Is this really a problem for tcc? An old version of VC produces the same sizes as tcc. The spec seems to say (not sure I'm reading this right, first time I've read the spec)

"An implementation may allocate any addressable storage unit large enough to hold a bitfield.....snip..... the order of bitfields within a unit is ( ... high to low or ..... low to high) implementation defined. The alignment of the addressable storage unit is undefined"

This seems to suggest that each implementation can do what it wants with bitfields and that passing them between different compilers is probably undefined. Having said all that, I'm not overly worried if it gets changed, just seems like risk for something that might not be broken. And, I am not an expert on C compiler internals. I do pass structures a lot between tcc and other compilers, but they are all carefully crafted with PACK directives/pragmas to ensure exact memory layout, and I dont use bitfields as part of compiler - hence why I looked at this.


David Mertens wrote:
Hello everyone,

I recently uncovered some segfaulting code when compiling code with macros that manipulate certain Perl structs on 64-bit Linux. I boiled the problem down to a discrepancy between how tcc and gcc determine the size needed by a series of bit fields. The tcc-compiled function would get the Perl interpreter struct produced by gcc-compiled code, then reach into the wrong memory slot for something. A reduced example is provided below.

Question 1: Would anybody be opposed to changing tcc's behavior to match gcc's behavior here? This could lead to binary incompatibility with object code previously compiled with tcc, but that seems to me highly unlikely to be a real problem for anyone.

Question 2: Does anybody know tccgen.c well enough to fix this? I can work on it, but if anybody knows exactly where this goes wrong, it would save me a few hours.

#include <stdint.h>
#include <stdio.h>
struct t1 {
    uint8_t op_type:1;
    uint8_t op_flags;
struct t2 {
    uint32_t op_type:1;
    uint8_t op_flags;
struct t3 {
    unsigned op_type:1;
    char op_flags;

int main() {
    printf("t1 struct size: %ld\n", sizeof(struct t1));
    printf("t2 struct size: %ld\n", sizeof(struct t2));
    printf("t3 struct size: %ld\n", sizeof(struct t3));
    return 0;

With tcc, this prints:
t1 struct size: 2
t2 struct size: 8
t3 struct size: 8

With gcc, this prints:
t1 struct size: 2
t2 struct size: 4
t3 struct size: 4

This suggests that with tcc, the number of bytes given to a series of bitfields in a struct depends upon the integer type of the bitfield. In particular, plain old "unsigned" is interpreted (in 64-bit context) to be "unsigned int", which has 32 bits. This is incompatible with gcc's interpretation.

The relevant code is, I think, in tccgen.c's struct_decl. However, I can't quite tease apart where the declaration comes in, and how it effect struct size calculations.


 "Debugging is twice as hard as writing the code in the first place.
  Therefore, if you write the code as cleverly as possible, you are,
  by definition, not smart enough to debug it." -- Brian Kernighan

Tinycc-devel mailing list

Tinycc-devel mailing list

Reply via email to