Skip known-unreasonable network location updates

[t-m-w]

If we have a GPS location of a given accuracy, and our proposed network location is more than twice as far away as the GPS accuracy, this is considered unreasonable. Stop sending that network location.

As part of this, also keep track of lower-accuracy GPS locations with accuracy as low as 10km.

[mar-v-in]

• I don't think a buffer of GPS locations make sense in this context. It should be enough to just store the latest GPS location and then compare to it if it's recent enough. Doing so will significantly reduce the complexity of the change. The current use of the buffer will also mean that every location will become "reasonable" if it is old enough (aka the buffer is overrun and there is no reference location anymore to compare against). However we don't want to say anything about how reasonable an outdated location was, we only care about how reasonable it is as a current location.
  About the GPS location buffer

  The current GPS location buffer (with accuracy requirements) is meant for learning network locations (and in the future also to submit them to a server). In that case we need the buffer as a reported wifi hotspot might be from a beacon received a few seconds ago, which when travelling at high speeds can mean an error of up to a few hundred meters.

• I am unsure if locations should be restricted at the network location provider if they mismatch GPS. I would assume it's the task of the network location provider to provide whatever it considers reasonable as a current network location, no matter what the GPS says. It's the purpose of a fused / hybrid location provider to understand what the current location could be in cases where GPS and network data (some of which might be outdated) is available. I would rather want to put more logic like this into our fused location manager (org.microg.gms.location.manager).

[t-m-w]

Sounds good, thanks for all the info; happy to look into moving things around and following those suggestions.

You had mentioned the fused location provider before, but I was having trouble identifying it; knowing that it's org.microg.gms.location.manager should be helpful.

[ale5000-git]

@t-m-w
The min/max functions usually return the opposite of what one expect :p
So if you use the "min" function you should pass the MAX_UNREASONABLE_DISTANCE_FROM_GPS.

[t-m-w]

* I am unsure if locations should be restricted at the network location provider if they mismatch GPS. I would assume it's the task of the network location provider to provide whatever it considers reasonable as a current network location, no matter what the GPS says. It's the purpose of a fused / hybrid location provider to understand what the current location could be in cases where GPS and network data (some of which might be outdated) is available. I would rather want to put more logic like this into our fused location manager (`org.microg.gms.location.manager`).

Is it the case that not all apps use the fused location provider? I've been trying to test this new branch with Android 14 and I don't see the log lines at all when using SatStat or Organic Maps. Added some extra ones locally and I don't see them either.

It would be a shame for the network location to be this far off (thousands of miles, here) and have some apps never be able to recover from that. Then again, this change's only goal was to fix it once GPS is known, not permanently; it doesn't resolve the recurring annoyance when all you have is network location, so maybe it's not really a worthwhile approach...?

[t-m-w]

@t-m-w The min/max functions usually return the opposite of what one expect :p So if you use the "min" function you should pass the MAX_UNREASONABLE_DISTANCE_FROM_GPS.

Oh, yep, I've done it again... Thanks for pointing that out!

[ale5000-git]

Even if the improvement doesn't cover everything it is still a good improvement.

An app can:

1. Use fused location from Google Play Services (this can be controlled directly by microG and works in every case);
2. Use fused location from Android directly (this cannot be controlled directly and it works only if microG is registered as location provider);
3. Use network location from Android directly (it works only if microG is registered as location provider; it is possible that we can still skip wrong results but I'm not sure);
4. Use GPS location from Android directly (this should never reach microG if I'm not wrong).

This is what I have understood.
@mar-v-in
Correct me if I get something wrong.

[t-m-w]

Any recommendations for an app that would take advantage of the approach in the current commits so that I can test it?

[Misalf-git]

FWIW, Organic Maps location jumps from Network to GPS occasionally (I guess).

[mar-v-in]

Apps that have their own location fusing logic (and I assume Organic Maps is one of them) are not affected by this change (and in fact can't really be fixed by us).

The underlying problem is how to understand the accuracy information in a location reported. It is often simplified that the real location is in a circle with a radius of accuracy meters centered around the location given. This is not correct, the real location is only within this circle in 2 out of 3 cases (68th percentile). The remaining cases, it is outside this circle, and this is where the bigger problem is: With GPS, it is never more than a few hundred meters from the given location (no matter the accuracy given), du to how GPS works. With network-based location, as the database might have wrong data, it can be on the other side of the planet.

Now location fusing is what combines network based location and gps based location to a single location, incorporating any metadata from each of the location sources. The basic implementation in AOSP has the simple logic of taking either GPS or network location, if one of them is 11s or more newer than the other, that one will be taken, if they're less than 11s from each other, the one with reported higher accuracy is taken. If apps take this implementation (by requesting fused location from AOSP based systems), they will likely see those jumps, same if they have their own, similarly naive implementations.

We do need a much more sophisticated implementation if we want to compete with Google's service - which is what apps expect. Ignoring locations from the network provider that are implausible given the location information we have from GPS is a good start, but ultimately, we also want to logically merge the information where possible.

[t-m-w]

I understand the idea that apps should be smart about how they fuse GPS and network location, if they choose to do it on their own, but in reality that doesn't seem to be the case, and such apps appear to work fine with reasonable network location providers but appear to be broken with microG. This is what makes me eager for a fix (or at least a workaround) in the network location provider, if the source of the provider's data is already known to be inaccurate in a large number of cases and with no solution on the horizon.

At this rate, Mozilla Location Services will be shut down soon (some day in June I think - I don't remember the date off-hand), so we might be able to get away with postponing this without it making much of a difference.

[mar-v-in]

Can you provide examples of apps that behave peculiar bad with microG location (e.g. a lot of jumping between locations)? So we can better understand in how far this behavior differs in Google and why.

If we just filter out locations from network location so that apps that do fuse can do this better, apps that don't fuse but only request network location won't get updates, which might be worse than giving them inaccurate updates.

[t-m-w]

Can you provide examples of apps that behave peculiar bad with microG location (e.g. a lot of jumping between locations)? So we can better understand in how far this behavior differs in Google and why.

With the test steps below that aim to simulate a loss of GPS coverage, I haven't encountered an app that doesn't jump between locations after a short amount of time. (Copied from https://gitlab.com/CalyxOS/calyxos/-/issues/2257#note_1929698556 )

Some potential repro steps (requires userdebug build / root) tried on Pixel 8 shiba:

1. Go to a place where your device is capable of getting a GPS fix, even if the fix can take a while, and where the network location is not accurate. (If this isn't near a computer with adb access, find some other way to get the needed access.)
2. Open Organic Maps, and observe the incorrect network location. Wait until it's corrected when a GPS fix is established.
3. Run adb shell stop gpsd to stop GPS from updating.
4. Wait a certain amount of time, and your location will be reverted to the incorrect network location. For me this was 20 seconds on one trial, and more than a minute on another trial. (At some point, the current-location indicator should go gray as well, but this may not overlap with when the app decides to shift the map to the incorrect network location.)
5. Run adb shell start gpsd to start GPS updates again, and once a fix is established, the location will update again.

There are the apps I tried, this time:

• Google Earth
• Google Maps
• GMaps WV
• Magic Earth
• Organic Maps
• OsmAnd
• SatStat