Seeing ICE due to u32 overflow

[smklein]

I have a repo I'm attempting to profile with these instructions: https://blog.rust-lang.org/inside-rust/2020/02/25/intro-rustc-self-profile.html#profiling-the-compiler

cargo +nightly rustc --lib -- -Zself-profile -Zself-profile-events=default,args

This results in the following output:

thread 'rustc' panicked at /rust/deps/measureme-10.1.1/src/stringtable.rs:105:62:                                                                                                                                                                                                                                                                                                                                                                               
called `Option::unwrap()` on a `None`

Which appears to be the unwrap within:

measureme/measureme/src/stringtable.rs

Lines 103 to 107 in 6b8bc6d

	#[inline]
	pub fn from_addr(addr: Addr) -> StringId {
	let id = addr.0.checked_add(FIRST_REGULAR_STRING_ID).unwrap();
	StringId::new(id)
	}

I noticed that the type of StringId is a u32:

measureme/measureme/src/stringtable.rs

Line 77 in 6b8bc6d

pub struct StringId(u32);

And since we're calling from_addr, we care about that type too:

measureme/measureme/src/serialization.rs

Lines 65 to 73 in 306a7d2

	/// An address within a data stream. Each data stream has its own address space,
	/// i.e. the first piece of data written to the events stream will have
	/// `Addr(0)` and the first piece of data written to the string data stream
	/// will *also* have `Addr(0)`.
	//
	// TODO: Evaluate if it makes sense to add a type tag to `Addr` in order to
	// prevent accidental use of `Addr` values with the wrong address space.
	#[derive(Clone, Copy, Eq, PartialEq, Debug)]
	pub struct Addr(pub u32);

Why do we believe this will be limited to u32? As a small data point, the "...mm_profdata" file I was emitting looks like it's just over 4 GiB in size, which sorta aligns with the idea of a u32 overflow.

[smklein]

For reproducing:

• The repo in question: https://github.com/oxidecomputer/omicron
• To repro:

# Starting within Omicron repo...
cd nexus
cargo +nightly rustc --lib -- -Zself-profile -Zself-profile-events=default,query-keys

And for context:

$ rustc +nightly --version
rustc 1.76.0-nightly (6cf088810 2023-11-26)

[smklein]

This looks similar to:

• Option::unwrap() on a None value when building vector #196
• Option::unwrap() on a None value #165

This is consistently reproducible for me

[smklein]

Also looks like rust-lang/rust#99282

[smklein]

My attempt to patch this exists in #216

[michaelwoerister]

Let's keep this open until the change has made it into the compiler.

[michaelwoerister]

The update is in the nightly compiler now.

Thanks again for implementing the fix, @smklein!