GeezerVison

Summary

A customer's network blocks the BrowserStackLocal binary download, so they asked for a way to point the CLI at a binary they had already placed on disk. browserstack-local has supported exactly that for a long time — a binarypath option that short-circuits LocalBinary entirely — but the CLI never forwarded one.

setLocalArgs forwards five connection_settings keys today (local_identifier, local_config_file, proxyHost, proxyPort, useCaCertificate). binaryPath appears nowhere in bin/, so setting it in browserstack.json is silently dropped and the download path is always taken. That silent drop is what made this hard to diagnose from the outside: there is no log line saying the key was ignored.

Reported and root-caused on SDK-7284.

What lands here

The forwarded key is deliberately all-lowercase. browserstack-local only special-cases binarypath in addArgs; any other casing falls through to its default: branch and is handed to the binary as an unknown CLI flag — while the download still happens:

binaryPath  -> this.binaryPath = undefined | leaks to the binary as ["--binaryPath","<path>"]
binarypath  -> this.binaryPath = <path>    | leaks: []
BinaryPath  -> this.binaryPath = undefined | leaks to the binary as ["--BinaryPath","<path>"]

Accepting both spellings in browserstack.json is for the same reason — a user copying binarypath from the browserstack-local README should not land back in the silent-drop case this PR exists to remove.

Verification

Against the published [email protected] with its real [email protected], using a stub binary that reports {"state":"connected"} so the assertion is which binary got spawned:

On the fixed run browserstack-local prints BINARY PATH IS DEFINED and start returns no error. The argv it hands the binary matches the shape in the customer's error line, including the duplicate --daemon / --source the two layers each add.

Test plan

• Three new specs on setLocalArgs (forwarded key, lowercase alias, absent key).
• test/unit/bin/helpers/utils.js: 398 passing / 3 failing on this branch vs 393 passing / 5 failing on a clean master. Net +5 passing, zero regressions. The 3 remaining failures are pre-existing getVideoConfig specs, red on master too.
• End-to-end check against the published tarball plus a stub binary (table above).

Why two pre-existing failures go green

bin/helpers/usageReporting.js requires ./utils (line 9) and then reassigns module.exports (line 21). Requiring usageReporting before utils therefore leaves utils holding the empty exports object captured mid-cycle, while the test holds the real one — so usageReporting.cli_version_and_path was undefined inside setLocalArgs, nothing stubbed in the test was visible to the code under test, and the existing spec threw. Requiring utils first makes it capture the finished exports.

This is a test-harness fix only; no production code depends on the ordering, since the CLI's entry points already load utils first. The stale-reference hazard in usageReporting.js itself is left alone as out of scope for this ticket.