#4085 Undiagnosed constraint violations on arithmetic operations

[Christopher Bazley]

1: Sample code that reproduces the problem.

/* N3886 6.5.7p2: the other operand of pointer addition is integer. */
int *pointer;
float floating;
int *test1(void) {
    return pointer + floating; /* ERROR */
}

/* N3886 6.5.7p3: pointer subtraction requires compatible pointed-to types. */
int *pointer;
char *other;
long test2(void) {
    return pointer - other; /* ERROR */
}

/* N3886 6.5.7p3: pointed-to object types must also be complete. */
struct incomplete;
struct incomplete *left, *right;
long test3(void) {
    return left - right; /* ERROR */
}

2: Exact command used to run SDCC on this sample code
./bin/sdcc -c bug.c

3: SDCC version tested

mcs51/z80/z180/r2k/r2ka/r3ka/r4k/r5k/r6k/sm83/tlcs90/ez80/z80n/r800/ds390/pic16/pic14/TININative/ds400/hc08/s08/stm8/pdk13/pdk14/pdk15/mos6502/mos65c02/huc6280/f8/f8l TD- 4.6.2 #0 (macOS ARM64)

4: Copy of the error message or incorrect output, or a clear description of the observed versus expected behavior.

Every arithmetic operation violates a constraint. According to the ISO C standard, all constraint violations must be diagnosed. Clang, GCC and TCC report the constraint violations but SDCC does not. See https://godbolt.org/z/M1azch6Kd

[Christopher Bazley]

Arithmetic on type void * also violates the ISO C constraint. Clang reports 'warning: arithmetic on pointers to void is a GNU extension [-Wgnu-pointer-arith]' for arithmetic on type void *. It's not clear to me whether SDCC deliberately permits this behaviour as an extension or not. Strictly speaking, it should still produce a diagnostic in some kind of strict mode (if it has one).

[Philipp Klaus Krause]

Currently it is (intentionally) allowed, but the (for some cases) missing warnings are a bug. A patch is being discussed at [bugs:#4056]

[Christopher Bazley]

I'm attaching the relevant one of five candidate patches for bugs 4085...4089.

For testing purposes, all five were applied in numeric order to the baseline of f68fca8a4368ece9fe93b21f2e0341282b104359 from the git mirror. The following commands then completed successfully:

make -C device/lib clean
make -C device/lib -j2
make -j2
make -C support/regression clean-results
make -C support/regression -j2 test-ucz80
make -C support/valdiag clean
make -C support/valdiag -j2

Test results:

• ucz80 regression: 0 failures, 36,745 tests, 6,409 test cases;
• valdiag: 0 failures on all 19 configured targets, with 377 test cases per target.