Clarify why SGX code specifies linkage/symbol names for certain statics

[jethrogb]

Specifying linkage/symbol name is solely to ensure a single instance between the std crate and its unit tests.

Also update the symbol names as items have moved around a bit. The actual name isn't that important, it just needs to be unique. But for debugging it can be useful for it to point to the right place.

[rustbot]

r? @joboet

rustbot has assigned @joboet.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[thaliaarchi]

What is unique about SGX that requires this, unlike other platforms?

[jethrogb]

I can't speak for all platforms, but most “regular” platforms like Linux, etc. relegate this kind of once-per-process state to libc.

[thaliaarchi]

As a possible counter-example, Unix platforms store their argc/argv values in static atomics.

[jethrogb]

Linux also does some linking tricks there:

rust/library/std/src/sys/args/unix.rs

Lines 115 to 125 in 5961e5b

	/// glibc passes argc, argv, and envp to functions in .init_array, as a non-standard extension.
	/// This allows `std::env::args` to work even in a `cdylib`, as it does on macOS and Windows.
	#[cfg(all(target_os = "linux", target_env = "gnu"))]
	#[used]
	#[unsafe(link_section = ".init_array.00099")]
	static ARGV_INIT_ARRAY: extern "C" fn(
	crate::os::raw::c_int,
	*const *const u8,
	*const *const u8,
	) = {
	extern "C" fn init_wrapper(

Presumably when looking at a build of the unit tests for std, you'd find that there are two copies of init_wrapper that are both being called by the dynamic loader, so that both versions of those symbols end up having the same value.

[thaliaarchi]

No, I think ARGV_INIT_ARRAY is different. It's a list of function pointers to be sent the args and env upon initialization, as the comment says, and needs the particular link section to be recognized as such. It looks like it's doing the same thing as linkme by specifying a single element, which the linker coalesces with all others marked the same into one array. So, it's the opposite of what you're doing.

In any case, this comment change looks good.

[joboet]

Adding a note is great even if there might be better ways to achieve the desired behaviour, so:
@bors r+ rollup

[bors]

📌 Commit 8dc7732 has been approved by joboet

It is now in the queue for this repository.

[bjorn3]

A lot of places in libstd use #[cfg(not(test))] and instead import from "realstd" (which is the libstd that libtest links against rather than the libstd that is being tested) when testing to avoid duplicate lang items. Is there any reason why SGX can't do the same?

[jethrogb]

@bjorn3 you can only import public items that way. None of the items at issue here are public.