Fixes and improvements for ESP32 in AP mode:

[tobiashennerbichler]

• Now correctly reports the device's IP in AP mode for UDP discovery requests. Previously reported as 0.0.0.0.
• Split logic between AP mode and Infra modes since WiFi.status() returns a meaningless value in AP mode. AP mode now immediately establishes TCP/UDP servers. Server establishment moved to own function.
• Also adds null-pointer guards to tcp/udp loop while servers are not established yet. Not relevant for AP mode, but included here due to simplicity.

Tested on ESP32 in WIFI_MODE_AP_ONLY. This PR does NOT address the infrastructure modes or the WIFI_MODE_ATTEMPT_INFRASTRUCTURE_FAIL_TO_AP mode in particular, which has some issues in the failover trigger. Local changes are ready and can be included in future PRs if desired.

[ClutchplateDude]

Code Review — "Fixes and improvements for ESP32 in AP mode"

Scope: ESP32 Wi-Fi in WIFI_MODE_AP_ONLY (infra / WIFI_MODE_ATTEMPT_INFRASTRUCTURE_FAIL_TO_AP explicitly excluded per the PR description). src/WifiControl.cpp / .hpp, +86 / −57.

Overview

• startAccessPointMode() — adds WiFi.mode(WIFI_AP) and reorders to mode → softAPConfig → softAP, then calls the new establishServers().
• establishServers() — extracted helper that (re)creates the TCP/UDP servers (was inline in loop()).
• getIP() — returns softAPIP() in AP-only, localIP() otherwise, "NONE" when disabled; used by getStatus() and udpLoop().
• loop() — the status-tracking / server-setup / failover block is now gated by if (WIFI_MODE != WIFI_MODE_AP_ONLY), so AP mode skips it entirely.
• tcpLoop() / udpLoop() — null-pointer guards (if (!_tcpServer) return; / if (!_udp) return;); available() → accept().

The core AP-mode fix is correct and complete, and the refactor is clean. One real regression for the failover path, plus a couple of minor loose ends.

Strengths

• Root-cause fix is right. Adding WiFi.mode(WIFI_AP) before softAP()/softAPConfig() is the documented ESP32 order; without WIFI_AP set, softAPIP() returns 0.0.0.0 — exactly the symptom described.
• Null guards are the load-bearing piece of the refactor. Because tcpLoop/udpLoop now no-op until servers exist, it becomes safe to stop early-returning out of loop() when not yet connected. Internally coherent design.
• establishServers() extraction removes duplication (the same code ran in loop() and now also runs in startAccessPointMode(), which is itself called from infraToAPFailover() — so failover correctly re-establishes servers too).
• _status = WL_DISCONNECTED member init fixes a previously-indeterminate wl_status_t.
• available() → accept() is a correct modernization: it accepts on connect rather than waiting for data, then while (client.available()) waits for bytes. Slightly more correct than before.
• Style matches house conventions (Allman braces, 4-space indent, right-aligned pointers, LOG(DEBUG_WIFI, …)).

Issues & Risks

1. [Important] Failover is now edge-triggered — likely regression for WIFI_MODE_ATTEMPT_INFRASTRUCTURE_FAIL_TO_AP

This is the headline finding. The old loop() polled infraToAPFailover() every iteration whenever _status != WL_CONNECTED:

// OLD
_mount->loop();
if (_status != WL_CONNECTED) {
    infraToAPFailover();   // called every loop — timeout gets re-evaluated each tick
    return;
}

The new code only calls it inside the status-change block:

// NEW
if (WIFI_MODE != WIFI_MODE_AP_ONLY) {
    if (_status != WiFi.status()) {
        _status = WiFi.status();
        ...
        if (_status != WL_CONNECTED) { infraToAPFailover(); return; }   // only on a transition
    }
}

infraToAPFailover() is timeout-gated (_infraStart + _infraWait < millis()). After a failed infra attempt, WiFi.status() typically settles at WL_DISCONNECTED and stops changing. At that point _status == WiFi.status() forever, the change-block never re-enters, and the timeout is never re-evaluated — so the fall-over-to-AP never fires. The device hangs on infra indefinitely.

The old every-loop poll is what made the timeout actually expire. The author notes FAIL_TO_AP "has some issues" and is out of scope, but this PR doesn't just leave it alone — it removes the mechanism that let the existing timeout work.

Suggested fix — keep a periodic (not edge) failover poll for the non-AP modes, e.g.:

if (WIFI_MODE != WIFI_MODE_AP_ONLY) {
    if (_status != WiFi.status()) {
        _status = WiFi.status();
        /* log */
        if (_status == WL_CONNECTED) establishServers();
    }
    if (_status != WL_CONNECTED) { infraToAPFailover(); return; }   // back to every-loop
}

The null guards mean re-adding the every-loop return is still safe.

2. [Minor] getStatus() still reports a meaningless Wi-Fi status in AP mode

getStatus() (unchanged) still calls wifiStatus(WiFi.status()). By the PR's own rationale (WiFi.status() is meaningless in AP mode), the status field in AP-only will show an arbitrary string. The PR fixed the IP half of this but not the status half. Consider returning a fixed connected/"Access Point" status for AP-only.

3. [Minor] getIP() returns the wrong IP after a FAIL_TO_AP failover

getIP() keys off the compile-time WIFI_MODE, so for WIFI_MODE_ATTEMPT_INFRASTRUCTURE_FAIL_TO_AP it always returns WiFi.localIP() — even after infraToAPFailover() has switched the radio to AP. The UDP discovery reply and getStatus() would then advertise 0.0.0.0/the stale STA IP in the failed-over state — the exact class of bug this PR is fixing for AP-only. Acceptable given the stated scope, but worth a // TODO or tracking runtime mode.

4. [Nit] Magic number 4031 now appears in two places

4031 (UDP discovery port) was already a bare literal; establishServers() duplicates it alongside the original udpLoop() use. A named constant (e.g. WIFI_DISCOVERY_PORT) would prevent the two from drifting. Pre-existing debt — optional.

Other Dimensions

• Correctness (non-infra): AP-only path is sound end-to-end. establishServers() deletes before new-ing (null-delete is safe), so re-establishment on failover doesn't leak.
• Performance: Negligible — one WiFi.mode() at startup; one extra branch per loop for the null guards.
• Security: No new surface. readStringUntil('#') into a heap String is pre-existing and unchanged.
• Test coverage: No unit tests added, but WifiControl is hardware-bound and the native suite doesn't cover Wi-Fi; manual ESP32 testing in AP-only is noted. Acceptable.
• Conventions: Compliant with the project style rules; should pass clang-format and -Werror.

Verdict

Approve with one requested change: the AP-only fix is correct and well-executed, but issue #1 (failover polling) is a concrete behavioral regression for FAIL_TO_AP that's worth fixing or explicitly documenting as known-broken in this PR. Issues #2–#4 are minor/cosmetic.

[ClutchplateDude]

Thanks for your submission and fixes!