test: add unit tests for version-check comparison and skip logic

[Montana]

Hey Sourcegraph,

This change adds cmd/src/version_check_test.go, covering both pieces of thenew logic with table-driven unit tests. The functions under test are pure and have no network or filesystem dependencies, so the suite runs fast and
deterministically.

isOutdated

Verifies the version comparison that decides whether a warning should fire:

• Major / minor / patch comparison — confirms an older patch (4.3.0 vs
  4.3.1), older minor (4.3.0 vs 4.4.0), and older major (3.43.0 vs
  4.0.0) are each reported as outdated, while equal (4.4.0 vs 4.4.0) and
  newer (4.5.0 vs 4.4.0) versions are not.
• v-prefixes — checks that a leading v (v4.3.0) parses correctly and
  compares the same as its bare form, so the result doesn't depend on how the
  version string is formatted.
• Prereleases — ensures comparison is limited to the major/minor/patch
  components: a prerelease on the same base (4.4.0-rc.1 vs 4.4.0) is not
  flagged as outdated, while a prerelease on an older base (4.3.0-rc.1 vs
  4.4.0) still is. This guards against spurious warnings for users running
  release candidates.
• Unparseable input — confirms the function returns ok = false (rather
  than guessing) when the current version can't be parsed (dev) or the
  recommended version is empty, so callers know to stay silent instead of
  warning on bad data.

shouldSkipVersionCheck

Verifies the guard that decides when to run the check at all:

• No-op subcommands — confirms the check is skipped for version, help,
  an empty subcommand, and flag-only invocations like -h / --help, where a
  warning would be noise (and where version already reports this information
  itself).
• Normal subcommands — confirms the check does run for representative
  real commands (search, batch, api), so the feature isn't accidentally
  disabled.
• SRC_SKIP_VERSION_CHECK opt-out — confirms setting the environment
  variable suppresses the check, covering the documented escape hatch for CI
  and scripts.
• Dev builds — confirms the check is skipped when version.BuildTag is the
  default dev value, so local and development builds don't warn on every
  invocation.

Because shouldSkipVersionCheck reads the package-level version.BuildTag, the test temporarily overrides it to a real release value (4.3.0) and restores it via t.Cleanup, keeping the cases isolated from the build-time version.

[burmudar]

@Montana This test should be part of #1338. Please update that PR as I am closing this.

[Montana]

Hi @burmudar,

You're right. I'd added the comparison and skip helpers but never invoked them, so the check was dead code. Fixed by calling maybeWarnVersion(os.Args[1:]) from main() before maybeRunMigratedCommand(). It reuses the existing getRecommendedVersion against .api/src-cli/version instead of duplicating the request.

Behavior matches the PR description: fail-open on any error, 3s timeout, stderr only so --json is untouched, and the skip rules (dev build via version.BuildTag == version.DefaultBuildTag, SRC_SKIP_VERSION_CHECK, and version/help/no-arg) are all applied at the call site. Comparison is major/minor/patch only, so prereleases don't warn. I've folded your #1339 tests in here so the comparison and skip logic are covered.

One thing to flag, because this runs before flag parsing, -endpoint/-config passed as flags aren't picked up by the early config read — it resolves the endpoint from SRC_ENDPOINT, the config file, or the default. Happy to move the call after config resolution if you'd rather flags be honored; I kept it simple to preserve the fail-open guarantees, but your call.

I've reopened another 2 PR's that should address this, you can fold them together if you want.

Cheers,
Michael