Type Punning in C and C++

Type Punning in C and C++blog.pwkf.org

I had a bug that took me a while to track down. The problem was type punning. A pointer cast worked fine at -O0 and silently broke at -O2. The C vs C++ distinction here is genuinely treacherous, and most blog posts on…

36 points•ingve•9 days ago•23 comments•

#type #punning #type punning #track #worked #pointer #cast #bug

23 comments

eqvinox9 days ago

The example included below "Why C and C++ Differ" is UB in both C and C++ (and says nothing about "why" C & C++ differ). The article isn't overall wrong but that bit is just...

(Feels a bit like LLM junk, but honestly not sure.)

LoganDark9 days ago

> A pointer cast worked fine at -O0 and silently broke at -O2.

Feels like LLM to me. I looked at some of the rest of the article and... pretty much certain now.

rfgplk9 days ago

> (Feels a bit like LLM junk, but honestly not sure.)

Why leave such disparaging comments?

munificent9 days ago

Because if the comment is correct, it's useful signal for other potential readers to know that they can skip it.

I try to be as charitable as possible when reading and commenting online, but we don't have infinite attention and thanks to AI, it's easier than ever to end up wasting it on things that aren't worth reading.

(I'm not claiming that the article here is or is not an example of that.)

quietbritishjim9 days ago

Even if it's not from an LLM, it's still wrong! You've focused on the last important of that comment, really.

scoopr9 days ago

I just watched video[0] that argues if you can consteval something, then it must be free of UB, which seems compelling, at least for these lower level helpers.

Unfortunately, it seems memcpy is not constexpr, so cannot be used like this, but std::bit_cast actually works[1] as constexpr, so I think in C++ that would now be the most preferred use. It won't allow the union either.

[0] https://www.youtube.com/watch?v=-LAXqqqX274 [1] https://godbolt.org/z/xGzjTMGvv

leni5369 days ago

Careful, only language UB is guaranteed to be detected in constant evaluation, UB in standard library functions isn't. It is a good sanity check though.

Also it seems that bit_cast in particular is going to be strengthened in this regard:

https://cplusplus.github.io/LWG/issue4539

rfgplk9 days ago

__builtin_memcpy is constexpr under clang with constraints

antiloper9 days ago

> The C vs C++ distinction here is genuinely treacherous

What's so genuine about this?

tialaramex9 days ago

Despite the fact they're two different languages, people are often taught as though it's basically just a subset relationship and then they carry this through into the actual software they write.

Because neither of these are memory safe languages, you are required to ensure you've made no mistakes or else anything might happen.

ndr9 days ago

It continues with

> and most blog posts on the topic get it wrong.

This smells like Claudism/LLMish.

StilesCrisis9 days ago

The conclusion is simply wrong. `union` should not be used for type-punning in C++, or in C code which might be compiled by a C++ compiler.

I can't downvote posts yet, but if you have the ability, consider it. This is clearly LLM slop without human review.

eqvinox9 days ago

> C code which might be compiled by a C++ compiler.

I really hope you're talking mostly about header files; for actual code this is such a horrible idea (and in a lot of cases just won't work without massive efforts.) Even for header files, arguably one ought to really know what they're doing.

StilesCrisis8 days ago

For sure, a union which is used as API surface to a C library which a C++ codebase uses is the most obvious and dangerous footgun here.

The article never even discusses how C++ actually implements unions to begin with, so it's just incomplete. (Only the "active" member is meaningful, the others are considered as meaningless and shouldn't be touched.)

LelouBil9 days ago

I don't understand the example in "Why C and C++ Differ".

Shouldn't the code return 3 in the base case and 4 if the check for 2 was assumed to always hold ?

bena9 days ago

You're halfway there. It's a little more messed up than that.

IIRC, this is all little-endian. So assume our 8 bytes labeled A-H. For a 32 bit/4 byte integer, they will be read as DCBA. For a 64 bit/8 byte integer, HGFEDCBA.

So the struct is set up like a = DBCA, b = HGFE. So when we assign 2 and 3 to a and b, it should look like this in memory:

00000010 00000000 00000000 00000000 and 00000011 00000000 00000000 00000000

When we cast that to a uint64 and assign 4 to it, we should wind up with:

00000100 00000000 00000000 00000000 00000000 00000000 00000000 00000000

Which effectively zeros out b.

So if the conditional is evaluated, it will evaluate to false, we return b, which is 0.

If the conditional is not evaluated, we should return the value in a, which is 4.

LelouBil8 days ago

Oooh thanks, I missed that the cast was into a 64 bit integer

Read the full thread on Hacker News →

23 comments

Related stories