firmware/test/fixtures/portduino-config

Portduino config fixtures

Input files for bin/test-config-check.sh, which drives a built meshtasticd binary and asserts what meshtasticd --check reports about each one. Every file here is referenced by name from that script, so renaming one means editing it too.

The theme is configuration that is accepted by the YAML parser but does not mean what it looks like it means - the failures that otherwise only show up as a radio that never transmits.

Each file carries a comment header naming its planted fault and what the checker is expected to say about it, so a fixture can be read on its own. One assertion, malformed-indent.yaml's reported line number, counts those header lines - editing that file's comments means updating the expected line in the script.

These files are exempt from trunk fmt (see .trunk/trunk.yaml): prettier rejects the duplicate keys and bad indentation that are the entire point of them.

Valid, one per radio module family

These must each report zero errors and zero warnings, and must resolve to the module named in the file. A silent fallback to sim would otherwise pass.

File	Module	Family
module-rf95.yaml	RF95	SX127x
module-sx1262.yaml	sx1262	SX126x
module-sx1268.yaml	sx1268	SX126x
module-llcc68.yaml	LLCC68	SX126x
module-sx1280.yaml	sx1280	SX128x
module-lr1110.yaml	lr1110	LR11xx
module-lr1120.yaml	lr1120	LR11xx
module-lr1121.yaml	lr1121	LR11xx
module-sim.yaml	sim	simulated
module-auto.yaml	auto	autodetect

valid.yaml is a minimal SX126x config, and empty-sections.yaml (Lora: with no body) exists to prove the checker does not invent a finding for it.

Module naming

Names are matched exactly and are not consistently cased - RF95 and LLCC68 are upper, sx1262 and lr1121 lower.

File	Expected
module-unknown.yaml	sx1263 is refused, and the valid set is listed.
module-wrong-case.yaml	llcc68 is refused with a "did you mean" for case.

LR11xx rfswitch table

File	Expected
rfswitch-valid.yaml	A full seven-mode table on an lr1121 is clean.
rfswitch-partial.yaml	Legal, but the omitted modes are named - they are driven all-LOW.
rfswitch-bad-pin.yaml	DIO9 is a real pin name but not a switch pin on an lr1121.
rfswitch-row-length.yaml	Rows shorter and longer than the declared pin count.
rfswitch-bad-level.yaml	high and On: anything not exactly HIGH is silently LOW.
rfswitch-no-pins.yaml	No pins list, so no switch pin is ever driven.
rfswitch-too-many-pins.yaml	Six pins declared; only the first five are read.
rfswitch-not-a-map.yaml	rfswitch_table given a scalar.
rfswitch-unknown-mode.yaml	MODE_TRANSMIT is not a mode.
rfswitch-stranded-modes.yaml	A MODE_ row one level out, sitting under Lora: doing nothing.
rfswitch-auto-partial.yaml	Silence guard - under auto the omitted modes are not named.
rfswitch-inert-table.yaml	Silence guard - an sx1262's rows are not judged individually.

rfswitch-partial.yaml is legal but noted: the omitted modes are driven all-LOW. module-mismatch-lr11xx.yaml (LR11xx with no table - cannot transmit) and module-mismatch-sx126x.yaml (a table on a radio that never applies one) cover the module/table disagreement in both directions.

Both silence guards exist because a mode list is only meaningful once the radio is known. Under Module: auto the part has not been probed, so naming the omitted modes against the union of both families would advise adding MODE_TX_HP, MODE_GNSS and MODE_WIFI rows to what may turn out to be an LR20x0. On a module that applies no table at all, singling out one row as ignored would imply the others are used, when the single module-mismatch-sx126x.yaml warning already says the whole table is inert.

LR20x0 rfswitch table and interrupt DIO

The table is handed to an LR20x0 as well as an LR11xx, and the two parts have neither the same modes nor the same switch pins, so which findings are correct depends on the module.

