Pico 2 W prints a newspaper every morning: pulls jobs from GitHub over HTTPS, streams them to a printer over IPP

[ierolsen]

I built a small system where a Raspberry Pi Pico 2 W running MicroPython 1.28 is the only part that runs at home. Every five minutes it asks a private GitHub repo whether a print job is waiting, holds the job until its print time, finds the printer on the LAN and streams the file over IPP in 4 KB pieces.

A few MicroPython details that took some work:

• Time without an RTC battery. The board takes the time from GitHub's Date header and does not print until it has heard from GitHub. The date math (http_date_to_unix) is tested against CPython's calendar on 13 edge dates and 5,000 random ones.
• No double prints. A job goes into state.json the moment the printer accepts it, so a crash or a reboot never sends it twice. A print that is cut off (I once switched the printer off mid-job) is sent again when the printer comes back.
• Finding the printer. Last known address first, then a quick scan of the subnet, then a patient one, because a sleeping printer answers slowly.
• Memory. The file never sits in RAM: it streams from the HTTPS response straight into the IPP request.
• Raster. My HP LaserJet M111w only takes raster, so GitHub Actions converts each page to PWG raster before the Pico sees it.

The pages are a two-page newspaper that a Claude Code routine writes overnight, but the Pico side is plain MicroPython and prints PDF, PWG raster or JPEG jobs, whichever the printer takes.

Code and a step-by-step Pico guide (docs/pico.md): https://github.com/ierolsen/coffee-edition-kit
Write-up with photos: https://erolsen.dev/writing/coffee-edition

I would like to hear how others handle time and watchdogs on the Pico W.

[peterhinch]

One way to handle time is an NTP query on power up. The Pico's RTC will keep time while powered up.

[ierolsen]

Thanks, Peter!

NTP would work too. I went with GitHub's Date header because the board already asks GitHub for jobs every five minutes, so every answer re-syncs the clock (the last Date header plus ticks_diff since it arrived), with no second protocol or server. After a reboot, scheduled jobs are held until the first answer, so nothing prints at a wrong time.

What I haven't solved is a watchdog. The board froze once after a print, but the RP2 watchdog only allows a few seconds and can't be stopped once started, while my loop sleeps for minutes between polls. Would you feed it from a uasyncio task, or wake up every few seconds?

[peterhinch]

If your application is asynchronous you can implement a software watchdog with the Delay_ms class. However I'm unclear what you mean by your loop "sleeping". An

await asyncio.sleep(long_time)

is fine. The hardware sleeps provided by machine do not play nicely with asyncio - nor does time.sleep().

[ierolsen]

Thanks! By "sleeping" I meant plain time.sleep(): the app is synchronous, so I'd port it to asyncio for Delay_ms. If it hangs inside a blocking call like the TLS handshake, though, the event loop stops too. Would you pair Delay_ms with the hardware WDT for that?

[peterhinch]

The asyncio stream mechanism supports nonblocking TLS.

In principle if you are stuck with a blocking function which may hang, I know of two solutions: a hardware WDC or using threaded/multi-core code. See Taming blocking functions - in this case you could use Delay_ms to trigger a hardware reset.

[ierolsen]

Thanks, Peter, that's the piece I was missing. I'll try asyncio streams with Delay_ms triggering machine.reset(), and keep getaddrinfo on the second core.