Commit 2aaa62b89d22b570e560731b03e3d2d6f5c3bbce | rust

author	bors <bors@rust-lang.org> 2025-10-24 18:23:10 UTC
committer	bors <bors@rust-lang.org> 2025-10-24 18:23:10 UTC
log	2aaa62b89d22b570e560731b03e3d2d6f5c3bbce
tree	d3e063725ce11b106f3efea2528d2cebe9776013
parent	ab925646fae038b02bd462cd328ae9eef1639236
parent	c6acffeb785e3efa2909f5ec4a4452cb38ddde17

Auto merge of #148018 - tardyp:lto_big_filesize_utf8, r=JonathanBrouwer

fix panic when rustc tries to reduce intermediate filenames len with utf8

The issue cannot be reproduced with the former testcase of creating external crates because rust refuses to use "external crate 28_找出字符串中第一个匹配项的下标" because it is not a valid indentifier (starts with number, and contain non ascii chars)

But still using 28_找出字符串中第一个匹配项的下标.rs as a filename is accepted by previous rustc releases So we consider it valid, and add an integration test for it to catch any regression on other code related to non ascii filenames.

Fix rust-lang/rust#147975

3 files changed, 37 insertions(+), 5 deletions(-)

compiler/rustc_session/src/config.rs+12-3

...	...	@@ -1211,13 +1211,22 @@ fn maybe_strip_file_name(mut path: PathBuf) -> PathBuf {
1211	1211	if path.file_name().map_or(0, \|name\| name.len()) > MAX_FILENAME_LENGTH {
1212	1212	let filename = path.file_name().unwrap().to_string_lossy();
1213	1213	let hash_len = 64 / 4; // Hash64 is 64 bits encoded in hex
1214		let stripped_len = filename.len() - MAX_FILENAME_LENGTH + hash_len;
	1214	let hyphen_len = 1; // the '-' we insert between hash and suffix
	1215
	1216	// number of bytes of suffix we can keep so that "hash-<suffix>" fits
	1217	let allowed_suffix = MAX_FILENAME_LENGTH.saturating_sub(hash_len + hyphen_len);
	1218
	1219	// number of bytes to remove from the start
	1220	let stripped_bytes = filename.len().saturating_sub(allowed_suffix);
	1221
	1222	// ensure we don't cut in a middle of a char
	1223	let split_at = filename.ceil_char_boundary(stripped_bytes);
1215	1224
1216	1225	let mut hasher = StableHasher::new();
1217		filename[..stripped_len].hash(&mut hasher);
	1226	filename[..split_at].hash(&mut hasher);
1218	1227	let hash = hasher.finish::<Hash64>();
1219	1228
1220		path.set_file_name(format!("{:x}-{}", hash, &filename[stripped_len..]));
	1229	path.set_file_name(format!("{:x}-{}", hash, &filename[split_at..]));
1221	1230	}
1222	1231	path
1223	1232	}

tests/run-make/lto-long-filenames/rmake.rs-2

...	...	@@ -9,8 +9,6 @@
9	9
10	10	//@ ignore-cross-compile
11	11
12		use std::fs;
13
14	12	use run_make_support::{rfs, rustc};
15	13
16	14	// This test make sure we don't get such following error:

tests/run-make/lto-long-filenames_cn/rmake.rs created+25

...	...	@@ -0,0 +1,25 @@
	1	//@ ignore-cross-compile
	2	// gnu ld is confused with intermediate files having multibytes characters in their names:
	3	// = note: ld.exe: cannot find f0d5ff18d6510ebc-???_???_??????????_?_?????_?_???????.d50c2 \
	4	// 4c0c4ea93cc-cgu.0.rcgu.o: Invalid argument
	5	// as this is not something rustc can fix by itself,
	6	// we just skip the test on windows-gnu for now. Hence:
	7	//@ ignore-windows-gnu
	8
	9	use run_make_support::{rfs, rustc};
	10
	11	// This test make sure we don't crash when lto creates output files with long names.
	12	// cn characters can be multi-byte and thus trigger the long filename reduction code more easily.
	13	// we need to make sure that the code is properly generating names at char boundaries.
	14	// as reported in issue #147975
	15	fn main() {
	16	let lto_flags = ["-Clto", "-Clto=yes", "-Clto=off", "-Clto=thin", "-Clto=fat"];
	17	for prefix_len in 0..4 {
	18	let prefix: String = std::iter::repeat("_").take(prefix_len).collect();
	19	let main_file = format!("{}ⵅⴻⵎⵎⴻⵎ_ⴷⵉⵎⴰ_ⵖⴻⴼ_ⵢⵉⵙⴻⴽⴽⵉⵍⴻⵏ_ⵏ_ⵡⴰⵟⴰⵙ_ⵏ_ⵢⵉⴱⵢⵜⴻⵏ.rs", prefix);
	20	rfs::write(&main_file, "fn main() {}\n");
	21	for flag in lto_flags {
	22	rustc().input(&main_file).arg(flag).run();
	23	}
	24	}
	25	}