fix(linting): don't flag sentence-initial "Backup" before a modal by AnayGarodia · Pull Request #3794 · Automattic/harper

AnayGarodia · 2026-07-08T22:05:19Z

Disclaimer: This PR was produced with LLM assistance, reviewed and verified by me before submission (per Harper's agent-PR policy).

What

Fixes #3661. PhrasalVerbAsCompoundNoun flagged sentence-initial "Backup" and suggested "back up" in sentences like "Backup will run automatically."

Why

The rule already treats a following present/simple-past/past verb form as evidence that the compound noun is the subject (ItsFollowedByVerb), e.g. "Backup is scheduled" and "Backup has finished" are correctly left alone. But modal/auxiliary verbs weren't covered, so "Backup will run", "Backup can be restored", and "Backup should complete" were still flagged.

How

Add next_tok.kind.is_auxiliary_verb() to the existing "followed by a verb" check. Modals (will/can/should/…) are tagged as auxiliary verbs, so this closes the gap without new control flow. True positives are unaffected — they are followed by a preposition or end of clause ("I backup", "you checkout"), not by an auxiliary.

Testing

Added dont_flag_backup_before_modal_3661. Full rule suite green, no regressions:

cargo test -p harper-core --lib phrasal_verb_as_compound_noun   # 65 passed, 1 ignored
cargo fmt -- --check                                            # clean
cargo clippy -- -Dwarnings ...                                  # clean

Compatibility

Narrows one false positive; no public API or other-input behavior change.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

hippietrail

Thanks for this but there are subtle things that the coding AIs are not yet good at in relation to grammar checking.

hippietrail · 2026-07-09T02:34:56Z


-        // If the next word is a present or simple past verb form compatible with a 3rd person singular noun
+        // If the next word is a present or simple past verb form compatible with a 3rd person singular noun,
+        // or an auxiliary/modal verb ("Backup will run", "Backup can be restored"), the word is the subject noun.


There is an assumption that a modal verb can't also be used after a phrasal verb. These kinds of assumptions need to be verified. AIs are terrible at this. I use Google manually and I think they are not able to even if they hadn't already "made up their minds". Here's what I found:

The data you back up should be determined by how important it is to you

The number of files you back up can affect both the time required to run the backup and the size of the resulting backup.

But if your estate is going to attract inheritance tax, any gift you leave to Back Up will be deducted from your estate before any tax liability is calculated.

The photos you back up might capture quiet moments at home

Swimming to the bottom, touching your elbows to a fresh fuel canister, and immediately swimming back up would probably be enough to kill you.

The data you can back up may vary depending on your operating system or phone version.

These all read very naturally and give the impression that legit use of "back up" followed by a modal is common

Done in 622a87a: the modal signal no longer fires mid-sentence — it's restricted to sentence-initial position (maybe_prev_tok.is_none()), i.e. the "Backup will run" case. All six of your back-up-before-modal sentences are in the test suite as of aad85f1 and stay unflagged.

AnayGarodia · 2026-07-09T04:08:06Z

fair caution. the subtle bug you're pointing at: I put the modal check in the general ItsFollowedByVerb block, so it fired mid-sentence too — but the test only justifies the sentence-initial case. mid-sentence "will/can/might" are as often nouns as modals, so that was broader than the reasoning behind it.

tightened it: the auxiliary/modal signal now only applies with nothing before the word (maybe_prev_tok.is_none()), which is the "Backup will run" case the test covers. the existing 3rd-person/past-form checks stay as they were. still suppression-only, no corpus snapshot changes.

if there's a specific construction you had in mind that's still off, point me at it — happy to add it as a test.

hippietrail · 2026-07-09T06:59:09Z

fair caution. the subtle bug you're pointing at: I put the modal check in the general ItsFollowedByVerb block, so it fired mid-sentence too — but the test only justifies the sentence-initial case. mid-sentence "will/can/might" are as often nouns as modals, so that was broader than the reasoning behind it.

tightened it: the auxiliary/modal signal now only applies with nothing before the word (maybe_prev_tok.is_none()), which is the "Backup will run" case the test covers. the existing 3rd-person/past-form checks stay as they were. still suppression-only, no corpus snapshot changes.

Yep if you've already got something like maybe_prev_tok that's a good approach. When you don't already have it, there's the AnchorStart expression.

if there's a specific construction you had in mind that's still off, point me at it — happy to add it as a test.

I'm on the road right now so might not have the brain space to dedicate to think it through fully but I'll definitely let you know if one does pop into my head.

hippietrail

More tests would help a lot.

hippietrail · 2026-07-09T07:01:03Z

        );
    }
+
+    #[test]


I believe it was in this thread where I posted half a dozen counterexamples earlier? Those were real-word sentences found by Googling so I would add at least one of them and in this case possibly all of them as additional tests.

