Rustc can't handle non-UTF-8 input paths, but only when RUST_BACKTRACE is not set

I ran the following python script, which runs rustc with a non-UTF-8 input file name:

import subprocess

p = subprocess.run(["rustc", bytes([255])], check=False)
print(p)

I got the following output (the empty line is not a copy-paste error):


thread 'main' (468943) panicked at /rustc/b6eeb93c992027a03e1fd120b6652546396b142c/library/std/src/env.rs:935:51:
called `Result::unwrap()` on an `Err` value: "\xFF"
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
CompletedProcess(args=['rustc', b'\xff'], returncode=101)

If I instead set RUST_BACKTRACE=1 before running the script, I get the following output (again, the empty line is not a copy-paste error. And for this one, in my terminal, the error message is bolded and has color):

error: argument 1 is not valid Unicode: "\xFF"

CompletedProcess(args=['rustc', b'\xff'], returncode=1)

Seems like whatever is trying to output the error message doesn't like it when RUST_BACKTRACE isn't set. (Also note that the returncode in the two outputs are different.)

Meta

rustc --version --verbose:

rustc 1.101.0-nightly (b6eeb93c9 2026-10-18)
binary: rustc
commit-hash: b6eeb93c992027a03e1fd120b6652546396b142c
commit-date: 2026-10-18
host: aarch64-apple-darwin
release: 1.101.0-nightly
LLVM version: 23.1.4

What on earth is this code doing? (Note the std::env::args() call.)

rust/compiler/rustc_driver_impl/src/lib.rs

Lines 1461 to 1476 in b26b4e7

    
           pub fn install_ice_hook(bug_report_url: &'static str, extra_info: fn(DiagCtxtHandle<'_>)) { 
        
               // If the user has not explicitly overridden "RUST_BACKTRACE", then produce 
        
               // full backtraces. When a compiler ICE happens, we want to gather 
        
               // as much information as possible to present in the issue opened 
        
               // by the user. Compiler developers and other rustc users can 
        
               // opt in to less-verbose backtraces by manually setting "RUST_BACKTRACE" 
        
               // (e.g. `RUST_BACKTRACE=1`) 
        
               if env::var_os("RUST_BACKTRACE").is_none() { 
        
                   // HACK: this check is extremely dumb, but we don't really need it to be smarter since this should only happen in the test suite anyway. 
        
                   let ui_testing = std::env::args().any(|arg| arg == "-Zui-testing"); 
        
                   if env!("CFG_RELEASE_CHANNEL") == "dev" && !ui_testing { 
        
                       panic::set_backtrace_style(panic::BacktraceStyle::Short); 
        
                   } else { 
        
                       panic::set_backtrace_style(panic::BacktraceStyle::Full); 
        
                   } 
        
               }

The nature of the hack aside, the only issue is that it uses args when it should be using args_os.

It also doesn't support passing -Z ui-testing (or -Zui-testing=yes, or anything else that option flags should support)

It also doesn't support passing -Z ui-testing (or -Zui-testing=yes, or anything else that option flags should support)

I'm assuming that's semi-intentional since it's an internal only flag and shouldn't be used by anyone (except the UI testing framework).

It shows up when you rustc -Zhelp, and it doesn't say it's internal. Is that the next problem?

-Z                                    ui-testing=val -- emit compiler diagnostics in a form suitable for UI testing (default: no)

(also notice that the =val form it shows up as literally doesn't work (and it shows up with spaces))

I mean the given code is specifically for the backtrace behaviour though, not ui-testing in general. The =val form should work in general since that's handled by our normal argument parsing.

Yeah, but that still sounds like buggy behavior, why would the backtrace format depend on how you passed the arg? ^^

It also depends on CFG_RELEASE_CHANNEL which is unlikely to be set by anyone who isn't building rustc from source.

But I think your position is that the hack added in #110989 shouldn't be a hack, despite the comment claiming it's "good enough"?

no my claim is that this code is super weird ^^

Navigation Menu

Repository navigation

Rustc can't handle non-UTF-8 input paths, but only when `RUST_BACKTRACE` is not set #164585

Meta

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	pub fn install_ice_hook(bug_report_url: &'static str, extra_info: fn(DiagCtxtHandle<'_>)) {
	// If the user has not explicitly overridden "RUST_BACKTRACE", then produce
	// full backtraces. When a compiler ICE happens, we want to gather
	// as much information as possible to present in the issue opened
	// by the user. Compiler developers and other rustc users can
	// opt in to less-verbose backtraces by manually setting "RUST_BACKTRACE"
	// (e.g. `RUST_BACKTRACE=1`)
	if env::var_os("RUST_BACKTRACE").is_none() {
	// HACK: this check is extremely dumb, but we don't really need it to be smarter since this should only happen in the test suite anyway.
	let ui_testing = std::env::args().any(\|arg\| arg == "-Zui-testing");
	if env!("CFG_RELEASE_CHANNEL") == "dev" && !ui_testing {
	panic::set_backtrace_style(panic::BacktraceStyle::Short);
	} else {
	panic::set_backtrace_style(panic::BacktraceStyle::Full);
	}
	}

Rustc can't handle non-UTF-8 input paths, but only when RUST_BACKTRACE is not set #164585

Description

Meta

Activity

theemathas commented on Oct 19, 2026

ChrisDenton commented on Oct 19, 2026

maxdexh commented on Oct 19, 2026

ChrisDenton commented on Oct 19, 2026

maxdexh commented on Oct 19, 2026

ChrisDenton commented on Oct 19, 2026

maxdexh commented on Oct 19, 2026

ChrisDenton commented on Oct 19, 2026

maxdexh commented on Oct 19, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Rustc can't handle non-UTF-8 input paths, but only when `RUST_BACKTRACE` is not set #164585