-
Notifications
You must be signed in to change notification settings - Fork 12.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
transmute documentation does not explicitly mention alignment requirement #82493
Comments
That looks like well-defined behavior to me. The alignment of Regarding the quote paragraph, this references a conversion of |
Thanks for the clarification! So the documentation could still be updated to clarify when alignment matters, right? |
I do not see a feasible way for the You example (the text you quoted) is specific to |
However, maybe the comment in the example could be worded a bit more carefully to explain that alignment matters in this specific case because we are transmuting a pointer (implicitly as part of the Honestly, that entire example is a bit weird. It says "still using transmute" but the code this comments doesn't even use |
If alignment is not a concern when transmuting values (because the compiler will move the memory to a stack location which is properly aligned), I think it would be worth explicitly saying this. Unless I was explicitly told (like Heroic explained), it would not have occured to me that it's moving into a proper aligned location. With regards to "explain all possible types", I don't think this is necessary. Saying "The resulting type must be a valid value" might be enough. Or maybe the:
…should be changed to encourage reading that page more strongly ("Make sure you understand all the dangers of using
It's not even use // To "transmute" the contents of a heap-allocated container to something else,
// you must make sure that both the size *and alignment* of the items match,
// so that the internal representation of the container is not affected:
let v_from_raw = unsafe {
// Ensure the original vector is not dropped.
let mut v_clone = std::mem::ManuallyDrop::new(v_clone);
Vec::from_raw_parts(v_clone.as_mut_ptr() as *mut Option<&i32>,
v_clone.len(),
v_clone.capacity())
}; |
I'm not a fan of listing all the things that are not a concern. That's a long list. However, you are not the first person to suggest this, so I concede. I guess I will never understand why people expect alignment to be an issue here, but it seems they do, and I don't have to understand everything. ;)
It already says that (in double-negated form): "Neither the original, nor the result, may be an invalid value."
There are people (myself included) that use the term "to transmute" for any kind of "type-changing reinterpretation of raw bytes", whether it is via |
In my case, it's because I'm not sure if it was undefined behaviour to read a type from memory with the wrong alignment. IIRC, there are some CPU architectures that fault when a misaligned memory access occurs? Or maybe I'm misremembering. Basically, I wasn't sure if "interpret this bunch of bytes as this other type" is something that is okay to do when "the bunch of bytes" is misaligned.
My bad, somehow missed it. |
It is UB to perform an unaligned read or write. But Anyway, you made some good suggestions -- do you plan to turn these into a PR, or should we mark this as an E-easy issue for someone else to write the PR? |
Improve transmute docs with further clarifications Closes rust-lang#82493. Please let me know if any of the new wording sounds off, English is not my mother tongue.
Improve transmute docs with further clarifications Closes rust-lang#82493. Please let me know if any of the new wording sounds off, English is not my mother tongue.
Improve transmute docs with further clarifications Closes rust-lang#82493. Please let me know if any of the new wording sounds off, English is not my mother tongue.
The documentation for
std::mem::transmute
reads:However, it does not mention that they must also have the same alignment. You would only learn about this if you try to dig into the code examples further down the page:
Must the types
T
andU
have the same alignment, or is that a particular quirk of the example?In any case, I think the documentation should explicitly mention alignment, because the compiler succeeds to compile the following with no warnings (playground), and I believe this is undefined behaviour:
@rustbot modify labels: +T-doc +C-enhancement
The text was updated successfully, but these errors were encountered: