Repository navigation
Bindgen can't translate #pragma pack(n) into #[repr(packed = "n")] #537
Description
Given the header
#pragma pack(1)
struct foo { int x; };bindgen generates
#[repr(C)]
pub struct foo {
pub x: ::std::os::raw::c_int,
}which uses repr(C) instead of repr(packed), and therefore has the wrong alignment:
running 1 test
test bindgen_test_layout_foo ... FAILED
failures:
---- bindgen_test_layout_foo stdout ----
thread 'bindgen_test_layout_foo' panicked at 'assertion failed: `(left == right)` (left: `4`, right: `1`): Alignment of foo', testcase.rs:12
note: Run with `RUST_BACKTRACE=1` for a backtrace.
failures:
bindgen_test_layout_foo
test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured
In fact, bindgen emits a warning to this effect:
WARN:bindgen::codegen: Type foo has an unkown attribute that may affect layout
Details
Steps to reproduce:
$ cat testcase.h
#pragma pack(1)
struct foo { int x; };
$ RUST_LOG=bindgen bindgen testcase.h -o testcase.rs
$ rustc --test testcase.rs
$ ./testcase
bindgen 0.22.0
rustc 1.15.1 (021bd294c 2017-02-08)
Activity
fitzgen commentedon Feb 22, 2017
Thanks for the bug report! I'm not super familiar with
#pragma pack-- is it standard? Or at least widely used? If not, we might want to consider whether implementing support is worth the pay off, and what ongoing maintenance costs might be.If we did support this, I think we would want to use
#[repr(C, packed)], not just#[repr(C)].asayers commentedon Feb 22, 2017
#pragma packis understood by GCC and Visual Studio, and I think Borland too. I don't know about clang.I hope not! Abusing structs for serialisation purposes doesn't seem to be a good idea, and can cause problems. Unfortunately, I know of least one library which does this extensively.
bindgen is currently generating structs which have the wrong layout; this could also lead to some fairly strange behaviour. I think there are a number of ways you could go here:
#pragma pack(*)it aborts.#pragma pack(*)it spits out a warning and stops generating bindings until it sees#pragma pack()(which resets things to normal).repr(packed)can lead to undefined behaviour in rust.Ah yes, you're right.
emilio commentedon Feb 22, 2017
I don't think we can even see
#pragma packedfrom libclang?I guess we can try to guess if we see the alignment is 1 and the relevant fields are packed when looking at their offsets, but...
radupopescu commentedon Apr 11, 2017
Hello,
I'm having a similar issue on macOS 10.12.4. When generating the bindings for a header which includes "fcntl.h", the following struct is encountered:
Rust-bindgen generates the following code:
Some of these assertions are failing:
::std::mem::size_of::<log2phys>()is 24 bytes, not 20.::std::mem::align_of::<log2phys>()is 8 bytes, not 4.unsafe { & ( * ( 0 as * const log2phys ) ) . l2p_contigbytes as * const _ as usize }is 8 bytes not 4.unsafe { & ( * ( 0 as * const log2phys ) ) . l2p_devoffset as * const _ as usize }is 16 bytes not 12.I don't know how helpful this is, but please let me know if I can supply more information.
Thanks for making bindgen!
mark-buer commentedon May 30, 2017
I'm running into this same issue for osx/ios.
Some of the system provided headers use
packquite heavily.Running:
Gives the count of
#pragma packper header (with more than 1 occurrence):And there are many more headers with a single
#pragma packoccurrence (not listed here to avoid excessive noise).fitzgen commentedon Oct 11, 2017
AFAICT, @emilo is right, and
libclangdoesn't even give us access to#pragma pack.For this header
we get this
libclangAST (note theUnexposedAttr):fitzgen commentedon Oct 11, 2017
Another version of this issue:
LegNeato commentedon Oct 19, 2017
Hmmm, reading http://llvm.org/viewvc/llvm-project?view=revision&revision=191345 (and the patch in http://lists.llvm.org/pipermail/cfe-commits/Week-of-Mon-20130923/089367.html) I would assume that
packedshould indeed be exposed by libclang?It appears
clang-sysat least has the constant: https://github.com/KyleMayes/clang-sys/blob/da6280de7ef5f5e7dde7e63ea00fb6df26eda601/src/lib.rs#L446emilio commentedon Oct 19, 2017
Detect `#pragma pack(...)` and make `pack(n)` where `n > 1` opaque
fitzgen commentedon Nov 1, 2017
I have a band-aid in #1136
It should make
#pragma packed(1)work.For
#pragma pack(n)wheren > 1, it detects that something is off and makes the type opaque. This is the best that we can do since Rust doesn't have#[repr(packed = "n")]yet.[-]Bindgen doesn't know about #pragma pack(1)[/-][+]Bindgen doesn't know about `#pragma pack(...)`[/+][-]Bindgen doesn't know about `#pragma pack(...)`[/-][+]Bindgen can't translate `#pragma pack(n)` into `#[repr(packed = "n")]`[/+]Detect `#pragma pack(...)` and make `pack(n)` where `n > 1` opaque
Auto merge of #1136 - fitzgen:pragma-pack-bandaid, r=emilio
gnzlbg commentedon Feb 26, 2018
The PR for
packedis there rust-lang/rust#48528It would be great if
bindgencould start mapping#pragma pack(N)and thepackedC attributes to#[repr(packed(N))]behind some sort of opt-in feature - the produced bindings are necessarily going to require nightly Rust untilrepr(packed)is stabilized.Given that getting struct packing wrong is essentially undefined behavior, I'd wish that once the PR is merged it can be quickly stabilized.
gnzlbg commentedon Apr 13, 2018
@fitzgen
repr(packed(N))just landed this week, would it be possible for bindgen to use it instead of making the struct opaque?fitzgen commentedon Apr 16, 2018
We can start doing that when the rust target = nightly now. I don't have time to write a PR myself, but I could review one.
Support #[repr(packed(N))] on Rust 1.33+
Support #[repr(packed(N))] on Rust 1.33+
Support #[repr(packed(N))] on Rust 1.33+