Per review: a 2D fix is valid, only less precise, and in some places a
receiver never gets a 3D fix. So the guard no longer rejects it; instead
the fix mode decides when the GPS may sleep:
- the fix mode is read from GSA for every NMEA receiver, not only with
GPS_GUARD (`fm:` in `gps diag`), and stored with the last fix;
- in periodic mode, and for `gps once`, the GPS sleeps only after a 3D
fix (or a fix from a receiver that does not report the mode); after a
2D fix it stays on up to the acquisition timeout;
- a 2D fix does not replace a 3D fix younger than 5 min
(GPS_3D_FIX_PREFERRED_MS).
A cold start can take up to ~30 min with a weak sky view, so the default
limit before the first fix since boot goes from 15 min to 1 h.
The limit is now a GPS preference, set with `gps firstfix <duration>`
(0 = no limit, at most 1 day) and saved with the other GPS settings, so
repeaters and companions can use different values without a rebuild.
GPS_FIRST_FIX_TIMEOUT_MS stays as the build default.
Before the first fix since boot, a periodic acquisition had no timeout. If no
fix is accepted (no sky view, no antenna, or a guard rejecting spoofed fixes),
the receiver then stayed on until a fix was accepted.
Give that window its own limit, GPS_FIRST_FIX_TIMEOUT_MS (15 min by default,
enough for a full almanac; 0 keeps the old behaviour). It counts as a miss and
the next acquisition starts after the configured period.
The guard checked satellites and HDOP but not the fix type. When the
receiver sends GSA, a fix it reports as 2D or no-fix is now rejected
(`rej 2d`). Receivers without GSA are unchanged (fix mode unknown).
GPS_GUARD_REQUIRE_3D=0 turns the check off. `gps diag` shows the fix
mode as `fm`.
Add a 16-byte zeroed tail behind the 100-byte NMEA buffer. MicroNMEA's
fixed-width field skips can read a few bytes past the end of a malformed
sentence; with the tail those reads land on NUL bytes.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
#3813 removed the telemetry channel for the GPS time alert. Remove the
docs sentence that still described it and the now unused
GpsGuard::getAlertOffset().
The remote-CLI reply buffers in onPeerDataRecv() (repeater, room server,
sensor) and BaseChatMesh were sized temp[166] for a 5-byte datagram
header plus 161 bytes of reply. But MyMesh::handleCommand() first
reflects a 3-byte correlation prefix (e.g. 'aa|') into the reply, after
which CommonCLI commands may write up to 160 chars + NUL into the
remaining 158 bytes -- overflowing the buffer by 2 bytes (e.g. a full
'regions' listing via exportTo(reply, 160)).
Triggering this remotely crashes the node (observed as an unresponsive
repeater after a prefixed 'regions' query); the 2 out-of-bounds bytes
are a printable region-name character and a NUL.
Size the buffers for the full worst case instead:
5 header + 3 prefix + 160 reply + 1 NUL = 169.
Verified with a host ASan harness reproducing the exact layout:
prefixed 'regions' with a 25-region listing overflowed temp[166]
(BufStream::write, RegionMap.cpp) and now passes clean with temp[169];
unprefixed (serial) path unaffected.
When the last GPS time check passed but there is no current trusted fix
(e.g. the GPS is asleep between periodic fixes), gps guard showed
"time rej nofix", which reads like a rejection although the clock is fine.
Show "time wait" instead. A real rejection without a trusted fix (GPS time
arriving while fixes are rejected) still shows "rej nofix".
The probe list was wired only in variants/promicro/target.cpp. It is now the
default list of NmeaGpsDriver when GPS_BAUD_PROBE_LIST is set, so a variant
enables baud autodetection with a build flag alone.
Station G2 and G3 enable it: their Grove UART connector is used with M5Stack
Unit GPS modules, which default to 115200.