a-b-street/abstreet
1
1
Fork 0
mirror of https://github.com/a-b-street/abstreet.git synced 2024-11-25 11:44:25 +03:00
Code Issues Projects Releases Wiki Activity
5,312 Commits 85 Branches 166 Tags 395 MiB
7854823c28
Commit Graph

1295 Commits

Author SHA1 Message Date
Dustin Carlino
3583d068a6 Fix a bug introduced in e3b9c42506 on the proposal saver, and allow blocking-the-box on another Ballard intersection 2020-11-11 13:20:56 -08:00
Dustin Carlino
e5ff76e00c Working on gridlock in the Ballard map. Allow blocking-the-box near the 
famous Green Lake intersection, and add a little debug tool to see
current queue reservation length.
 2020-11-11 12:43:28 -08:00
Dustin Carlino
07ac288e72 Refactor: rename Pt2D::from_gps to just gps.to_pt(bounds) 2020-11-10 12:00:43 -08:00
Dustin Carlino
e15efa9002 Organize raw maps by city. With this, data/input/ is totally namespaced by city! Nothing prevents us from having a map named "downtown" in multiple cities now. #326 2020-11-05 14:58:28 -08:00
Dustin Carlino
cda586dfc9 Cleanup by moving some methods to belong to MapName. #326 2020-11-05 14:18:13 -08:00
Dustin Carlino
4f2cda8d53 Reorganize data/system/ files by city! The only interesting change is 
abstutil::path_all_maps becoming abstutil::list_all_maps. #326
 2020-11-05 14:05:48 -08:00
Dustin Carlino
67530bec07 Future-proof file organization by changing map_name from a string to a 
struct. Whatever choices we make next about naming cities hierarchially
or not can be managed in just one place. #326

This is a pretty huge change, but the compiler gives reasonable
confidence it's correct. More bugs are likely to crop up in the next
step, when filenames start being namespaced by the city too.
 2020-11-04 17:26:32 -08:00
Dustin Carlino
3ed7e73aa8 Use a u64 RNG seed, not a u8 2020-10-29 09:25:07 -07:00
Dustin Carlino
3b838c1225 Finish formatting distance/duration/speed units nicely. And remove the 
number of lanes from trip info; it's misleading and not useful. #331
 2020-10-28 11:33:48 -07:00
Dustin Carlino
9dbb156058 Skip wakeup_waiting while we're handling live map edits. Based on the 
order of vehicle deletion, we try to ask about stop sign and traffic
signal movements that may be deleted and would be irrelevant anyway if a
different deletion order happened. #312

We should really defer wakeup_waiting until after all cleanup is done,
but for now, willing to risk some stuckness at a stop sign...
 2020-10-23 12:15:51 -07:00
Dustin Carlino
3db8fe6669 Also need to clean up vehicles that're laggy heads on live-edited parts 
of the map. #312
 2020-10-23 11:58:43 -07:00
Dustin Carlino
781604907c While deleting cars for live edits, don't opportunistically lane-change for woen-up followers. They're likely about to be deleted, too. #312 2020-10-23 11:38:21 -07:00
Dustin Carlino
5973da0717 Handle bike trips that wind up with no actual biking portion due to live 
edits. #312

... And with this, traffic seitan makes it through an entire day at
montlake! Next up, lakeslice I guess.
 2020-10-22 21:39:54 -07:00
Dustin Carlino
773176f321 Handle live edits affecting multiple cars at one stop sign, and cars 
turning to/from an edited lane. #312
 2020-10-22 21:04:12 -07:00
Dustin Carlino
50c9c3236d Deal with a race condition. #312 
1) A car tries to spawn, but fails because there's something in the way
2) The player makes live edits
3) The retry occurs, but the path has become invalid due to edits

Need to make the detection of this more efficient later.
 2020-10-22 19:53:49 -07:00
Dustin Carlino
058103b84f Cars in the process of parking were being double-counted for deletion in 
live edits! #312
 2020-10-22 14:28:16 -07:00
Dustin Carlino
22ae06c3a4 Remove the --pathfinding_upfront option. #312 and #258 
Originally, all trips in the entire scenario had their paths calculated
at the beginning of a simulation. The sum time is faster than
calculating them individually, because we could use multiple threads.
But a while ago, this was disabled by default to improve the startup
latency. Especially if a player isn't making it through an entire day
anyway, calculating all of the paths upfront is a waste and slows down
their initial experience.

