OSM Scout Server Android Port

[woodyst]

Hi! Thank you very much for your program. It has been very useful for me.

I have made an Android port (https://www.egpsistemas.com/site/osmscout) because I need it for my Navius GPS Navigator (https://www.egpsistemas.com/site/navius).

I want to publish it in google play store because it is an important piece for my navigator for offline navigation.

I write to you because I want to know if you agree with it and if you have any consideration.

Thanks.

Edu.

[rinigus]

Very interesting. Looking at the code, there are significant changes from the original. Maybe we one day can merge it to have a single code base that spans from linux to android.

As for you distributing your work - sure, as long as you comply with GPL :). Good luck and maybe you will chip in later with maps updates ...

[woodyst]

Thank you very much.

I prefer not publishing it if the owner disagrees, because may be you have other plans, and I respect your work. But It was needed if I want to enable offline navigation.

About the changes, in case they are useful to you:

Two of them are not Android-specific, and I think you may want them
regardless.

The first one is a real bug in geocoder-nlp, in
search_rank_location_bias(). The zoom is clamped from below:

zoom = std::max(zoom, 18);

so (18 - zoom) is never positive and the location bias radius is always
250 m. Beyond a couple of kilometres exp(-distance/250) is zero and the
reference point stops having any effect. Searching for "Barcelona" while
in Barcelona returned streets in Salamanca, Leon and Almeria, exactly the
same with and without a reference point. There is also undefined
behaviour: with zoom > 18 the shift 1 << (18 - zoom) is negative. The
formula comes from Photon, which clamps from above at 18, which is what
makes sense -- less zoom, larger radius. What I use is:

zoom = std::min(std::max(zoom, 1), 18);

This affects Sailfish and Ubuntu Touch too, not only Android.

The second one is a compile error with Qt 6 in uhttp's
microhttpserver.cpp:

size_t tosend = std::min<size_t>(max, (size_t)data.size() - p);

QByteArray::size() returns qsizetype in Qt 6, while libmicrohttpd passes
a size_t, so std::min cannot deduce the type. Anyone building that file
with Qt 6 will hit it.

Everything else is Android-specific and quite contained. The engines,
the algorithms and the /v1 and /v2 API are yours; I did not change them.
What I had to replace is the activation: there is no D-Bus on Android,
so the server runs as a foreground service in its own process and the
client starts it with an explicit Intent. The client also has to declare
<package .../> in its manifest or Android 11+ blocks
the Intent silently. /v1/activate is kept for compatibility but answers
immediately, since if HTTP responds the server is already up.

One thing that is an addition to your API and not just a port: I added
/v1/status, because with the server in its own process its own UI no
longer has the engines at hand. If we ever merge the code bases, that is
a design decision you should make, not me.

The map files are used exactly as your server prepares them -- nothing
is generated or imported on the device. Valhalla mmaps the tiles, so the
map directory has to be a real filesystem path; a content:// URI from
the Storage Access Framework does not work.

The deviations from your code are kept in a CAMBIOS.md file next to each
vendored directory, and marked in the code, so they can be re-applied if
you update upstream. I am happy to send them as patches in whatever form
suits you.

Thanks again.

Edu.

[rinigus]

generic bugs and improvements - please submit them to corresponding repos. Otherwise we will loose those

[woodyst]

ok. I'll do, thanks