[saagarjha]

I think it’s a really easy sell, actually: if you go past the end of the array far enough you end up accessing the stack which includes parts of the program like “where does this function return to” or “what is the index used to perform this access” or “there is no page mapped there”. None of these are arbitrary values.

[lelanthran]

The "anything can happen" means that the compiler can simply silently refuse to emit the code does the access.

Documenting that the instructions to access will always be eliminated makes it easier to predict what will happen.

[bee_rider]

Can you unravel this further (for those of us who don’t know compilers)? I’ve always assumed access past the end of an array can’t always be detected in C, so I don’t see how those instructions could be eliminated.

For example, a dynamically linked library that takes in a pointer, and then writes to the 10 ints after it—whether or not this behavior is defined is determined after that library is compiled, right?

[horsawlarway]

I think the disconnect here is that you're operating on the assumptions built by using common architectures that have solved these problems in implementation specific forms, and you're used to those solutions.

But just because those forms are common, doesn't mean the behavior is actually defined.

Ex - I might be using a vendor specific compiler for custom embedded devices where dynamic linking isn't available at all, and which might have complicated storage mechanisms that look nothing like standard memory pages.

[bee_rider]

I’m not sure there’s a disconnect at all (note that I’m not saagarjah, they and lelanthran seem to be pushing back on each other’s opinions; I’m just asking a clarifying question).

[horsawlarway]

Yes, and I'm saying your clarifying question hints at a misunderstanding.

You're already deep into the bowels of implementation specific behavior by the time we talk about dynamic linking. The C standard doesn't have anything to say about it at all.

My read on the above conversation is basically a discussion about asking/requiring vendors to properly document their implementation, as opposed to leaving it undocumented (the default - given my experience with hardware manufacturers...).

I don't think the real takeaway is that "instructions should be eliminated in case [blah blah blah]" it's that "Something is going to happen, please tell me what that is on your system, instead of leaving it as UB" (Basically - make UB in the standard implementation defined behavior from the vendor).

My read is that this won't happen because it's genuinely incredibly difficult to do, and this isn't a space overflowing with capital to allocate to the problem. But I do think there's merit to the idea of pushing vendors to provide coverage in this space AT SOME POINT.

[lelanthran]

> I’ve always assumed access past the end of an array can’t always be detected in C, so I don’t see how those instructions could be eliminated.

"Can't always be detected" is jut a different way of saying "Can sometimes be detected".

Upon detection, I'd rather that the compiler still emit the instructions, not elide the code altogether.

[twoodfin]

Now the behavior of your compiler/runtime stack is dependent on the sophistication of your compiler or runtime relative to the particular code at issue + the specific information available statically or dynamically in the instance.

That does not seem like an improvement if your goal is predictable, consistent behavior.

[lelanthran]

> Now the behavior of your compiler/runtime stack is dependent on the sophistication of your compiler or runtime relative to the particular code at issue + the specific information available statically or dynamically in the instance.

> That does not seem like an improvement if your goal is predictable, consistent behavior.

Getting unpredictable outcomes only on some items is better than getting unpredictable outcomes on every item. This is what I meant when I said moving UB to IB might not fix everything, but it is certainly a solid start!

[saagarjha]

Yes, but usually you don't want this. You think you do, but you don't: you can't always eliminate these, and often eliminating the extra accesses is not the most efficient thing to do either. Sometimes it's faster to have the loads and not check, sometimes you can check and skip that path, etc.