-
Notifications
You must be signed in to change notification settings - Fork 12.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Even nicer errors from assert_unsafe_precondition #103035
Conversation
Maybe libtest should force all captured test output to be written in case of an aborting panic? |
With a SIGABRT handler? Or do you have something more portable/graceful in mind? |
No, libtest registers a panic handler which normally only records the panic info and marks the test as failed. My suggestion is that in the case of an aborting panic it should flush all captured output including panic message before it actually aborts. |
I'm trying to implement this, and I can't seem to generate the desired effect. Are panic handlers even called in the double-panic situation we're creating? |
This shouldn't be a double panic but a panic where the PanicInfo returns false from |
This comment has been minimized.
This comment has been minimized.
6fe3754
to
0772741
Compare
Hey! It looks like you've submitted a new PR for the library teams! If this PR contains changes to any Examples of
|
r? @thomcc You asked a while ago to review such a PR |
@rustbot ready |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Oh this is much cleaner than I expected. Just one nit.
0772741
to
d831c3b
Compare
Might have gotten messy if not for @bjorn3's advice 😄 |
d831c3b
to
764b5f3
Compare
@bors r+ |
📌 Commit 764b5f3c400feb8f34ec6dd34dd9541b23590c0c has been approved by It is now in the queue for this repository. |
@bors r- |
🤷 sure if you both think that we should just stringify the expression, we can also go for that. I don't really understand the concerns but I also don't care enough to fight for the human-readable messages. Regarding the code size concerns, that's just something that was brought up when we added the first of these -- I'm not an expert for the domains where those concerns come up so all I can really do here is try to be careful. |
I just remembered that I have a conflict with the libs meeting, so I'm writing a lot more here. This is not so much directed at you, as it is my effort to make sure these points come up at/before a meeting. The best user experience here involves displaying the invalid values alongside the location of the bug in the user's code, as we do in panic messages for things like array index out of bounds. Certainly I would like to have that kind of behavior because it would let me debug the code of others directly from the error message, as I often do with Miri diagnostics. I have no idea if that is a possibility though, because of a concern about code size. But we already substantially regressed code size in #102732, the failure branch for one of these assertions is now (on x86_64): cc79: 50 push %rax
cc7a: 48 8d 3d 7f 53 04 00 lea 0x4537f(%rip),%rdi # 52000 <anon.0248cb95e8b925b4d7128f8507660a7d.0.llvm.3434341789188788945>
cc81: be 1c 00 00 00 mov $0x1c,%esi
cc86: ff 15 1c 4f 06 00 call *0x64f1c(%rip) # 71ba8 <_GLOBAL_OFFSET_TABLE_+0x2a0>
cc8c: 0f 0b ud2 The previous failure path was just a 2-byte There are other strategies for producing a static but human-readable error message here; for example we could have every invocation of Recently, I tried profiling these checks and I identified two micro-optimizations for the two hottest checks: #103287 #103285 I suspect both of those have code size implications, but exactly what they are is not at all straightforward, because LLVM can (and as best I can tell by squinting at assembly, actually does) optimize away parts of these checks, for some callers. The implications of those optimizations on code size are entirely unclear to me, for example do the code size savings there open up the possibility to spend a code size budget elsewhere? So I think if we're going to keep making decision on the basis of code size here, I want to do that with data. I know that Oxide has at least in the past compiled and run a bootloader with these debug assertions enabled, and found UB that way: rust-embedded/heapless#280 so I'm sure they care about code size at some level. And of course there are plenty of other embedded Rust users who may have even more stringent code size requirements. Are those users even enabling debug assertions? Embedded users of debug assertions may be able to at least provide an example codebase whose size they care about, and some context about what kinds of regressions would be concerning; we could keep an eye on that to at least have some idea. In terms of concrete maintenance burden, we have had #100759 which touched every use of this macro, and I expect we will have the occasional PR along the lines of #103329 to add these assertions for |
☔ The latest upstream changes (presumably #103562) made this pull request unmergeable. Please resolve the merge conflicts. |
306b613
to
987876a
Compare
This comment has been minimized.
This comment has been minimized.
987876a
to
306b3c0
Compare
Brace yourselves, I'm going to click the button on all the suggestions then squash them down. |
dca333a
to
fe6d17e
Compare
Github supports committing multiple suggestions at once. :) You have to be on the 'diff' view for that. |
I assume this is ready for review? LGTM. |
Co-authored-by: Ralf Jung <[email protected]>
e37bbfa
to
458aaa5
Compare
LGTM @bors r+ |
…n3, r=thomcc Even nicer errors from assert_unsafe_precondition For example, now running `cargo test` with this patch I get things like: ``` $ cargo +stage1 test Finished test [unoptimized + debuginfo] target(s) in 0.01s Running unittests src/lib.rs (target/debug/deps/malloc_buf-9d105ddf86862995) running 5 tests thread 'tests::test_null_buf' panicked at 'unsafe precondition violated: is_aligned_and_not_null(data) && crate::mem::size_of::<T>().saturating_mul(len) <= isize::MAX as usize', /home/ben/rust/library/core/src/slice/raw.rs:93:9 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace thread panicked while panicking. aborting. error: test failed, to rerun pass `--lib` Caused by: process didn't exit successfully: `/tmp/malloc_buf-1.0.0/target/debug/deps/malloc_buf-9d105ddf86862995` (signal: 6, SIGABRT: process abort signal) ``` This is still not perfect, but these are better for another PR: * `stringify!` is trying to do clever pretty-printing on the `expr` inside `assert_unsafe_precondition` and can even add a newline. * It would be nice to print a bit more information about where the problem is. Perhaps this is `cfg_attr(debug_assertions, track_caller)`, or perhaps it the function name added to `Location`. cc `@RalfJung` this is what I was thinking of for rust-lang#102732 (comment)
…iaskrgr Rollup of 9 pull requests Successful merges: - rust-lang#103035 (Even nicer errors from assert_unsafe_precondition) - rust-lang#103106 (Try to say that memory outside the AM is always exposed) - rust-lang#103475 (Make param index generation a bit more robust) - rust-lang#103525 (Move a wf-check into the site where the value is instantiated) - rust-lang#103564 (library: allow some unused things in Miri) - rust-lang#103586 (Process registered region obligation in `resolve_regions_with_wf_tys`) - rust-lang#103592 (rustdoc: remove redundant CSS selector `.notable-traits .notable`) - rust-lang#103593 (Remove an unused parser function (`Expr::returns`)) - rust-lang#103611 (Add test for issue 103574) Failed merges: r? `@ghost` `@rustbot` modify labels: rollup
For example, now running
cargo test
with this patch I get things like:This is still not perfect, but these are better for another PR:
stringify!
is trying to do clever pretty-printing on theexpr
insideassert_unsafe_precondition
and can even add a newline.cfg_attr(debug_assertions, track_caller)
, or perhaps it the function name added toLocation
.cc @RalfJung this is what I was thinking of for #102732 (comment)