Now I'm hitting all sorts of bugs with live map edits, because the map
change between initial planning and when a trip starts. Previously, the
PathRequest was calculated up-front and resolved when a trip starts. But
even this PathRequest can become stale. So now always calculate it when
a trip actually starts, looking at the current map. To do this sanely,
totally rip out support for --pathfinding_upfront.

If we really want it later for performance, we could add it back in, and
be very careful about detecting stale PathRequests and recomputing. But
for now, there's no use case for this, so it'd needlessly complicate the
code.
 2020-10-22 13:47:03 -07:00
Dustin Carlino
cd62ad26d1 Lazily fix up TripLeg when the exact starting position for somebody 
biking changes due to live map edits. #312
 2020-10-22 12:47:41 -07:00
Dustin Carlino
83c1c09936 Notify the player how many trips got oofed after live map edits. #312 2020-10-22 11:58:16 -07:00
Dustin Carlino
8e6aa08d59 Breakdown agent counts much more in the tooltips. #331 2020-10-21 14:26:59 -07:00
Dustin Carlino
3d1ca3af9e Handle live map edits that rip away parking while a car is actively parking in it. #312 2020-10-21 13:00:49 -07:00
Dustin Carlino
adcd0f65fe Clean up some of the trips state when abruptly deleting agents. #312 2020-10-21 12:32:14 -07:00
Dustin Carlino
3c81480550 Make sure DrivingSimState has the correct queues after live map edits. #312 2020-10-21 12:00:56 -07:00
Dustin Carlino
011ed8c7cf Handle agents deleted mid-turn. #312 2020-10-21 11:17:27 -07:00
Dustin Carlino
a579cb4f74 Add a sanity check to catch live map edit bugs faster. Deleted turns 
aren't getting cleaned up properly.

Also fix broken headless build, woops.
 2020-10-21 10:23:56 -07:00
Dustin Carlino
e9f9d3884b Simplify the traffic seitan tool's logging and make it easier to find 
the last savestate and edits of a crash.
 2020-10-21 09:59:17 -07:00
Dustin Carlino
4760d6b2cf Fix one bug with live edits, found using traffic seitan. Agents may be 
deleted in the middle of a turn.
 2020-10-21 09:40:45 -07:00
Dustin Carlino
8a1525ddfa Lift parsing of --scenario_modifiers to a more common place, so it can be done in run_scenario too. #368 2020-10-21 08:35:38 -07:00
Dustin Carlino
ebc632d9f8 Add a flag to skip sim analytics, to work even faster on large scenarios. #368 
On the downtown benchmark, now we drop from 96s to 88s. Since we don't
need to collect analytics to debug gridlock on this map, another huge
win!
 2020-10-20 16:38:28 -07:00
Dustin Carlino
45e6a79e7d Use a HashMap to store driving queues. Modest speedup on the downtown 
benchmark from ~101s to ~96s. #368
 2020-10-20 16:23:18 -07:00
Dustin Carlino
36c1ce02ac Fix how scenario modifiers cancel trips, from 
c435d3fce2. If we cancel one trip, have to
cancel all others after that.
 2020-10-20 15:39:27 -07:00
Dustin Carlino
d8910ad04f Build a quadtree of unzoomed agents while drawing, to later speedup mouseover. This seems to make performance fine in debug mode, so get rid of the flag guarding of this feature! [rebuild] 2020-10-20 11:38:38 -07:00
Dustin Carlino
262c48721b Track some very basic stats on how many events we're processing. #368 
Didn't run the benchmarks yet, but hopefully this should't slow things
down.
 2020-10-19 19:12:19 -05:00
Dustin Carlino
c435d3fce2 Consolidate two scenario modifiers, allowing all walking trips to be cancelled, for #368 2020-10-19 18:03:17 -05:00
Dustin Carlino
87d233299a Use a HashMap instead of a BTreeMap in Scheduler's internals. Doesn't 
affect determistic simulation, but yields a crazy speedup. #54

