[
https://issues.apache.org/jira/browse/SPARK-57488?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Max Gekk resolved SPARK-57488.
------------------------------
Resolution: Won't Fix
> checkout-pr.sh review helper aborts with bash "unbound variable" on the
> no-force checkout path under bash 3.2
> -------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-57488
> URL: https://issues.apache.org/jira/browse/SPARK-57488
> Project: Spark
> Issue Type: Bug
> Components: Project Infra
> Affects Versions: 5.0.0
> Reporter: Max Gekk
> Priority: Major
>
> h3. Summary
> The {{checkout-pr.sh}} helper used by the spark-dev PR-review tooling aborts
> with {{force_args[@]: unbound variable}} whenever it takes the no-force
> checkout path. That path is *always* taken for self-PRs (and for any external
> PR before a force is needed), so self-PR review is consistently broken on
> macOS. The caller misreports the crash as a {{branch-diverged}} stop (exit
> 11), so a clean branch that is already at the PR head looks as if it has
> diverged.
> h3. Environment
> * macOS, where {{/usr/bin/env bash}} resolves to {{/bin/bash}} = GNU bash
> *3.2.57* (Apple's stock version).
> * Triggered via the PR setup/checkout step of the review tooling.
> h3. Steps to reproduce
> # On macOS (bash 3.2), check out your own open PR locally so the branch is
> already at the PR head.
> # Run the review checkout step against that PR.
> h3. Expected
> Checkout is a no-op (branch already at head); setup exits 0 and proceeds.
> h3. Actual
> {noformat}
> checkout-pr.sh: line 63: force_args[@]: unbound variable
> branch-diverged: local branch <name> could not be fast-forwarded to the PR
> head ...
> {noformat}
> Setup exits 11 (branch-diverged), halting the review even though the working
> tree is clean and HEAD exactly matches the PR head_sha.
> h3. Root cause
> The helper builds an optional force flag in an array:
> {code:bash}
> local force_args=()
> [[ "$1" == "1" ]] && force_args=(-f)
> gh pr checkout ... "${force_args[@]}"
> {code}
> Under {{set -euo pipefail}}, *bash 3.2 treats expansion of an empty array*
> ({{"${force_args[@]}"}}) *as an unbound variable* and aborts -- a bug fixed
> in bash 4.4+. On the no-force path the array is empty, so the command never
> runs.
> h3. Fix
> Use the {{set -u}}-safe expansion idiom, which yields nothing when the array
> is empty and the quoted elements otherwise:
> {code:bash}
> gh pr checkout "${PR_NUMBER}" -R "${OWNER}/${REPO}"
> ${force_args[@]+"${force_args[@]}"} 2>&1
> {code}
> h3. Test gap
> The script's self-test stubs out the checkout function entirely, so it never
> exercises the real argv assembly -- which is why this shipped green. A
> regression test should stub the {{gh}} binary instead of the function so the
> empty-array expansion runs under {{set -u}}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
