the time goes forward 5 minutes each week

I can confirm that commit 5be8869ae36b7d9fe955e1b1071b9f8a58a0­38cf produces the problem:

let firstFix = true;
let mesureStartTime;
let latestLogTime;

function calcTimeDifference(fix) {
  if (!fix.time) {
    return;
  }

  const currentDateTime = new Date();
  const delta = currentDateTime - fix.time;
  if (firstFix || delta < 0) {
    // Set the time to the first time received, or if the time is behind the GPS time 
    // because we have a forward drift
    console.log(`${fix.time.toISOString()} Start measuring`);

    setTime((new Date() - delta) / 1000);
    firstFix = false;
    mesureStartTime = latestLogTime = new Date();
  } else if (!latestLogTime || (currentDateTime - latestLogTime) > 60000) {
    const measureSeconds = (currentDateTime - mesureStartTime) / 1000;
    console.log(`${currentDateTime.toISOString()} difference ${(delta / 1000).toFixed(3)} second/s within ${measureSeconds.toFixed(3)} second/s`);
    latestLogTime = currentDateTime;
  }
}
console.log(process.env.VERSION);
Bangle.on("GPS", calcTimeDifference);
Bangle.setGPSPower(1);

2v21.123
2024-06-23T18:08:40.000Z Start measuring
2024-06-23T18:09:40.012Z difference 0.013 second/s within 60.009 second/s
2024-06-23T18:10:40.020Z difference 0.020 second/s within 120.016 second/s
2024-06-23T18:11:40.025Z difference 0.025 second/s within 180.022 second/s
2024-06-23T18:12:40.033Z difference 0.034 second/s within 240.030 second/s
2024-06-23T18:13:41.026Z difference 0.026 second/s within 301.023 second/s
2024-06-23T18:14:42.019Z difference 0.019 second/s within 362.016 second/s
2024-06-23T18:15:42.023Z difference 0.024 second/s within 422.020 second/s

2v21.124
2024-06-23T18:49:06.063Z Start measuring
2024-06-23T18:50:06.106Z difference 0.044 second/s within 60.039 second/s
2024-06-23T18:51:06.146Z difference 0.084 second/s within 120.079 second/s
2024-06-23T18:52:06.182Z difference 0.119 second/s within 180.116 second/s
2024-06-23T18:53:06.216Z difference 0.153 second/s within 240.149 second/s
2024-06-23T18:54:06.250Z difference 0.187 second/s within 300.183 second/s
2024-06-23T18:55:06.286Z difference 0.224 second/s within 360.219 second/s
2024-06-23T18:56:06.328Z difference 0.266 second/s within 420.262 second/s

https://github.com/espruino/Espruino/com­mit/5be8869ae36b7d9fe955e1b1071b9f8a58a0­38cf

v21.124
This is the culprit, but I have no idea why, it doesn't seem like it should be the culprit.

Good find. The first change looks harmless if the SYSCLK_FREQ is divisible by 1000 (which it maybe is - 64000000 for nrf52?). but the second is now multiplication by small floating point number with low precision so maybe that may be the cause.