Debian Codegraph @

gcc/testsuite/g++.old-deja/g++.bugs/900227_01.C

raw
1 // { dg-do assemble }
2 // { dg-options "" }
3 // g++ 1.37.1 bug 900227_01
4
5 // g++ allows pointer type values to be converted to integral types which are
6 // not actually large enough to hold the converted values.
7
8 // Section 3.3.4 of the ANSI C standard says:
9
10 // A pointer may be converted to an integral type. The size of the
11 // integer required and the results are implementation defined. If
12 // the space provided is not long enough, the behavior is undefined.
13
14 // I believe that the only proper thing to do in such cases is to generate
15 // errors. After all, if the converted value gets truncated, it is not
16 // likely to be useful after that.
17
18 // Furthermore, as the following example demonstrates, allowing pointers
19 // to be converted to integral types which are not of sufficient size to
20 // completely hold the converted values may cause additional troubles.
21
22 // I tried the following code on 5 different machines and it failed on
23 // all five (unless I also use the GNU assembler and the GNU linker). Three
24 // of the five (Sun3, Sun4, and Symmetry) got link-time errors about byte
25 // offset overflows. The other two (368/SystemV and AViiON) got assembly
26 // time errors about relocatable names used in "constant" expressions.
27
28 // keywords: casts, pointer types, integral types
29
30 // Update 2/10/95: The compiler will now compute these expressions at
31 // runtime. I think this is in the spirit of the GNU compilers (jason).
32
33
34 int main ();
35
36 short s = (short) &main; // { dg-error "11:cast \[^\n\r]* loses precision" "lose" { target { ! ptr_eq_short } } }
37 char c = (char) &main; // { dg-error "10:cast \[^\n\r]* loses precision" "lose" }
38
39 int main () { return 0; }