feat(port): mutable specials and unhardcored labs and anthills #3520
Conversation
Co-authored-by: John Bytheway <[email protected]>
Previously this would fail with an error which didn't explain where the problem was.
Linear terrain (roads, ant tunnels, sewers, etc.) was not previously possible to define in JSON. This was because the rotation wasn't correctly calculated for linear terrain. Make it be.
Co-authored-by: John Bytheway <[email protected]>
When placement of mutable overmap specials fails, it's hard to figure out why. The small amount of context previously given in the debugmsg wasn't enough. I did have some further debug printfs in non-compiled code behind a #if, but that's no use to people who can't compile the game themselves, nor does it help debug failures in CI. Reformulate the code so that these more detailed messages are always saved (rather than printed) and when a placement error occurs, it includes a full history of the placement to allow for debugging.
I think I figured out why mutable special placement was sometimes failing. Some joins could be postponed but others could appear pointing at the same tile and be satisfied in the absence of the postponed joins. Then, when the postponed joins are restored they cannot be satisfied. Avoid this by keeping track of the tiles with postponed joins and ensuring that they new joins pointing at the same tiles are added directly to the postponed list rather than the unresolved list.
The previous attempt to fix the rare anthill placement errors did not work. I now know this was due to a simple typo, but tracking down the issue took a lot of refactoring. That refactoring improved the code, so I'm including it here. Changes include: * The previous fix was that new unresolved joins that might conflict with postponed joins would themselves be automatically postponed. We have now reversed this logic so the postponed joins are restored. This allows another chance for the current phase to satisfy the joins, now that might be possible. * Postponed joins now need to have their positions indexed for efficient access, so I factored out that indexing capability into a new struct indexed_joins. * Removed the list of orphaned joins. They should no longer ever occur. * Added a bunch of calls to a consistency_check function that verifies the class invariants (except it's disabled for now, with the code left in case it's useful for future debugging). * The order of unresolved joins used to depend on their addresses in memory, which made the tests non-reproducible. Change that to be deterministic. * Add unit test which places ~100k anthills to more easily observe placement errors. * Improve the debugging output for failed placements to make it easier to see what happened (add phase boundary markers and capitalize FAILED).
Rather than always having a fixed number of overmaps for a particular rule during placement, allow a random number. Currently this can only be a poisson-distributed number (or a fixed value, as was already supported).
Explain how to specify that the max number of instances should be drawn from a Poisson distribution.
To add more flexibility to the layout of mutable specials it is convenient to allow some joins for particular overmaps to be optional. This helps in particular with the final phase when you want to clean up all the leftover unresolved joins. Generalize the definition of joins to allow this in the JSON, and handle appropriately in the code. Convert craters to use this feature. Extend the unit tests to spawn a test crater to exercise this feature.
Sometimes it's important for the two sides of a join to be different, for example a hallway_to_room join should not join two hallways. It is also be important if the set of locations is different for the two sides. Add support for this type of asymmetry, and use it to allow anthills to have multiple entrances. Update test anthill also.
The Windows stack size limit is smaller than Linux, so need to tweak this test to avoid a stack overflow.
To be able to use these types in regular mapgen they need to be promoted to headers.
When a mutable special is placed, keep a record of every join used and store that in the overmap. Display this information in the overmap editor to aid with debugging.
Similarly to neighbourhood checks in chunk placement, add join checks to chunk placement.
* Acid Anthills * Update ants.json * Update ants.json
When placement fails for a mutable overmap special, add a note to the overmap with some details to assist debugging.
There used to be a function called direction_XY that converted a direction to a point. This didn't match the naming scheme used for analogous functions for om_direction and cube_direction. Rename it to displace_XY and also create a tripoint version displace.
Previously the mapgen feature of placing chunks based on neighbours used the absolute cardinal directions on the map. This is unhelpful when the map is rotated. Rotate the neighbours to match the rotation of the map so that these checks make sense for the mapgen.
For symmetry with the - overload.
Previously these checks were based on absolute map directions. After this change they are instead rotated with the map, so that they are easier to use logically within the map.
It can be useful to know the rotation of a particular tile for debugging purposes. This may not be obvious for linear terrain. Add it to the debugging output in the overmap editor.
It will help a lot with mutable overmap special design if a rule can place multiple overmaps together in a chunk. This attempts to implement that. It entails a significant rewrite of the guts of overmap special placement. Add a test_microlab special to test this new feature.
Vollch
force-pushed
the
mutable_specials
branch
from
6607040 to
d1c8830
Compare
src/mapgen.cpp
Outdated
| /* TODO: Throws errors on central lab depot, presumably due to loading order | ||
| for( const std::set<oter_str_id> &p : neighbors ) { | ||
| for( const oter_str_id &id : p ) { | ||
| if( !id.is_valid() ) { | ||
| debugmsg( "Invalid oter_str_id '%s' in %s", id.str(), oter_name ); | ||
| } | ||
| } | ||
| }*/ |
There was a problem hiding this comment.
Not quite the load order issue; this check happens at the "check consistency" stage and by then all JSONs have been loaded and all copy-froms have been resolved.
This legitimately triggers on subway because subway is not an overmap terrain, but rather overmap terrain type. The overmap terrains belonging to this type will have suffixes like _ns, _esw, _nesw, etc.