File	Expected
rfswitch-lr2021.yaml	Clean. MODE_RX_HF is a real mode here, and IRQ_DIO_NUM keeps the IRQ clear.
rfswitch-lr2021-irq-collision.yaml	IRQ_DIO_NUM: 5 names a pin the table also drives as a switch line.
rfswitch-lr2021-irq-default.yaml	The same collision reached by omitting the key: the radio default is DIO5.
rfswitch-lr2021-irq-clear.yaml	False-positive guard - DIO5 as the IRQ, table on DIO6/7/8, is clean.
rfswitch-lr2021-irq-all-low.yaml	DIO5 listed in pins but driven LOW everywhere: still a collision.
rfswitch-lr2021-irq-out-of-range.yaml	IRQ_DIO_NUM: 3 is outside DIO5-DIO11, so it is discarded and DIO5 is used.
rfswitch-lr2021-irq-alias.yaml	LR2021_IRQ_DIO_NUM alongside IRQ_DIO_NUM: the older spelling does nothing.
rfswitch-lr2021-no-table.yaml	An LR20x0 with no table cannot transmit, same as an LR11xx without one.
rfswitch-lr2021-wrong-mode.yaml	MODE_TX_HP and MODE_GNSS are LR11xx modes an LR20x0 does not have.

begin() needs only SPI and BUSY, so a radio whose interrupt lands on a switch pin still reports init success and then never receives a packet.

An out-of-range IRQ_DIO_NUM is refused twice, in loadConfig() and again in the driver before it reaches RadioLib, so the checker judges it per file: by the time the merged config is built a rejected value and an absent key look the same.

Why -irq-all-low is a fault and -irq-clear is not: LR2021::config() (from begin()) points the IRQ DIO at FUNCTION_IRQ, then setRfSwitchTable() calls setDioFunction(..., FUNCTION_RF_SWITCH) for every non-NC pin in the list whatever the levels are, and nothing re-asserts the IRQ function afterwards. Listing the pin is what breaks it, not driving it.

rfswitch-lr2021.yaml also carries LR2021_MAX_POWER and LR2021_MAX_POWER_HF in a case that must stay at zero warnings, so dropping either from the checker's schema fails it.

TCXO probing (Lora.TCXO_OPTIONAL)

A variant declares "a TCXO may or may not be fitted" at compile time with TCXO_OPTIONAL. A Portduino carrier cannot: the same meshtasticd binary runs on hardware populated either way, so the statement arrives as YAML and is answered at runtime.

File	Expected
tcxo-optional.yaml	Clean. No Vref given, so the TCXO attempt uses the 1.6 V radio default.
tcxo-optional-sx1262.yaml	Clean. Another family, explicit Vref, which is the one reported.
tcxo-optional-unsupported.yaml	An SX128x has no TCXO reference to probe for, so the key is inert.
tcxo-optional-contradiction.yaml	DIO3_TCXO_VOLTAGE: false asks for DIO3 to be left alone; the probe drives it anyway.

With the probe asked for and no voltage given, the driver tries the radio default rather than skipping the TCXO attempt - otherwise there is nothing to fall back from.

PA gain table (TX_GAIN_LORA)

Two shapes are accepted and they fail differently. A list is read element-by-element with .as<int>() and NO fallback, so one bad entry throws and meshtasticd will not start. A bare scalar is read as .as<int>(0) and merely falls back to 0. The table is uint16_t[22], so extra points are dropped and out-of-range values wrap.

File	Expected
txgain-scalar.yaml	Clean, and a regression guard - an earlier checker called this fatal.
value-type-fatal-list.yaml	A non-numeric list entry: throws, so meshtasticd will not start.
txgain-out-of-range.yaml	-5 and 70000 wrap to a different gain than written.
txgain-too-many.yaml	25 points; everything past the 22nd is dropped.

Value types, ranges and units

File	Expected
value-type-fatal.yaml	Logging.AsciiLogs is the other no-fallback read: a bad value stops meshtasticd starting.
value-type-silent.yaml	Wrong-typed values where the read has a default: silently replaced, so the setting does nothing.
tcxo-millivolts.yaml	DIO3_TCXO_VOLTAGE is in VOLTS and multiplied by 1000, so 1800 silently asks for 1800V. Write 1.8.
port-out-of-range.yaml	APIPort outside 1024-65535 is silently ignored; Webserver.Port has no guard at all.
statusmessage-long.yaml	Copied into a char[80], so it is safe but silently shortened to 79 characters.
configdir-missing.yaml	Crash regression guard - an unreadable ConfigDirectory used to abort meshtasticd (and --check) with SIGABRT via an uncaught filesystem_error.

