  verify_true (R) is for scalar contexts, where it
-   may be cast to void if need be.  verify (R) is for declaration
+   contexts in C.  verify_true (R) is for scalar contexts, including
+   integer constant expression contexts.  verify (R) is for declaration
    contexts, e.g., the top level.
 
-   The symbols verify_error_if_negative_size__ and verify_function__
-   are private to this header.  */
+   Symbols ending in "__" are private to this header.
+
+   The code below uses several ideas.
+
+   * The first step is ((R) ? 1 : -1).  Given an expression R, of
+     integral or boolean or floating-point type, this yields an
+     expression of integral type, whose value is later verified to be
+     constant and nonnegative.
+
+   * Next this expression W is wrapped in a type
+     struct verify_type__ { unsigned int verify_error_if_negative_size__: W; }.
+     If W is negative, this yields a compile-time error.  No compiler can
+     deal with a bit-field of negative size.
+
+     One might think that an array size check would have the same
+     effect, that is, that the type struct { unsigned int dummy[W]; }
+     would work as well.  However, inside a function, some compilers
+     (such as C++ compilers and GNU C) allow local parameters and
+     variables inside array size expressions.  With these compilers,
+     an array size check would not properly diagnose this misuse of
+     the verify macro:
+
+       void function (int n) { verify (n < 0); }
+
+   * For the verify macro, the struct verify_type__ will need to
+     somehow be embedded into a declaration.  To be portable, this
+     declaration must declare an object, a constant, a function, or a
+     typedef name.  If the declared entity uses the type directly,
+     such as in
+
+       struct dummy {...};
+       typedef struct {...} dummy;
+       extern struct {...} *dummy;
+       extern void dummy (struct {...} *);
+       extern struct {...} *dummy (void);
+
+     two uses of the verify macro would yield colliding declarations
+     if the entity names are not disambiguated.  A workaround is to
+     attach the current line number to the entity name:
+
+       #define GL_CONCAT0(x, y) x##y
+       #define GL_CONCAT(x, y) GL_CONCAT0 (x, y)
+       extern struct {...} * GL_CONCAT(dummy,__LINE__);
+
+     But this has the problem that two invocations of verify from
+     within the same macro would collide, since the __LINE__ value
+     would be the same for both invocations.
+
+     A solution is to use the sizeof operator.  It yields a number,
+     getting rid of the identity of the type.  Declarations like
+
+       extern int dummy [sizeof (struct {...})];
+       extern void dummy (int [sizeof (struct {...})]);
+       extern int (*dummy (void)) [sizeof (struct {...})];
+
+     can be repeated.
+
+   * Should the implementation use a named struct or an unnamed struct?
+     Which of the following alternatives can be used?
+
+       extern int dummy [sizeof (struct {...})];
+       extern int dummy [sizeof (struct verify_type__ {...})];
+       extern void dummy (int [sizeof (struct {...})]);
+       extern void dummy (int [sizeof (struct verify_type__ {...})]);
+       extern int (*dummy (void)) [sizeof (struct {...})];
+       extern int (*dummy (void)) [sizeof (struct verify_type__ {...})];
+
+     In the second and sixth case, the struct type is exported to the
+     outer scope; two such declarations therefore collide.  GCC warns
+     about the first, third, and fourth cases.  So the only remaining
+     possibility is the fifth case:
+
+       extern int (*dummy (void)) [sizeof (struct {...})];
+
+   * This implementation exploits the fact that GCC does not warn about
+     the last declaration mentioned above.  If a future version of GCC
+     introduces a warning for this, the problem could be worked around
+     by using code specialized to GCC, e.g.,:
+
+       #if 4 <= __GNUC__
+       # define verify(R) \
+	   extern int (* verify_function__ (void)) \
+		      [__builtin_constant_p (R) && (R) ? 1 : -1]
+       #endif
+
+   * In C++, any struct definition inside sizeof is invalid.
+     Use a template type to work around the problem.  */
+
 
 /* Verify requirement R at compile-time, as an integer constant expression.
-   Return true.  */
+   Return 1.  */
 
 # ifdef __cplusplus
 template <int w>