fix(eslint): skip autofix on ${} in string literal#8627
Conversation
For literal operands, `buildAutofixMessage` embeds the source text
verbatim into the synthesised template's static segments. When that
text contains `${...}`, it turns into a real interpolation —
silently changing the message, or throwing `ReferenceError` if the
identifier isn't in scope before `assert.ok` even runs.
Bail out (return null) in that case rather than try to escape, and
drop the now-obsolete comment about backslashes — the broader
literal check supersedes it. Includes regression tests for `==`,
`>`, and the escaped `\${...}` variant.
Follow-up to #8537.
|
Overall package sizeSelf size: 5.86 MB Dependency sizes| name | version | self size | total size | |------|---------|-----------|------------| | import-in-the-middle | 3.0.1 | 82.56 kB | 817.39 kB | | opentracing | 0.14.7 | 194.81 kB | 194.81 kB | | dc-polyfill | 0.1.11 | 25.74 kB | 25.74 kB |🤖 This report was automatically generated by heaviest-objects-in-the-universe |
BenchmarksBenchmark execution time: 2026-05-25 08:16:40 Comparing candidate commit 57da6fa in PR branch Found 0 performance improvements and 0 performance regressions! Performance is the same for 1489 metrics, 104 unstable metrics. |
BridgeAR
left a comment
BridgeAR
left a comment
There was a problem hiding this comment.
LGTM
We should likely limit the rule quite a bit to where it applies. I deactivated it for now.
2 participants
What does this PR do?
Fixes a bug in the
eslint-require-boolean-assert-messageautofixer (introduced in #8537): when a boolean assertion has a string-literal operand that contains${...}, the autofixer silently produces a broken message — or, worse, code that throwsReferenceErrorbefore the assertion runs.buildAutofixMessageembeds the source text ofLiteraloperands verbatim into the static segments of the synthesised template literal (non-literal operands are wrapped in${...}and are expression text, so they're unaffected). For example:was being autofixed to:
…where the literal's
${expected}becomes a real interpolation against whateverexpectedhappens to mean in scope. This PR adds an early bail-out (return null) when aLiteraloperand's source text contains${, leaving the violation flagged but unfixed so the author can write the message themselves. The companion comment about backslashes — now subsumed by the broader literal-level check — is dropped.Three regression tests cover the cases that motivated the fix:
==with an interpolation-shaped RHS,>with the same, and the escaped\${...}variant (the escape is normalised away by the parser, so the synthesised template still sees${).Motivation
The original autofixer in #8537 protected against one hazard — a literal backtick in an operand, which would terminate the synthesised template early — but missed the symmetric case of
${...}syntax appearing inside a string literal. The window is narrow but real: any test asserting against a string that happens to contain template-literal syntax (error messages, snapshots, codegen output, fixture strings) would hit it on the first--fix.Additional Notes
#8620 proposes deactivating
eslint-require-boolean-assert-messageoutright and reworking it before re-enabling. This PR is independent of that decision: the autofixer source still ships either way, the bug is real, and the fix is part of the hardening #8620 references. Happy to sequence behind it if reviewers prefer.