`-Zchecksum-hash-algorithm` used normalized file size in dep-info

[weihanglo]

Problem

-Zchecksum-hash-algorithm writes magic comments to dep-info files to info build tool like Cargo about source file checksum and file length

# checksum:{checksum_hash} file_len:{file_len} {path}

rust/compiler/rustc_interface/src/passes.rs

Lines 747 to 758 in d6deffe

	// If caller requested this information, add special comments about source file checksums.
	// These are not necessarily the same checksums as was used in the debug files.
	if sess.opts.unstable_opts.checksum_hash_algorithm().is_some() {
	files
	.iter()
	.filter_map(|(path, file_len, hash_algo)| {
	hash_algo.map(|hash_algo| (path, file_len, hash_algo))
	})
	.try_for_each(|(path, file_len, checksum_hash)| {
	writeln!(file, "# checksum:{checksum_hash} file_len:{file_len} {path}")
	})?;
	}

However, the above code uses SourceFile.source_len from SourceMap that the length of the file is normalized length (BOM-removed, newline normalized). That makes external tools harder to verify the checksum, as they need to follow all the normalizations rustc did to compute exact the same checksum

Possible solution

Instead of using normalized file length, SourceFile should keep the original file length, and serialize it to dep-info file correctly.

Meta

rustc 1.93.0-nightly (bd3ac0330 2025-11-01)

6d41834

Cargo issue: rust-lang/cargo#16253

[weihanglo]

I was thinking about adding an original_source_len field to SourceFile, but I am not familiar with the entire hash/encode around it.

[saethlin]

I would just hash/encode however source_len does. Does that not work?