fix(expansion): preserve leading IFS boundaries

[privatenumber]

Problem

Unquoted array default expansion can begin with an empty element. When it followed literal text, the word splitter dropped the resulting boundary: prefix=${values[@]-fallback} could merge prefix= and the next array element into one argument. Commands then received different arguments from Bash.

Changes

The splitter now commits preceding literal text when the next unquoted expansion begins with an IFS delimiter. The change preserves the intended argument boundary and enables the matching Oils conformance case.

Stack position: child of #366; prerequisite for the structured-extglob repair.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
just-bash-website	Ready	Preview	Aug 12, 2026 9:10am

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
just-bash	Ignored		Aug 12, 2026 9:10am

[auto-maintain]

🤖 auto-maintain review

Automated, advisory triage for @privatenumber's PR. Facts below are read from the GitHub API.

Check	Result
Author's merged PRs (this repo)	11
Account established	✅ (age 5436d · 3940 followers · 268 public repos)
Commits signed/verified	✅ 9/9
Changeset included	✅ (.changeset/leading-ifs-array-defaults.md)

Review panel: 🟡 medium highest severity

just-bash maintainer code review: 🟡 medium

The leading-boundary fix leaves the symmetric preceding-expansion boundary case broken for mixed operation words.

• packages/just-bash/src/interpreter/expansion/word-split.ts:342 — A trailing IFS boundary from the preceding expansion is discarded when the next segment is a mixed parameter-operation word: `pendingWordBreak` is never consumed before combining `split.words` and is overwritten here. For example, with `a='x '; b=y; unset v`, `set -- $a${v:-""$b}` incorrectly merges the arguments as `xy` instead of producing `x` and `y`.

General code review: 🟡 medium

The boundary fix introduces an actionable regression for mixed operation words containing an empty expansion after a trailing IFS delimiter.

• packages/just-bash/src/interpreter/expansion/word-split.ts:505 — A trailing IFS boundary is lost when followed by an empty unquoted expansion inside a mixed operation word. For example, with `x='b '`, `empty=''`, and unset `value`, `set -- prefix=${value:-$x$empty}suffix` should produce `prefix=b` and `suffix`, but this pre-flush clears `pendingWordBreak`; the empty segment does not restore it, so the helper returns `['b']` without a trailing-boundary marker and the outer splitter merges everything into `prefix=bsuffix`.

Adversarial security: 🟢 low

No actionable adversarial security issues found in the complete diff.

Adversarial security (second opinion): 🟡 medium

No security concerns: the diff is a self-contained word-splitting fidelity fix (no new deps, network, process execution, filesystem or argument-sanitization surface), and the spec-test edit only gates an OSH-specific `shopt` and asserts empty stderr; the one actionable issue is that the new IFS-boundary flush is not applied in the mixed default-value expansion branch, leaving arguments merged in that case.

• packages/just-bash/src/interpreter/expansion/word-split.ts:313 — The new pending-word-break flush was added to the plain splittable branch (line 346) and to the inner splitter (line 505), but not to the `preparedOperationWord` branch. When a preceding unquoted expansion ended with an IFS delimiter and the next segment is a mixed-quoting default expansion, the pending boundary is silently swallowed: `x='b '; unset v; set -- a$x${v:-"c"d}` yields one argument `abcd` instead of bash's two (`ab`, `cd`). Adding the same `if (pendingWordBreak && currentWord !== "")` flush before computing `flushedLeadingDelimiter` here would make the three branches consistent.

Standard Bash and host portability: 🟢 low

No Bash compatibility or ordinary-host portability issues found.

_{Posted by auto-maintain. This automated code review is advisory; a human maintainer makes the call.}

[auto-maintain]

⚠️ auto-maintain: review severity raised to 🟡 medium

A new push changed this PR and the review now contains a higher-severity finding. See the updated review comment above.

[vercel]

Additional Suggestion:

In smartWordSplitWithUnquotedLiterals, a quoted prefix followed by a splittable segment that starts with an IFS char produces a spurious empty argument (e.g. with IFS=_ yields ["q","","a"] instead of ["q","a"]).