Debug-format fat pointers with their metadata for better insight

[vojtechkral]

New iteration of #84927 (since I can no longer reopen that one).

Basically, add metadata via ptr::metadata() to debug prints of fat pointers (ie. to slices and trait objects).
Thin ptr debug format as well as all {:p} should be unadffected.

Example prints:

(0x56486780e7e3, 5)                            // pointer to a slice
0x7f0abc000d20, DynMetadata(0x564ef9f9e938))   // pointer to a trait object

This time I managed to leverage specialization instead of size_of shennanigans.

Blocked on #81513

[rust-highfive]

r? @m-ou-se

(rust-highfive has picked a reviewer for you, use r? to override)

[vojtechkral]

Man, I get the weirdest compiler errors...

[m-ou-se]

Looks good to me. Just two small nits:

[rust-log-analyzer]

The job x86_64-gnu-llvm-12 failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

   Compiling rustc_mir_build v0.0.0 (/checkout/compiler/rustc_mir_build)
   Compiling rustc_plugin_impl v0.0.0 (/checkout/compiler/rustc_plugin_impl)
   Compiling rustc_typeck v0.0.0 (/checkout/compiler/rustc_typeck)
   Compiling rustc_borrowck v0.0.0 (/checkout/compiler/rustc_borrowck)
error: internal compiler error: compiler/rustc_codegen_llvm/src/context.rs:924:21: `fn_abi_of_instance(std::ptr::metadata::<chalk_ir::GenericArg<ChalkRustInterner>>, [])` failed: the type `<chalk_ir::GenericArg<ChalkRustInterner> as std::ptr::Pointee>::Metadata` has an unknown layout
  --> /checkout/library/core/src/ptr/metadata.rs:93:1
   |
93 | pub const fn metadata<T: ?Sized>(ptr: *const T) -> <T as Pointee>::Metadata {

thread 'rustc' panicked at 'Box<dyn Any>', /rustc/28c8a34e18fc05277c81328d1bbf5ed931f4d22e/compiler/rustc_errors/src/lib.rs:1115:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


note: the compiler unexpectedly panicked. this is a bug.

note: we would appreciate a bug report: https://github.com/rust-lang/rust/issues/new?labels=C-bug%2C+I-ICE%2C+T-compiler&template=ice.md

note: rustc 1.59.0-beta.5 (28c8a34e1 2022-01-27) running on x86_64-unknown-linux-gnu

note: compiler flags: -Z macro-backtrace -Z tls-model=initial-exec -Z unstable-options -Z binary-dep-depinfo -Z force-unstable-if-unmarked -C opt-level=3 -C embed-bitcode=no -C debuginfo=0 -C debug-assertions=on -C symbol-mangling-version=v0 -C link-args=-Wl,-z,origin -C link-args=-Wl,-rpath,$ORIGIN/../lib -C prefer-dynamic -C llvm-args=-import-instr-limit=10 --crate-type lib
note: some of the compiler flags provided by cargo are hidden

query stack during panic:
end of query stack

[oli-obk]

r? @oli-obk

I'll look at the failure tomorrow

[eddyb]

The bug here was fixed by #92248, you should be able to use #[cfg(not(bootstrap))] on your additions to avoid applying them to the beta compiler that doesn't have the fix yet.

[oli-obk]

r? @m-ou-se

[eddyb]

(bikeshedding) I wonder how silly something like *[_] { data: 0x..., len: 123 } would be. (and have meta instead of len in the general case, I guess?)

[vojtechkral]

I'm not sure I understand, what's the *[_] part?

[eddyb]

You should be able to improve this in one of two ways:

• split the () case from the default:
  • () is "guaranteed simple" (but maybe specializing on the original pointee being Sized would be stronger)
  • the default would only be hit by custom DSTs (and unreachable today? could print some indication of metadata existing but not being understood, or just fit an unreachable!() in there)
• add a new sealed trait, and a bound for it on Pointee::Metadata (just like it requires Copy, Eq, etc. today)
  • ("sealed" as in a pub trait in a private mod, not nameable outside core - for now, anyway)
  • the new trait could just be this PR's PtrMetadataFmt for now, even if we may want to change it in the future (see below)
  • no specialization should be needed, since Pointee would be claiming it's always implemented for valid metadata types
  • whenever we get custom DSTs, it would be a nice bonus to force the user to implement such a trait, effectively making them decide how raw pointers to their custom DST should be formatted
    • this is when we'd remove the "sealing"
    • maybe it should just be a fmt::Debug bound and not let the user see the data pointer or choose how it is printed?
    • but this part is really all a bikshed for the future

[vojtechkral]

I know exactly what you mean, in fact, I've been meaning to propose in #81513 that Pointee::Metadata should maybe be its own trait (and yeah initially sealed by all means), which could enable attaching arbitrary things to it that might be needed in the future, be it debug printing or something else.

I initially did try the fmt::Debug bound approach but I think there was issues with it, altough I don't remember off the top of my head at the moment...

[vojtechkral]

I initially did try the fmt::Debug bound approach but I think there was issues with it, altough I don't remember off the top of my head at the moment...

Right, because Debug on Pointee::Metadata gets me the ability to debug-print the metadata, but in this case I need to format the whole pointer+metadata combo somehow, ideally preserving the simple "just hex number" format for simple pointers...

Edit: But I could use a custom sealed trait, sure, I'll try that later and wil update...

[vojtechkral]

FYI I have paused work on this due to stuff not being resolved in #81513

[bors]

☔ The latest upstream changes (presumably #125443) made this pull request unmergeable. Please resolve the merge conflicts.