Economic Districts #6345

bunnybot · 2024-02-06T19:06:44Z

Mirrored from Codeberg
Created on Tue Feb 06 19:59:44 CET 2024 by Benedikt Straub (Nordfriese)

Type of change
New feature for v1.3

Issue(s) closed
This addresses several points that arise frequently in the context of very large games – namely the problem of wares (and also workers) being transported across long distances instead of preferring local supplies.

How it works
These points all happen in the background. The UI is unchanged.

Each economy automatically divides itself into districts. A district is centered around a warehouse (multiple warehouses in close proximity are clustered for this purpose), and each flag belongs to the closest warehouse. Thus, a district is a circle around a warehouse or cluster of warehouses.

Districts try to be self-sufficient and minimize imports.
Supply/request matching always prefers to use supplies from the same district. Imports are only accepted if no local supply is available.

New: All active long-distance imports are frequently monitored, and if a ware becomes available in the local district, the import is cancelled and replaced with the local ware. This type of supply exchanging was previously always rejected as too CPU-intensive, but this approach is performant and catches all the cases where it matters most.

Additionally, economy targets are distributed across districts. Global targets (global here meaning within one economy) continue to work as before. In addition, if a district is short of a specific ware, then the productionsites within this district keep producing even if the global stock is above target. Each district's local target is simply the global target divided by number of districts, rounding up.

Possible regressions
Economic request/supply matching, economy targets.

Screenshots

Additional context
Nothing here needs to be saveloaded. This is all recalculated lazily every few seconds.
Benchmark for a nearly completely conquered Accurate Europe 1.0 map, Release build:

[00:02:38.430 real] INFO: NOCOM Recalcing districts for 5376 flags and 34 warehouses took 7 ms
[00:02:38.434 real] INFO: NOCOM Checking imports for 5376 flags and 34 warehouses took 3 ms

Flags at the border between two districts tend to jitter between the two possible districts due to pseudo-randomness in the routing algorithm. I would not consider this a problem.

Could do with more testing, both with big economies and with small ones. I'd therefore like to have this early in the release cycle for v1.3.

bunnybot · 2024-02-06T19:06:50Z

Assigned to Nordfriese

bunnybot · 2024-02-20T14:17:23Z

Mirrored from Codeberg
On Tue Feb 20 15:14:18 CET 2024, ** (frankystone)* wrote:*

Does this change anything related to economy worker settings?

In the attached save game the barracks produce soldiers endlessly, 'stealing' the wares needed to create heroes. Starting this save game in master the soldier production stops, because economy setting says 10 soldiers and there are already more than 40 sitting around.

In debug builds the game stutters for me every time the district calculation is made. Probably because of the debug things.

bunnybot · 2024-02-20T14:50:26Z

Mirrored from Codeberg
On Tue Feb 20 15:48:29 CET 2024, Benedikt Straub (Nordfriese) wrote:

Ah yes…
if you have 4 warehouses and a soldier target setting of 10, then each warehouse's district target is ceil(10/4) = 3 soldiers. However the barracks is in the HQ district and the HQ's stock policy for soldiers is kRemove, so the barracks district can never reach its local target.

So for districts where a ware/worker is not meant to be stored in warehouses this check won't work. And if some other (foreign) warehouse has a Prefer policy for that ware/worker but we don't Prefer it locally then too it won't work.

I'll create a commit to take these stock policies into account and skip the local target check if the warehouse policies interfere with district target distribution.

In debug builds the game stutters for me every time the district calculation is made. Probably because of the debug things.

For me debug builds with such large maps stutter on every economy update anyway ;) What times does the log print out for district recalculation and imports checking? For me this is always 1ms each with your savegame in a debug build (i.e. negligibly fast).

bunnybot

Mirrored from Codeberg
On Tue Jun 11 15:45:41 CEST 2024, Tóth András (tothxa) commented:

I tested it in a game too and it seems to work nicely. Thanks a lot for solving this! :)

Some comments inline.

bunnybot · 2024-06-11T16:29:05Z

src/economy/flag.h

@@ -204,6 +216,10 @@ struct Flag : public PlayerImmovable, public RoutingNode {
 	Coords position_;
 	Time animstart_{0};

+	Warehouse* district_center_[2] = {
+	   nullptr,
+	   nullptr};  ///< Warehouse at the center of our district, indexed by WareWorker (may be null).


