leichter@yale-comix.UUCP (Jerry Leichter) (04/20/84)
Brian Jones (drutx!qwerty) presents a program to test C compilers for proper
evaluation of expressions involving side-effects and coercions. I've tried
his sample program on two more C compilers, with the following results:
VAX11-C (VMS) DECUS C (VMS, compatibility mode)
------------- ---------------------------------
20 2 20 2
20 3 0 3
30 5 30 5
30 6 33 6
The VAX11-C results are identical to those Brian found for "Unix on a Dec
11/70, Ver. 3.0"; he believes this is the (a?) "right answer". (I'm not
questioning his judgement; I haven't looked at the code closely enough to form
any sort of opinion.)
The DECUS C results - which would be the same on all the operating systems
and machines (VAXes and 11's) DECUS C runs under - are identical to what
Brian found for "Mark Williams Co. C on an Intel 86/330 System, with 8087
support". I find this an interesting statement, since I think the same person
(Dave Conroy) did most of both compilers.
DECUS C required two changes to the code as distributed. The first is a known
DECUS C restriction: Local variables cannot be initialized. This required
that j in main be initialized in a separate statement. The second is either a
bug or a change in the language definition (DECUS C pre-dates K&R): DECUS C
requires a parenthesized expression after return; this requires that:
return a;
in ifunc() be changed to:
return (a);
For people who might have missed the original article and want to try this
test out on their favorite compiler, here is the version I used, with the
edits for DECUS C in place (might help on other old compilers):
/*
* Simple E1 op= E2 test
*/
#include <stdio.h>
main()
{
int i,
j,
*ifunc();
j = 0;
i = 100;
*ifunc(&i, &j) = *ifunc(&i, &j) * .2;
printf("%d %d\n", i, j);
i = 100;
*ifunc(&i, &j) *= .2;
printf("%d %d\n", i, j);
i = 100;
*ifunc(&i, &j) = *ifunc(&i, &j) / 3.3;
printf("%d %d\n", i, j);
i = 100;
*ifunc(&i, &j) /= 3.3;
printf("%d %d\n", i, j);
}
int *ifunc(a, b)
int *a, *b;
{
++*b;
return (a);
}
Brian's results, in summary:
1 2 3 4 5 6
---- ---- ---- ---- ---- ----
19 2 20 2 20 2 20 2 20 2 19 2
0 3 0 3 20 3 20 4 0 3 19 4
30 5 30 5 30 5 30 6 30 5 30 5
33 6 33 6 30 6 30 8 33 6 30 6
1) Maxi-Unix on an Ahmdahl (sp?), Ver. 5.0
2) Unix on a Vax 11/780, Ver. 5.0
3) Unix on a Dec 11/70, Ver. 3.0 **right answer**
4) Unix on a 3B20, Ver. 5.0
5) Mark Williams Co. C on an Intel 86/330 System, with 8087 support
6) Ancient Yourdon (I think) compiler on a Dec 11/44 **right answer**
-- Jerry
decvax!yale-comix!leichter
leichter@yaleqwerty@drutx.UUCP (JonesBD) (04/23/84)
I would like to correct the results posted in column 6. It should read: 6 ----- 19 2 19 3 (was 4) 30 5 30 6 My original text was wrong. SBT (Sorry 'Bout That) Brian Jones
gwyn@brl-vgr.ARPA (Doug Gwyn ) (05/02/84)
DECUS C does not "predate K&R". The requirement that one write return (a); rather than return a; has not been in the C language since 1975 and I doubt it was there before.