- 8 hours of downtown: 122s to 102s!!!
- prebake: 181s to 160s
 2020-10-19 16:30:41 -05:00
Michael Kirk
82827bd60b
perf: only allocate vector for inner_get_car_positions when necessary (#369) 2020-10-19 14:58:16 -05:00
Dustin Carlino
bb669b7c1a Dump a savestate in the run_scenario tool when interrupted. Useful for 
investigating why downtown has 45k live agents by 5pm... #368
 2020-10-19 14:52:26 -05:00
Dustin Carlino
0a64cda947 Distinguish cancellation reasons for capped trips better. #325 2020-10-17 09:39:37 -05:00
Dustin Carlino
0259750eb3 Add a flag to detect drivers who're destined to pass through a currently delayed area. #325 2020-10-17 09:22:49 -05:00
Dustin Carlino
32b11f03e8 Avoid cloning a car's path when creating it. #368 
- 8 hours of downtown: 135s down to 126s
- prebaking: still hovering around 3m4s, plus or minus a few seconds
 2020-10-16 13:12:19 -05:00
Dustin Carlino
e66b058439 Disable a simulation sanity check that happens frequently and hasn't had 
problems in a long time. Marginal speedup of a few seconds. #368
 2020-10-16 12:29:46 -05:00
Dustin Carlino
491f4bbffe Speed up the simulation by replacing BTreeMaps with a simple Vec lookup. #368 
No behavioral changes. Dramatic performance boost:

- 8 hours of downtown: 170s down to 135s
- prebaking: 3m33s down to 3m4s
 2020-10-16 12:10:05 -05:00
Dustin Carlino
422882c03b Ditch cpuprofiler in favor of cargo flamegraph. Add a tool that just 
runs a scenario headlessly, since that's easier than sending a few curl
commands to the API. #368
 2020-10-15 17:12:38 -05:00
Dustin Carlino
b3525f95a2 Enable profiling in the headless server. #368 
(Although actually, I'm liking cargo flamegraph better than cpuprofiler.
Might rip out the support for cpuprofiler.)
 2020-10-15 16:21:57 -05:00
Dustin Carlino
028d1df62c Allow selecting and clicking unzoomed agents. [rebuild] 2020-10-15 14:58:07 -05:00
Dustin Carlino
c0513a3d64 Reduce time to park/unpark offstreet when --infinite_parking is used. 
Downtown forms gridock just from everyone pouring into Harborview! The
point of --infinite_parking is to simplify the sim by not modelling the
impact of parking. #368
 2020-10-14 12:40:26 -05:00
Dustin Carlino
84a65c293b Make it easier to test scenario modifiers by specifying them on the CLI 2020-10-14 12:16:38 -05:00
Dustin Carlino
1906fb0913 Remove the GetDrawAgents indirection. Originally the UI could pull the 
set of agents to draw either from the simulation or from this
"time-traveler" plugin, which would let you rewind sim time. That plugin
is long gone, and it never worked well, because much of the UI would use
the GetDrawAgents to select something, then query the live sim for lots
of details anyway. The plugin never served all of those calls, so the
results would be kind of out-of-sync anyway.

There are some parts of the UI that need to temporarily not draw agents.
Use the same suspended_sim trick that edit mode does.
 2020-10-12 11:52:05 -05:00
Dustin Carlino
a1c53d5e1c Humble start to adjusting how distances and durations are printed in the UI. #331 
Also trimmed out a few old UI settings; they kind of started as
experiments, but now the alternative feels useless to maintain.
 2020-10-11 10:45:13 -07:00
Dustin Carlino
df873a5cf9 Add an option to totally disable turn conflicts in the simulation. Can we make progress on large scenarios only with vehicle queueing rules? (Didn't evaluate this yet, don't want to burn through laptop battery currently) 
Also fix zooming in on traffic signals crashing with freeform policy
 2020-10-11 10:14:48 -07:00