The issue comes from how this class misuses oter_str_id in the first place. In the test method below, it doesn't use the id as an id, it casts it back to std::string for pattern matching, which voids the whole reason behind having it be typesafe id. This should either all be std::strings, or be switched to oter_type_str_id and the code in test should check if terrain belongs to this type (e.g. by passing ot_match_type::type to is_ot_match).
There was a problem hiding this comment.
Good catch. That's actually my derp, original PR changed oter_str_id to oter_type_str_id, but mapgen.cpp have so many conflict all around(due do missing parametric mapgens) that it slipped by me during resolving. And the fact that DDA eventually removed that check() hinted me in wrong direction. I just confirmed it works at all(thanks to test()'s string casting) and called it a day :)
There was a problem hiding this comment.
Turns out that it can't really check for ot_match_type::type - It breaks oter checks in central labs, as they share same mapgens with regular labs, but have different types
There was a problem hiding this comment.
Suggestions here are mostly me hunting for typos and opportunities to elaborate on the documentation.
From the testing, the system is quite robust, the unique specials like the refugee center or the endgame lab are placed even with wacky density/spacing settings, so people running with 9999 mods should be fine.
The subways have changed somewhat, on z=-4 the labs often can't quite connect to each other, but other times it properly generates the usual "central lab <-> secondary labs" network. There's also that single rail line that ends up in a dead end under a random city. But then again, subway gen has many other glaring issues (tunnels opening into lake, rails not matching on diagonals, sharp turns making vehicles ram the walls...)
Something's weird seems to be going on with the lake specials, on default settings and on x5 density I get maybe a lighthouse and a marine on overmaps with substantial lakes.
| std::poisson_distribution<int> dist( mean ); | ||
| return string_format( "Poisson(%.0f)", dist.mean() ); |
There was a problem hiding this comment.
?
| std::poisson_distribution<int> dist( mean ); | |
| return string_format( "Poisson(%.0f)", dist.mean() ); | |
| return string_format( "Poisson(%.0f)", mean ); |
There was a problem hiding this comment.
std::poisson_distribution::mean
double mean() const;
https://en.cppreference.com/w/cpp/numeric/random/poisson_distribution/mean
lakes always was pretty empty, and normalization to lake\land ratio reduced occurrences even lower. Plus, lake specials for some weird reason depends on cities, even islands. That restricts their placement, and even more with current distribution of specials between different cities: at x5 density 0-1 lighthouse becomes 0-5, but then there's 5+ cities on overmap, and mapgen tries to put one lighthouse per city, yet not all cities have a lake around them to actually place it. |
|
As for subways. z-4 central labs are intendedly allowed to spawn with no depot. It can be changed to json side, if needed. Like i mentioned old connection rules was pretty nuanced, and while it's possible to replicate it'd be pretty messy - involving om_pos checks on all regular labs, and i'm not really sure whether i need to port all those convoluted nuances or not. For now it just spawns For subways in general, those dead ends due to how subways linked to each others. Specials S1 and S2 links to city C1, then S3 links to C2, then C1 links to C2. Now we have everything in one global network, but those C1 and C2 may have a dead ends if they reused old tracks. I'm not perfectly happy with that solution, but it just works, and i didn't came up with something more elegant so far. Edit: Finally got my hands to rework subways. No more dead ends. |
Vollch
force-pushed
the
mutable_specials
branch
from
f211367 to
a4fe122
Compare
|
Regarding disconnected things. Well, it shouldn't happen, obviously :) Rechecking generation i noticed that i missed one old connector under city mall, fixed now. And stumbled on microlabs spawned right under city center(c1 and c2 points as per pic above), preventing to link there - this one also should get better, by reusing road's fallback for no cities. But still no guarantees, due to limited search range. Edit: ...and yeah. This problem is gone now, after adding new connections type for subways. |
Vollch
force-pushed
the
mutable_specials
branch
from
a4fe122 to
c80817b
Compare
|
And also fixed one of those old "glaring issues" with rails connections. Chunks should check for actual connection, not just any subway. After: |
Vollch
force-pushed
the
mutable_specials
branch
from
39c1082 to
5a6f0a4
Compare
Vollch
force-pushed
the
mutable_specials
branch
from
5a59302 to
8c451c4
Compare
Vollch
force-pushed
the
mutable_specials
branch
from
3ac93db to
cf2e3a1
Compare
07735ca
Summary
SUMMARY: Infrastructure "Ported mutable specials from DDA, and unhardcored labs and anthills"
Purpose of change
Allows porting mutable specials from DDA, adding new ones, and also solves bunch of issues caused by hardcoded anthills\labs, such as type mismatch or building over subways.
Describe the solution
This PR contains both ported things, and new one. As i'm not fully satisfied with current DDA implementation to leave original changes as is.
First, ported PRs:
What was made on top of that:
Describe alternatives you've considered
Specials should get native nesting support to fully get rid of hardcode in labs.Done.Connections could be improved to support different layouts on top of hooking everything to city positions.Done.Rules of when central lab generate subways z-4 might need to be tweaked, as current implementation doesn't fully replicated old nuanced behavior.
Testing
Generated bunch of overmaps, ran mutable generation and endgame spawning tests, looks stable.
Additional context
Nothing to showcase really. Most of efforts were focused on making it look as similar to current labs\anthills as possible.