Add Generate Map tab to the lobby map chooser.

[pchote]

This PR introduces a third classification of map (together with local and remote): MapClassification.Generated, which is constructed at runtime by passing an opaque yaml blob to a mod-defined map generation trait.

The lobby random map generator UI is split into its own followup PR, so this includes a simple testcase that overrides the "Change Map" button to select a random map with default settings. Note that the default settings in RA fail for at least one of the tilesets (probably Interior?), so it may take a couple of button presses for it to work (and/or just test in TD).

Depends on #21849, #21850, #21859, #21860, #21861, #21868, #21869, #21870, #21871, #21876, #21877.

Strongly recommend reviewing by commit.

[pchote]

Rebased.

[PunkPun]

Got several crashes

OpenRA.FieldLoader+MissingFieldsException: LandTile, WaterTile
   at OpenRA.FieldLoader.Load(Object self, MiniYaml my) in OpenRA/OpenRA.Game/FieldLoader.cs:line 591
   at OpenRA.Mods.Common.Traits.ExperimentalMapGeneratorInfo.Parameters..ctor(Map map, MiniYaml my) in OpenRA/OpenRA.Mods.Common/Traits/World/ExperimentalMapGenerator.cs:line 234
   at OpenRA.Mods.Common.Traits.ExperimentalMapGeneratorInfo.Generate(ModData modData, MapGenerationArgs args) in OpenRA/OpenRA.Mods.Common/Traits/World/ExperimentalMapGenerator.cs:line 515
   at OpenRA.Mods.Common.Widgets.Logic.LobbyLogic.<>c__DisplayClass55_2.<.ctor>b__53() in OpenRA/OpenRA.Mods.Common/Widgets/Logic/Lobby/LobbyLogic.cs:line 286
OpenRA.FieldLoader+MissingFieldsException: LandTile, WaterTile
   at OpenRA.FieldLoader.Load(Object self, MiniYaml my) in OpenRA/OpenRA.Game/FieldLoader.cs:line 591
   at OpenRA.Mods.Common.Traits.ExperimentalMapGeneratorInfo.Parameters..ctor(Map map, MiniYaml my) in OpenRA/OpenRA.Mods.Common/Traits/World/ExperimentalMapGenerator.cs:line 234
   at OpenRA.Mods.Common.Traits.ExperimentalMapGeneratorInfo.Generate(ModData modData, MapGenerationArgs args) in OpenRA/OpenRA.Mods.Common/Traits/World/ExperimentalMapGenerator.cs:line 515
   at OpenRA.Mods.Common.Widgets.Logic.LobbyLogic.<>c__DisplayClass55_2.<.ctor>b__53() in OpenRA/OpenRA.Mods.Common/Widgets/Logic/Lobby/LobbyLogic.cs:line 286

Exception of type `System.NullReferenceException`: Object reference not set to an instance of an object.
at OpenRA.MiniYaml.FromString(String text, String name, Boolean discardCommentsAndWhitespace, HashSet`1 stringPool) in OpenRA/OpenRA.Game/MiniYaml.cs:line 398
at OpenRA.Network.UnitOrders.ProcessOrder(OrderManager orderManager, World world, Int32 clientId, Order order) in OpenRA/OpenRA.Game/Network/UnitOrders.cs:line 394
at OpenRA.Network.OrderManager.ReceiveImmediateOrders(Int32 clientId, OrderPacket orders) in OpenRA/OpenRA.Game/Network/OrderManager.cs:line 180
at OpenRA.Network.NetworkConnection.OpenRA.Network.IConnection.Receive(OrderManager orderManager) in /OpenRA/OpenRA.Game/Network/Connection.cs:line 350
at OpenRA.Sync.RunUnsynced[T](Boolean checkSyncHash, World world, Func`1 fn) in /OpenRA/OpenRA.Game/Sync.cs:line 192

[PunkPun]

I assume this is still in proof of concept stage. Lots of polish still seems missing

[pchote]

OpenRA.FieldLoader+MissingFieldsException: LandTile, WaterTile

This is

Note that the default settings in RA fail for at least one of the tilesets (probably Interior?), so it may take a couple of button presses for it to work (and/or just test in TD).

and is unrelated to this PR.

[PunkPun]

the second crash happens when in skirmish you exit a lobby with a generated map and return.

I recon we should not to hurry to merge this, but wait for a feature complete version and do full testing. There really isn't a need to break bleed. Some of the commits should be split off into separate PR's

[pchote]

Agreed, but do feel it important for the core plumbing to be tested and sanity checked independently from the UI.

I can split out some of the commits, but note that these won't then have any testcases.

[PunkPun]

We can kee changes without a testcase here. Other's can be quickly reviewed

[pchote]

Converted to draft, added the (still WIP) lobby UI, and split out a bunch of prereq PRs.

[PunkPun]

wouldn't having a round map actually increase symmetry?

[pchote]

Yes, but you can't make a round map out of a rectangular area.

[pchote]

Small partial update:

• Removed non-square map sizes (deferring rectangular symmetry fixes to a future PR)
• Only update the seed when pressing Generate button
• Fixed map settings not persisting when reopening the chooser (this was surprisingly hard to fix without hardcoding implementation-specific logic!)

[PunkPun]

This feels much better. One more kink left; when you click change map in skirmish lobby, the map is still regenerated despite it being already generated

[pchote]

One more kink left; when you click change map in skirmish lobby, the map is still regenerated despite it being already generated

The issue is that, at this point, it hasn't been generated (in the form needed by the generator UI). The previous Map instance disappears when the dialog is closed, and the rest of the game deals instead with a MapPreview. Getting a Map from a MapPreview triggers a load screen unless it is done on a background thread, so we're still going to need to show the "Generating..." label just for a shorter time. The additional complexity IMO isn't worth this slight improvement.

[PunkPun]

In that case we can handle that at a later point in time if it's necessary

[pchote]

Another small partial update:

• Dispose GeneratedMapPreviewWidget sheet.
• Disable "Ok" button while a map is generating / has failed.

Remaining tasks:

• Investigate crash on loading saved game.
• Move Generate button below the preview.
• Move the Boundary Shape option.
• Split additional prereq changes from the final commit into their own commits.

[pchote]

UI updated, including a fix for previously generated maps.

[pchote]

Updated with a fix for the save game crash.

This happened if you try to load a save after closing and reopening the game. The generated map was not in the cache, so the server selected a different map to use instead...

Moved the MapCache update logic earlier in the process to prevent this.

Ready for re-review.

[pchote]

Added a special case to avoid regenerating the map for skirmish/hosts, which significantly speeds up selecting a generated map.

[PunkPun]

If you generate the map, and leave lobby, then when you return to lobby the map stays. However, when you click to change the map it regenerates.

I assume this bug happens because we don't remember map generation options when one exists skirmish lobby. Maybe it should be put into the lobby skirmish map options serialization?

[PunkPun]

LGTM, additional fixes can be done in followups

[PunkPun]

https://github.com/OpenRA/OpenRA/wiki/Changelog-(bleed)/_compare/34486edf9a958ad87247d117e41d4193887bcc11...9b3ac488c3ea382f211e08cab3b599a8287c4d1a