thomersch / scapegoat Goto Github PK

The GPS traces layer's attribution uses a hack to display a legend:

"html": "<span style='display: inline-block; padding: 0 8px; background-color: rgba(0,0,0,0.5);'><span style='color: #eee;'>GPS Direction:</span> <span style='font-size: 15px; padding-left: 2px; font-weight: bold;'> <span style='color: #0ee;'>&larr;</span> <span style='color: #96f;'>&darr;</span> <span style='color: #6e0;'>&uarr;</span> <span style='color: #f63;'>&rarr;</span> </span></span> \u00a9 <a href='https://www.openstreetmap.org/copyright'>OpenStreetMap contributors</a>."

It's the only layer that uses the 'html' attribution. JOSM doesn't have html attributions:

https://josm.openstreetmap.de/wiki/Maps/Worldwide#OpenStreetMapGPSTraces

We should instead ask iD to hardcode this if the layer id is "osm-gps" or the url contains "gps-tile.openstreetmap.org" or something like that.

Another hack could be to use an icon? It might get shrunk into unreadability.

The layer ID is currently used to generate the on-disk path for the icon. This could open a security problem, so those strings should be normalized.

ELI's check.py checks for missing attributes - missing license, etc.

https://github.com/osmlab/editor-layer-index/blob/gh-pages/scripts/check.py

These checks should be done on the JOSM stuff, and be shown on some kind of monitoring/QA page, a bit like https://josm.openstreetmap.de/wiki/ImageryCompare

Probably not something we can fix here, but only in thomersch/grandine#3

iD does some filtering and enriching of tile layers that we should reproduce as much as possible

• Only include wms layers that support the projections that iD can use, with only one projection - already done for MVT
• Layers of type wms_endpoint or wmts stripped out
• Remove a few layers, mostly layers that will be of category 'osm' - see discard in iD
• Remove layers that are more than 20 years old - see too old
• Add Maxar layers - see keep
• Rename attribution attributes: terms_url = attribution.url; terms_text = attribution.text

https://xkcd.com/2170/

ELI simplifies all features to 5 decimals:

for file in sys.argv[1:]:
    with io.open(file, 'r') as f:
        # simplify all floats to 5 decimal points
        source_features.append(json.load(f, parse_float=lambda x: round(float(x), 5)))

4 might actually be enough; I was just being conservative when I wrote that.

iD supports these projections:

const supportedWMSProjections = {
  'EPSG:4326': true,
  'EPSG:3857': true,
  'EPSG:900913': true,
  'EPSG:3587': true,
  'EPSG:54004': true,
  'EPSG:41001': true,
  'EPSG:102113': true,
  'EPSG:102100': true,
  'EPSG:3785': true
};

These are basically all Web Mercator + 4326

iD also needs to know which ones the wms server supports to create its requests in 3857 or 4326 - our current replacement won't work.

We should:

• If Web Mercator is supported in any way, advertise the layer in MVT with only one of the supported Web Mercator projections
• If Web Mercator isn't supported but 4326 is, advertise 4326 in MVT
• If neither is, don't advertise the layer in MVT at all - it's useless to iD.

The JOSM xml contains a lot of base64-encoded icons. Every single tile we generate will contain, for example, the maxar and mapbox icons in base64.

Using https://github.com/osmlab/editor-layer-index/blob/gh-pages/scripts/unembed-icons.py , unembed them and host them on https://imagery.openstreetmap.lu/icons/

Using https://nginx.org/en/docs/http/ngx_http_gzip_static_module.html and maybe https://github.com/google/ngx_brotli , pre-compress tiles on server.

One tile I'm looking at contains:

"geometry": {
                "type": "Polygon",
                "coordinates": [
                    [
                        [
                            5.62500000,
                            48.92249926
                        ],
                        [
                            8.43750000,
                            48.92249926
                        ],
                        [
                            8.43750000,
                            50.73645514
                        ],
                        [
                            5.62500000,
                            50.73645514
                        ],
                        [
                            5.62500000,
                            48.92249926
                        ]
                    ]
                ]
            }

We should have 5.625 instead of 5.62500000.

We know from the Accept-Language headers what the human in front of the screen expects, and can pre-generate some languages, or generate the tiles dynamically for more complex requests (like mine, en-GB,en;q=0.9,fr;q=0.8,de;q=0.7,lb;q=0.6).

We can send a Vary: Accept-Language header to prevent the wrong language from getting cached.

We can pre-embed all available translations in a tile, creating attributes like name:de-CH. This might produce big tiles, and would have to be understood by clients.

We can generate tiles per language, e.g. 7/42/42-en-GB.geojson. This means that a client would have to download as many duplicate tiles as the human understands.

We can generate translation tiles that would contain only the translated strings, e.g. 7/42/42-en-GB.json. A client would download the 'normal' scapegoat tile and the translation tile for all wanted languages.

We can use the translations from JOSM, from iD, or both.

See also #4

Worldwide tiles won't change. Having the clients download one worldwide and one local tile will make further local downloads lighter.

Have https://josm.openstreetmap.de/jenkins/job/JOSM-Imagery-Integration/ trigger a Scapegoat build whenever the wiki is updated.

Scapegoat currently runs on openstreetmap.lu and one of Thomas's VMs.

We should move to OSM infrastructure when we're ready.

Many places aren't covered by any local layers: see the map at https://osmlab.github.io/editor-layer-index/

For example, 17/16 and 18/16 both cover Congo-RDC, and contain the same stuff.

Redirecting to a worldwide.(geojson|mvt) would potentially save three tile downloads. Premature optimisation? See also #5.