salt-ssh: Single.run_wfunc() never dispatches to the existing _run_wfunc_relenv() method (dead code)

[twangboy]

Description

salt.client.ssh.Single.run_wfunc() always calls self._run_wfunc_thin(), even when the deployment type is relenv:

def run_wfunc(self):
    """
    ...
    """
    return self._run_wfunc_thin()

Single._run_wfunc_relenv() already exists as a complete implementation specifically intended for this case -- its own docstring says "Execute a function using salt-call from relenv deployment. Bypasses the wrapper system entirely since relenv includes a full salt-call binary." It is never called from anywhere in the codebase.

Why this isn't a trivial one-line fix

Wiring up _run_wfunc_relenv() naively (dispatch on self.opts.get("relenv")) was attempted while investigating #70186 and found to be unsafe as-is:

1. No deploy-bootstrap fallback. self.deploy() (which sends the relenv tarball and, via cmd_block()'s SCP logic, the minion config) is only ever triggered from inside cmd_block()'s "undefined SHIM state" error-detection/retry logic. _run_wfunc_relenv() has no equivalent -- it just runs {thin_dir}/salt-call and assumes the target is already deployed. Routing state.apply/etc. straight to it would break the very first command against a fresh target.
2. Config-dir mismatch. _run_wfunc_relenv() passes --config-dir={thin_dir}/conf to salt-call, but the actual shim (SSH_SH_SHIM_RELENV) writes the minion config to {thin_dir}/minion and invokes salt-call -c "{THIN_DIR}" (i.e. config lives at {thin_dir}, not {thin_dir}/conf). _run_wfunc_relenv()'s own config path appears to have never been exercised/tested.
3. Some wrapper functions may be master-side-only. salt/client/ssh/wrapper/state.py (and others) are heavily __context__["fileclient"]-dependent -- pillar/state compilation happens on the master using the master's fileserver access, which a bare salt-call --local on the target cannot replicate. It's not obvious which wrapper functions are safe to bypass this way without wrapper-by-wrapper review.

Suggested next step

Either: (a) fix _run_wfunc_relenv()'s config-dir path and add deploy-bootstrap/retry logic so it can safely replace _run_wfunc_thin() for the functions where that's valid (likely just state.*), or (b) remove _run_wfunc_relenv() entirely as unused/superseded dead code if no one intends to finish wiring it up.

Related

• Split out from salt-ssh relenv minion config embeds the master's own opts (__master_opts__), causing ARG_MAX failures on state execution #70186 during investigation -- the actual ARG_MAX bug there was a separate __master_opts__ embedding issue, fixed in Stop embedding master opts in relenv minion config to fix ARG_MAX #70194. This dispatch gap is real but was not the cause of that bug.