newly generated XCMs unusable on waeker platforms, features missing

[bomilkar]

Comparing xcm files between 2022 and now (Feb 2026) using the example files:

GER_HighRes.xcm from https://download.xcsoar.org/source/map/region/GER_HighRes.xcm
Generated on 2022–07-01, uncompressed, including “small villages”, latitude range: 47.0 to 55.0, longitude range: 5.5 to 15.5

GER_HighRes_2026.xcm from the map generator:
Generated on 2026–02-24, same features as the 2022 xcm.

Here are the differences in size:

1. xcm file size: 80M vs. 455M, this is 5.5 x lager file size.
2. the new xcm has additional features:
   “water_area_small” accounting for 174M
   “airstrip_area” accounting for 5M
3. the following features grew:
   “roadbig_line” from 0.7M to 79M
   “roadmedium_line” from 3M to 108M
4. unchanged in size:
   roadsmall_line, railway_line, city_point, town_point, suburb_point, village_point
   this accounts for 35M
5. the files “terrain.jp2” are same size.

Looking at the map, comparing Map_1, the most obvious differences:

2022:

2026:

1. line features such as roads and rivers have a much “smoother” shape. The point-density of these features is significantly higher in the new xcm.
2. some smaller creeks are missing in the new xcm.

Looking at the map, comparing Map_2, :

2022:

2026:

1. a major road is missing in the new xcm: it’s B15n a 4 lane road with a center divide east of “Ma12”.
2. smaller lakes are missing in the new xcm: around “Ingol” and east an west of “Lands”.
3. the new xcm shows a lot of medium size roads: the light brown colored roads.

Summary:

1. some features were added with a much higher point-density. This comes at a significant cost in terms of compute cycles while rendering the map. For instance roads and waters. The map may look nicer with the higher point-density but to me as a pilot it doesn’t add any value.
2. some features are missing: a 4 lane major road with a center divide “B15n” easy to discover on the ground, hence a significant loss on the map; smaller creeks are missing, which I’m interested in, because they are humidity sources and indicative of the topography.

Conclusion:

1. there should be a switch in the map generator to control the point-density, similar to “No small villages or smaller roads”. A high point-density can make the xcm unusable because it consumes too much CPU power on a weaker platforms such as Cubie2.
2. we need to figure out why important features are missing.

[bomilkar]

Here is more detailed comparison between the 2 archives.
GER_compare_2022vs2026.xlsx

[lordfolken]

So why are you using a HighRes xcm on a weaker platform?
Full Germany with minimal settings is 55Mb.

The old dataset VMAP0 hasn't been updated since the early 00s. The lakes, rivers and roads are offset by several km, so even visual navigation wasn´t really possible. Besides countless bug reports of features missing.

Input is welcome: https://github.com/XCSoar/xcsoar-mapgen-topology

[bomilkar]

The only difference between LowRes and HighRes is "terrain.jp2", which I don't see impacting the performance. The difference is 7.4M vs. 37.4M. Isn't terrain used to calculate H AGL ? 3 arc sec is equivalent to 93m. It shouldn't be much less (in the mountains).

55M is the compressed size. Uncompressed it's 126M for nearly 0 features. And the issue is not the file size. It's the CPU speed while redrawing the map.