MAC address

The MAC no longer determines NodeNum - that comes from the public key - but a MAC that fails to apply still falls through to the BlueZ and LoRa-serial fallbacks, and if those yield nothing meshtasticd exits with "Blank MAC Address not allowed!".

File	Expected
mac-conflict.yaml	Both MACAddress and MACAddressSource; meshtasticd refuses.
mac-malformed.yaml	AA:BB:CC is under 12 hex digits, so it is silently dropped.
mac-source-missing.yaml	Names an interface with no /sys/class/net/<n>/address. Warning, not an error: it is machine-dependent and may be checked on another host.

Joystick buttons

Input.JoystickButtons is keyed by action, not by button, so an action can name a single evdev code or a list of them and every one of those buttons drives it.

File	Expected
joystick-buttons.yaml	Clean, and a regression guard - four actions, two of them with several codes.
joystick-buttons-bad.yaml	Three silent no-ops: an action name nothing reads, an evdev name where a code belongs, one code claimed twice.

CH341 USB-SPI adapters

spidev: ch341 is a different hardware model, not a variant of the same one. The Lora pins become indexes on the adapter and are driven by the usermode USB driver - portduinoSetup() skips initGPIOPin() for every one of them - so nothing is claimed from a gpiochip. This is also the only shape that works on Windows and macOS, which have no gpiochip, gpiodetect or gpioinfo to check anything against.

File	Expected
usb-ch341.yaml	Clean, and the report lists adapter pins rather than resolved gpiochip lines.
ch341-gpiochip.yaml	A gpiochip and line mapping alongside ch341: read, stored, and never used.

Structure

File	Expected
duplicate-key.yaml	yaml-cpp keeps the FIRST duplicate, so the later value is lost.
nonmap-section.yaml	Lora: invalid - a known section whose body cannot be read.
unknown-section.yaml	A top-level section meshtasticd never reads.
stranded-key.yaml	spidev left at the top level instead of inside a section.
top-level-list.yaml	Document root is a sequence.
empty-file.yaml	No content: comments only, which parse to a null document. A warning, not an error.
malformed-indent.yaml	Will not parse; the report must still name the file and line.
pin-unknown-subkey.yaml	A pin mapping accepts only pin, gpiochip and line.
pin-unreadable.yaml	A non-numeric pin resolves to -1 and trips an assertion at startup.
hub75-unknown-key.yaml	An unknown Display.HUB75 option. On a build without rgbmatrix this also reports the missing HUB75 support, so the test asserts only the unknown key.

Across a config directory

configd-conflict/ is a whole tree: a config.yaml pointing at a config.d/ holding two more Lora: sections. It covers the trap that a key not repeated in the last-loaded file is reset to its default - here config.yaml sets DIO3_TCXO_VOLTAGE: 1800 and the effective configuration ends up without it. The load order within config.d/ comes from the filesystem, so the report warns rather than assuming alphabetical order.

rfswitch-last-wins/ covers Lora.rfswitch_table across two config.d/ files. It follows the same "last file loaded wins" rule as every other Lora: key - the loader resets a table's pins and mode rows before applying a replacement, so an earlier file's MODE_RX setting cannot leak through a later file that omits it. --check reports this as the standard cross-file-overlap info, not a special-cased error.

rfswitch-replace/ pins the replacement itself rather than the diagnostic. Which of two config.d/ files wins is up to the filesystem, so the fixture above cannot assert the effective table by value; here the losing table sits in config.yaml, which is always loaded before config.d/, and the winner is therefore deterministic. The loser is the wider of the two - four pins and three mode rows, all HIGH - so any carryover appears as a surviving pin, a surviving mode row, or a HIGH that should be LOW. --output-yaml is what reports it: the --check report says no more than set.

Running these as a normal boot

malformed-indent.yaml, nonmap-section.yaml, module-unknown.yaml, mac-conflict.yaml and hub75-unknown-key.yaml are also run without --check, where each must be rejected with a non-zero exit. That is the guard on --check mode not having quietly made the normal path permissive. No other fixture is run that way: a config meshtasticd accepts makes it boot a node and block.