mirror of
https://github.com/affaan-m/ECC.git
synced 2026-09-28 20:45:11 +02:00
042924e1f4
* fix(install): make install.sh robust to sh/dash invocation Two related portability fixes so `sh install.sh` behaves correctly even though the script is written for bash: - The cygpath detection used the bash-only `&>/dev/null` redirection. dash misparses `&>`, so `command -v cygpath &>/dev/null` always took the true branch and tried to run the nonexistent `cygpath` binary, failing with "cygpath: not found". Switched to the POSIX-portable `>/dev/null 2>&1` form. - Some dash builds don't support `set -o pipefail`, so `sh install.sh` can fail immediately at that line before even reaching the cygpath check (or the `[[ ... ]]` symlink-resolution logic further down). Added a guard that re-execs the script under bash when the current shell lacks bash capabilities, so the rest of the bash-only syntax always runs under a real bash regardless of the invoking shell. The guard probes for the `[[` compound command directly (via `eval '[[ 1 == 1 ]]'`) rather than trusting the $BASH_VERSION environment variable, since that variable could be inherited or spoofed under a non-bash shell and cause the guard to be skipped. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019XBaBRjaZgwhrZtYoqcQfp * test(install): cover the sh install.sh re-exec path Greptile flagged that the bash re-exec guard added in install.sh has no automated coverage, since the existing test helper only invoked the wrapper via `bash`. Adds two regression tests: - "delegates to the Node installer when invoked via a POSIX sh wrapper" — runs the script via `sh` and asserts the args/cwd still reach the Node installer correctly. - "re-execs into bash under sh even when BASH_VERSION is spoofed in the environment" — exercises the eval '[[ 1 == 1 ]]' capability probe directly, guarding against a regression back to trusting the (spoofable) $BASH_VERSION variable. CodeRabbit then pointed out that the second test used generic `sh`, which could trivially pass without exercising the re-exec branch at all if `sh` ever resolves to bash on some system. Added a findPosixOnlyShell() helper that prefers `dash` (falling back to checking `sh`, and skipping with an explicit message if neither genuinely lacks bash's `[[`), so the test reliably exercises the branch it claims to cover instead of passing vacuously. CodeRabbit then flagged that the skip path itself was miscounted as a pass (the callback returned normally, so `test()` reported success and `passed` was incremented even though nothing executed). Moved the findPosixOnlyShell() check outside the test() registration, so the test is only registered — and only counted — when a genuinely POSIX-only shell is actually available; otherwise it's excluded from both the passed and failed counts with an explicit skip line. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019XBaBRjaZgwhrZtYoqcQfp --------- Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>