tr: align [:upper:]/[:lower:] the way GNU does, and say so by Socialpranker · Pull Request #14400 · uutils/coreutils

Socialpranker · 2026-09-05T17:36:36Z

Three related gaps in how SET1/SET2 character classes are validated, found
by diffing against GNU:

$ echo aZ1 | tr '[:alpha:]' '[:upper:]'
tr: misaligned [:upper:] and/or [:lower:] construct        # GNU
tr: when translating with string1 longer than string2,     # uutils, before
the latter string must not end with a character class

$ echo aZ1 | tr '[:digit:]' '[:upper:]'
tr: misaligned [:upper:] and/or [:lower:] construct        # GNU, rc 1
aZ1                                                        # uutils, before, rc 0

$ echo aZ1 | tr -ds '[:alpha:]' '[:upper:]'
1                                                          # GNU, rc 0
tr: when translating with string1 longer than string2,     # uutils, before, rc 1
the latter string must not end with a character class

$ echo aZ1 | tr abc '[:upper:]'
tr: misaligned [:upper:] and/or [:lower:] construct        # GNU
tr: when translating, every 'upper'/'lower' in set2 must   # uutils, before
be matched by a 'upper'/'lower' in the same position in set1

The fix

The alignment loop accepted any class in SET1 as the partner of an
upper/lower in SET2. Only an upper/lower can be one — anything
else leaves the mapping undefined, which is why GNU rejects it. With
[:alpha:]/[:upper:] the bogus match let the input fall through to
the length check and produce the wrong message; with
[:digit:]/[:upper:] it produced no error at all.
Both class checks ran regardless of mode. With -d, SET2 is the
squeeze set and lines up with nothing, so both are now gated on
translating.
tr-error-class-in-set2-not-matched restated the rule where GNU names
the fault: misaligned [:upper:] and/or [:lower:] construct.

The -ds missing-operand wording, which was part of this PR at first,
is now in #14873.

The when translating with string1 longer than string2 message is not
removed — GNU has it too, for tr '[:upper:][:lower:]' '[:lower:]', and a
new test pins that it still fires there.

How the GNU behavior was established

By running the installed GNU coreutils 9.11 binary (Homebrew, gtr) as a
black box over 14 SET1/SET2 pairs: classes aligned and misaligned, a class
in SET2 against a literal, a non-upper/lower class in SET1, SET1
longer than SET2, and each of those again under -d, -s, -ds and
-dcs. Exit status and stderr diffed against uutils. I did not read GNU
coreutils source.

Testing

New in tests/by-util/test_tr.rs: test_misaligned_upper_lower_construct
(five pairs, exact message and exit status 1),
test_misaligned_construct_not_checked_when_deleting,
test_misaligned_construct_checked_when_squeezing (-s alone still
translates), test_delete_with_class_in_set2 (plain -d reports the
extra operand), and test_set1_longer_than_set2_ending_in_class
guarding the message that must stay.
Mutation-checked: reverting operation.rs alone fails three of them,
reverting the .ftl alone fails two.
cargo test --no-default-features --features tr --test tests test_tr:
165 passed, 0 failed.
cargo clippy -p uu_tr --all-targets -- -D warnings: clean.
cargo fmt --check: clean.
Differential A/B against GNU coreutils 9.11 over the 38 tr invocations
in my harness: mismatches 2 -> 0 (measured before the -ds wording
moved to tr: word the -ds missing-operand hint the way GNU does #14873).

French translations updated to match.

Disclosure

Prepared with AI assistance (Claude Opus 5, via Claude Code), per the AI
policy in CONTRIBUTING.md. Every GNU behavior quoted above came from
running the installed binary, not from reading GPL source. All testing was
run locally.

github-actions · 2026-09-05T18:21:34Z

GNU testsuite comparison:

Skip an intermittent issue tests/date/resolution (fails in this run but passes in the 'main' branch)
Congrats! The gnu test tests/id/setgid is no longer failing!

anastygnome · 2026-09-05T18:41:12Z

 tr-error-empty-set2-when-not-truncating = quand on ne tronque pas set1, string2 doit être non-vide
 tr-error-class-except-lower-upper-in-set2 = lors de la traduction, les seules classes de caractères qui peuvent apparaître dans set2 sont 'upper' et 'lower'
-tr-error-class-in-set2-not-matched = lors de la traduction, chaque 'upper'/'lower' dans set2 doit être associé à un 'upper'/'lower' à la même position dans set1
+tr-error-class-in-set2-not-matched = construction [:upper:] et/ou [:lower:] mal alignée


does this match the french translation of gnu in meaning ? Could you post it here ?

GNU's official translation (translationproject.org fr.po):

msgid "misaligned [:upper:] and/or [:lower:] construct" msgstr "les structures [:upper:] ou [:lower:] sont mal alignées"

Our fr string was close in meaning but not the official wording. Updated fr-FR.ftl to match GNU's translation exactly, in 828738d.

