[amitmishra11]

Summary

Fixes #2627.

rtk diff <file1> <file2> reported two files as identical (exit 0) when they differed only in line-ending bytes (CRLF vs LF) — or, more generally, in anything str::lines() normalizes away (it strips \r along with \n, and treats a missing trailing newline as just "one fewer empty trailing line").

printf 'a\r\nb\r\nc\r\n' > x   # CRLF
printf 'a\nb\nc\n'       > y   # LF
rtk diff x y; echo "exit=$?"
# before: [ok] Files are identical / exit=0   (wrong — bytes differ, `diff` sees this)

Root cause

render_file_diff splits both files with content.lines() before comparing (src/cmds/git/diff_cmd.rs). str::lines() treats \r\n and \n as equivalent line terminators, so a CRLF file and an LF file with the same text collapse to identical line vectors, and compute_diff finds zero changes — the early-return then reports "identical" with no byte comparison.

Fix

In render_file_diff, when the line-based diff finds no changes, fall back to an explicit byte comparison (content1 == content2) before declaring the files identical:

• Byte-equal → still "[ok] Files are identical", exit 0 (unchanged behavior).
• Byte-different but line-identical → report the files differ, with a short description of the cause (describe_byte_only_difference: CRLF vs LF, trailing newline, or generic), exit 1 (diff convention).

Scope note

The issue also flags condense_unified_diff (this file) and compact_diff (src/cmds/git/git.rs) as sharing the same str::lines() pattern, but says those paths weren't independently verified to reproduce the bug. I scoped this PR to the confirmed, reproducible case (rtk diff <file> <file> / render_file_diff) rather than changing unverified code paths. Happy to follow up on those separately if confirmed.

Test plan

• Added regression tests in src/cmds/git/diff_cmd.rs: CRLF-vs-LF (reproduces the exact issue), trailing-newline-only diff, and a same-CRLF-on-both-sides case to confirm true identical files still short-circuit correctly. Also unit tests for the new describe_byte_only_difference helper.
• cargo test — could not run locally (no working MSVC/64-bit linker in my sandbox; see comment on fix: COPILOT_INSTRUCTIONS example drops kubectl get subcommand #2642 for details). The change is isolated to render_file_diff and a new pure helper function with no external dependencies; existing tests for that function (test_render_identical_files_exit_zero, test_render_modified_only_yaml_not_identical, etc.) are untouched and should still pass since the byte-equal fast path is preserved. Appreciate CI confirming.

[CLAassistant]

All committers have signed the CLA.

[KuSh]

Closing as a duplicate of #3519, which adds the same byte-equality guard before the "identical" verdict.

The bug is confirmed live on 0.46.0:

$ printf 'a\r\nb\r\nc\r\n' > x; printf 'a\nb\nc\n' > y
$ rtk diff x y; echo "exit=$?"
[ok] Files are identical
exit=0

#3519 is the narrower change (57 lines, one file) and has a full green test matrix; this PR only ever ran check-target, which skipped, so it was never CI-verified.

That said, describe_byte_only_difference is the better message: #3519 says "whitespace or line endings" generically, where yours names the actual cause (line endings (CRLF vs LF) / trailing newline). I've noted it for #3519 to fold in with credit to you — a user told two files differ invisibly benefits from being told how.