Re: PDUMP questions

KJ> Olivier Galibert writes: KJ> I remeber reviewing some code that made this assumption and doing KJ> some reading of the H&S C reference. I came to the conclusion KJ> that treating a struct like an array was portable. I'm away from KJ> my books at the moment so I can't quote anything yet. But I'm KJ> curious, Martin, why you think it isn't portable? Sorry, it would be more correct to argue that we can portably rely on the ORDER of elements in a struct, but not the presence of padding. For example, the compiler might decide to compile struct cons { Lisp_Object car; Lisp_Object cdr; } as struct cons { 4-byte car; 4-byte pad; 4-byte cdr; 4-byte pad; } because, for example, the hardware is 64-bit, but the OS is still 32-bit, and pointers are more efficient when 64-bit aligned. In practice, today, nobody is actually going to do that. Still, it's nice to code to the ANSI C standard when possible. No one used to do alias analysis either (except xlc...). OG> On Tue, Jan 25, 2000 at 06:47:48PM -0800, Martin Buchholz wrote: OG> If you do such a "fix", do it everywhere. There are a lot of places OG> where I do this assumption. I do others, btw, like the fact that all OG> pointers are of the same size. OG> In particular, I'm interessed in how you're going to handle that one: OG> #define BUFFER_SLOTS_SIZE (offsetof (struct buffer, BUFFER_SLOTS_LAST_NAME) - offsetof (struct buffer, BUFFER_SLOTS_FIRST_NAME) + sizeof (Lisp_Object)) OG> #define BUFFER_SLOTS_COUNT (BUFFER_SLOTS_SIZE / sizeof (Lisp_Object)) This code has the same portability `problem'. The `correct' thing is to have the buffer slots in an array of Lisp_Object, and then use countof to portably count them. If something is being used like an array, maybe it should _be_ an array. If you claim this is not worth fixing, I might agree with you - after all this is old code. But you're adding new code. But a more fundamental reason is maintainability. Pretend you're going to delete a member from a Lisp Object type struct. You do a global grep for the member name, and remove those. And then you get strange crashes because you didn't change a `6' to a `5'. There are times when I argue for verboseness in code. This is one of them. Although your code works, it is not as `clean' as it could be. However, don't consider that you have to change this. There are lots of real bugs to fix. But I might, whenever I hack on this. I assume I have your permission. Pedantically yours, Martin