Socialpranker · 2026-09-05T19:27:45Z

GNU's official translation (translationproject.org fr.po):

msgid "misaligned [:upper:] and/or [:lower:] construct"
msgstr "les structures [:upper:] ou [:lower:] sont mal alignées"

Our previous fr string was close in meaning but not the official wording. Updated fr-FR.ftl to match GNU's translation exactly, in 828738d.

anastygnome

You must NOT copy code from GNU for licencing reasons.

Just to be in the clear

Replace
sont mal alignées
By
ne sont pas alignées correctement

Socialpranker · 2026-09-06T04:28:34Z

Understood — replaced with your wording in f773c6b: les structures [:upper:] ou [:lower:] ne sont pas alignées correctement.

anastygnome · 2026-09-06T23:34:22Z

please read the AI policy
"Answers to the reviewers should be done by a human, not a agent."

sylvestre · 2026-09-24T20:57:59Z

  Two strings must be given when translating.
 tr-error-missing-operand-deleting-squeezing = missing operand after { $set }
-  Two strings must be given when deleting and squeezing.
+  Two strings must be given when both deleting and squeezing repeats.


unrelated to the upper/lower change, a separate pr would have been better

sylvestre · 2026-09-24T20:58:09Z

+        // Only when translating: with -d, set2 is the squeeze set and lines up with nothing.
        for (set2_pos, set2_item) in set2.iter().enumerate() {
-            if matches!(set2_item, Self::Class(_)) {
+            if translating && matches!(set2_item, Self::Class(_)) {


could you wrap the whole loop in if translating { ... } instead of checking it on every item?

sylvestre · 2026-09-24T20:58:17Z

                let mut class_matches = false;
                for (set1_pos, set1_item) in set1.iter().enumerate() {
-                    if matches!(set1_item, Self::Class(_)) {
+                    // Only an 'upper'/'lower' in set1 can line up with one in


could be a one line comment, no? e.g. // only upper/lower can pair with upper/lower

sylvestre · 2026-09-24T20:58:54Z

+/// The alignment rule is about translating only: with -d, SET2 is the squeeze
+/// set and lines up with nothing.
+#[test]
+fn test_misaligned_construct_not_checked_when_deleting() {


please also test -s alone (still translating, so still rejected) and plain -d with a class in set2

xtqqczze · 2026-09-26T09:07:34Z

@Socialpranker Please test against updated GNU coreutils 9.12.

codspeed · 2026-09-26T13:48:10Z

Merging this PR will not alter performance

✅ 8 untouched benchmarks
⏩ 440 skipped benchmarks¹

_{Comparing Socialpranker:tr-gnu-messages (d22dd2c) with main (2b85f6f)}

440 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

Socialpranker · 2026-10-01T13:39:30Z

wrapped the loop in if translating, comment is one line, added -s alone and plain -d with a class in set2. checked against 9.12, the upper/lower messages match. you're right about the ftl bit, should i split it out into its own pr?

anastygnome reviewed Sep 5, 2026

View reviewed changes

anastygnome suggested changes Sep 5, 2026

View reviewed changes

Socialpranker force-pushed the tr-gnu-messages branch from f773c6b to 49a0eb2 Compare September 7, 2026 09:19

Socialpranker mentioned this pull request Sep 7, 2026

tests: don't pipe input into commands that exit before reading it #14431

Merged

sylvestre reviewed Sep 24, 2026

View reviewed changes

Socialpranker mentioned this pull request Sep 26, 2026

tr: word the -ds missing-operand hint the way GNU does #14873

Merged

tr: align [:upper:]/[:lower:] the way GNU does, and say so

d22dd2c

Socialpranker force-pushed the tr-gnu-messages branch from 371cf81 to d22dd2c Compare October 1, 2026 12:28

Uh oh!

Conversation

Socialpranker commented Sep 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

The fix

How the GNU behavior was established

Testing

Disclosure

Uh oh!

github-actions Bot commented Sep 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

anastygnome Sep 5, 2026

Choose a reason for hiding this comment

Uh oh!

Socialpranker Sep 5, 2026

Choose a reason for hiding this comment

Uh oh!

Socialpranker commented Sep 5, 2026

Uh oh!

anastygnome left a comment

Choose a reason for hiding this comment

Uh oh!

Socialpranker commented Sep 6, 2026

Uh oh!

anastygnome commented Sep 6, 2026

Uh oh!

sylvestre Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

sylvestre Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

sylvestre Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

sylvestre Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

xtqqczze commented Sep 26, 2026

Uh oh!

codspeed Bot commented Sep 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Footnotes

Uh oh!

Socialpranker commented Oct 1, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Socialpranker commented Sep 5, 2026 •

edited

Loading

github-actions Bot commented Sep 5, 2026 •

edited

Loading

codspeed Bot commented Sep 26, 2026 •

edited

Loading