Add debug assertions to write_bytes and copy*

[RalfJung]

Looks like @nitnelave went MIA in #58783, so I am re-submitting their PR, tweaked just a bit. I took care to preserve commit authorship.

Cc #53871

[rust-highfive]

r? @alexcrichton

(rust_highfive has picked a reviewer for you, use r? to override)

[alexcrichton]

Seems reasonable to me! r=me with @Centril's nits

[RalfJung]

@alexcrichton I had to disable some codegen tests in debug mode, because even with optimizations the new debug assertions get in the way.

[RalfJung]

So the nits are fixed, but I'd like to get explicit confirmation that it is okay to ignore some (more) codegen tests in debug mode. We do run some CI runners in release mode, right?
(These are not the first codegen tests to be disabled in debug mode.)

[alexcrichton]

@bors: r+

[bors]

📌 Commit d3e1bf9 has been approved by alexcrichton

[RalfJung]

Building libcore with debug assertions works just fine though locally for this PR.

[mati865]

__aeabi_memcpy4 is only defined on arm and calls ptr::read: https://github.com/rust-lang-nursery/compiler-builtins/blob/39ad0538f2cfbc81b62e4f0ea1ee7a2f95133094/src/arm.rs#L144
Then ptr::read calls

rust/src/libcore/ptr/mod.rs

Line 601 in b820c76

copy_nonoverlapping(src, tmp.as_mut_ptr(), 1);

where you have added assertions.

This together with bjron3 comment should explain it.

[bjorn3]

I think adding a second -llibcompiler_builtins-*.rlib at the beginning of the linker invocation could fix this.

[RalfJung]

__aeabi_memcpy4 is only defined on arm and calls ptr::read

I see, that makes sense.

I think adding a second -llibcompiler_builtins-*.rlib at the beginning of the linker invocation could fix this.

I am afraid I have no idea how linkers are invoked in general and in Rust specifically. My current inclination is to add some cfg so that ptr::read only does these checks on x86, but that would be sad. Since you seem to have a pretty clear idea for what it takes, could you open a PR against https://github.com/RalfJung/rust/tree/debug-assert ?

[nikic]

@RalfJung Replacing ptr::write(dest, ptr::read(src)) with *dest = *src would probably do it as well.

[RalfJung]

So is there some kind of guarantee that the arguments of __aeabi_memcpy4 are 4-aligned?

[nikic]

@RalfJung Yes, that's exactly the difference between memcpy and memcpy4, the latter has guaranteed 4-byte alignment.

[alexcrichton]

I would personally prefer to avoid updating any linkage with compiler-builtins, in general compiler-builtins should never be referencing panicking symbols. I agree with @nikic that the best solution for this is probably source-level changes in the compiler-builtins crate to avoid these assertions getting embedded, even in release mode.

[RalfJung]

All right, submitted PR for compiler-builtins: rust-lang/compiler-builtins#300.

[RalfJung]

I rebased the PR and added a commit that bumps compiler_builtins to the fresh 0.1.17 release.

[alexcrichton]

@bors: r+

[bors]

📌 Commit 85d76a1 has been approved by alexcrichton