Done in aad85f1: all six are in dont_flag_back_up_verb_before_modal_3661, alongside still-flagged guards pinning the assumption ("Backup your files daily." -> "Back up your files daily.", "You should backup before upgrading." -> "back up").

AnayGarodia · 2026-07-09T19:15:19Z

added the tests. pulled in all six of your back-up-before-modal sentences as real-world cases — they stay unflagged, as they should.

also widened the sentence-initial check past "will" to "can"/"should", and added the bit that actually pins down the assumption: a bare "backup" with no modal after it still gets flagged ("Backup your files daily" → "Back up your files daily", "You should backup before upgrading" → "back up"). so the exception only fires sentence-initial with a modal right after, and real errors still come through.

one honest note: your six are already two words, so the rule never considers them candidates in the first place — they pass trivially. i kept them because they document that "back up" + modal is legit, but the sentence-initial and still-flagged cases are the ones that actually exercise the change. happy to drop the trivial ones if you'd rather.

PhrasalVerbAsCompoundNoun rewrote "Backup" to "back up" in sentences like "Backup will run automatically." The rule already treats a following present/past verb form as a signal that the word is the subject noun, but modal/auxiliary verbs ("will", "can", "should") were not covered, so a compound noun followed by a modal was still flagged. Add is_auxiliary_verb() to that check. Closes Automattic#3661.

…-initial an auxiliary/modal after the word (e.g. "Backup will run") signals a preceding subject noun, but only with nothing before it. mid-sentence will/can/might are just as often nouns, so gate the modal case on maybe_prev_tok.is_none() to match what the test actually claims. refs Automattic#3661.

@hippietrail

…ption verify the Automattic#3661 modal exception with real data rather than a single case: - sentence-initial "Backup" before will/can/should stays unflagged (the fix) - six real "back up" + modal sentences (from @hippietrail's review) stay unflagged, documenting that a following modal isn't itself noun evidence - a genuine "backup" verb error with no modal after it is still flagged, so the exception stays narrow. refs Automattic#3661.

AnayGarodia · 2026-07-23T18:45:33Z

@hippietrail Following up after the earlier revision: the modal guard remains limited to sentence-initial backup cases, and the reviewer-provided real-world back up + modal examples were added as regression tests. Could you please re-review?

Copilot AI review requested due to automatic review settings July 8, 2026 22:05

Copilot AI reviewed Jul 8, 2026

hippietrail requested changes Jul 9, 2026

View reviewed changes

pratyushsinghal7 and others added 3 commits July 17, 2026 12:31

AnayGarodia force-pushed the fix/backup-sentence-initial-3661 branch from 3bce11a to aad85f1 Compare July 17, 2026 19:38

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(linting): don't flag sentence-initial "Backup" before a modal - #3794

fix(linting): don't flag sentence-initial "Backup" before a modal#3794
AnayGarodia wants to merge 3 commits into
Automattic:masterfrom
AnayGarodia:fix/backup-sentence-initial-3661

AnayGarodia commented Jul 8, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

hippietrail left a comment

Uh oh!

hippietrail Jul 9, 2026

Uh oh!

AnayGarodia Jul 29, 2026

Uh oh!

AnayGarodia commented Jul 9, 2026

Uh oh!

hippietrail commented Jul 9, 2026

Uh oh!

hippietrail left a comment

Uh oh!

hippietrail Jul 9, 2026

Uh oh!

AnayGarodia Jul 29, 2026

Uh oh!

AnayGarodia commented Jul 9, 2026

Uh oh!

AnayGarodia commented Jul 23, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

                       );
                   }
+                  #[test]

Uh oh!

Conversation

AnayGarodia commented Jul 8, 2026

What

Why

How

Testing

Compatibility

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

hippietrail left a comment

Choose a reason for hiding this comment

Uh oh!

hippietrail Jul 9, 2026

Choose a reason for hiding this comment

Uh oh!

AnayGarodia Jul 29, 2026

Choose a reason for hiding this comment

Uh oh!

AnayGarodia commented Jul 9, 2026

Uh oh!

hippietrail commented Jul 9, 2026

Uh oh!

hippietrail left a comment

Choose a reason for hiding this comment

Uh oh!

hippietrail Jul 9, 2026

Choose a reason for hiding this comment

Uh oh!

AnayGarodia Jul 29, 2026

Choose a reason for hiding this comment

Uh oh!

AnayGarodia commented Jul 9, 2026

Uh oh!

AnayGarodia commented Jul 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

AnayGarodia commented Jul 23, 2026 •

edited

Loading