Mirrored from Codeberg
On Tue Jun 11 15:45:41 CEST 2024, Tóth András (tothxa) wrote:

Is indexing this directly by an unscoped enum safe?

bunnybot · 2024-06-11T16:29:05Z

src/economy/flag.cc

@@ -1015,6 +1018,19 @@ void Flag::log_general_info(const Widelands::EditorGameBase& egbase) const {

 	Widelands::PlayerImmovable::log_general_info(egbase);

+	if (district_center_[0] == nullptr) {


Mirrored from Codeberg
On Tue Jun 11 15:51:36 CEST 2024, Tóth András (tothxa) wrote:

Maybe wwWARE (and wwWORKER below) should be used for consistency?

bunnybot · 2024-06-11T16:29:05Z

src/economy/economy.cc

+				++target_district;  // Rounding up is important for wares with small targets
+			}
+		} else {
+			target_district = 0;


Mirrored from Codeberg
On Tue Jun 11 16:26:04 CEST 2024, Tóth András (tothxa) wrote:

Simply initialise as 0 for readability instead of this else clause?

bunnybot · 2024-06-11T16:29:05Z

src/economy/economy.cc

+
+			Flag& supply_flag = imm->base_flag();
+			Warehouse* supply_district = supply_flag.get_district_center(type());
+			if (supply_district == nullptr || supply_district == target_district) {


Mirrored from Codeberg
On Tue Jun 11 17:24:19 CEST 2024, Tóth András (tothxa) wrote:

Is there a valid case when supply_district can be nullptr? Doesn't that mean there's some inconsistency somewhere?

bunnybot · 2024-06-11T16:29:05Z

src/economy/economy.cc

+		wh->base_flag().set_district_center(type(), wh);
+		astar.push(wh->base_flag());
+	}
+	while (RoutingNode* current = astar.step()) {


Mirrored from Codeberg
On Tue Jun 11 17:34:44 CEST 2024, Tóth András (tothxa) wrote:

Isn't this an implicit cast to bool?

bunnybot · 2024-06-11T16:29:05Z

src/economy/economy.cc

+	constexpr uint32_t kClusterThreshold = 12 * 1800;  // 12 nodes on flat terrain.
+	std::map<std::pair<Warehouse*, Warehouse*>, uint32_t> warehouse_distances;
+
+	for (Flag* flag1 : flags_) {


Mirrored from Codeberg
On Tue Jun 11 18:02:55 CEST 2024, Tóth András (tothxa) wrote:

Why don't you only iterate over warehouses_?

bunnybot · 2024-06-11T16:29:05Z

src/economy/economy.cc

+			}
+		}
+
+		// Now find all existing clusters to merge this one with


Mirrored from Codeberg
On Tue Jun 11 18:21:07 CEST 2024, Tóth András (tothxa) wrote:

Do I get it right that if there's a long chain of warehouses where neighbouring ones are within the threshold, then all of them will get clustered even if the distance between the opposite ends is many times the threshold? Do we really need clusters at all?

bunnybot · 2024-06-11T17:22:08Z

src/economy/economy.cc

+			}
+
+			upcast(PlayerImmovable, imm, location);
+			if (imm == nullptr) {


Mirrored from Codeberg
On Tue Jun 11 19:19:35 CEST 2024, Benedikt Straub (Nordfriese) wrote:

Since districts are recalculated lazily, they can be nullptr for short periods during and after splits and merges and warehouse deletions. In this case the import will simply be caught on the next cycle, after the districts have been updated again.

bunnybot · 2024-06-11T17:22:08Z

src/economy/economy.cc

+		Cluster current_cluster = {current};
+		for (auto& pair : warehouse_distances) {
+			if (pair.first.first == current) {
+				current_cluster.push_back(pair.first.second);


Mirrored from Codeberg
On Tue Jun 11 19:17:28 CEST 2024, Benedikt Straub (Nordfriese) wrote:

The case I was thinking of was the strategy that some players use of building two or three warehouses in close proximity for training centres, one to store food, one to store weapons, etc. If each of these warehouses defined its own district then this would mess up all productionsites around this centre. This new feature should not break existing valid strategies and clustering was the solution I came up with.

A long chain would be clustered too, but IMHO a really long warehouse chain is an indication that the player is doing something highly unusual and unexpected and we can't really guess what he might wish to accomplish with this, except that such an economy seems to be tightly chain-linked and therefore may be meant to be treated as one unit.

bunnybot · 2024-06-11T18:38:29Z

src/economy/economy.cc

+		Cluster current_cluster = {current};
+		for (auto& pair : warehouse_distances) {
+			if (pair.first.first == current) {
+				current_cluster.push_back(pair.first.second);


Mirrored from Codeberg
On Tue Jun 11 20:34:08 CEST 2024, Tóth András (tothxa) wrote:

The case I was thinking of was the strategy that some players use of building two or three warehouses in close proximity for training centres, one to store food, one to store weapons, etc. If each of these warehouses defined its own district then this would mess up all productionsites around this centre.

Have you experienced that in testing before adding clusters?

bunnybot · 2024-06-11T23:16:11Z

src/economy/economy.cc

+		Cluster current_cluster = {current};
+		for (auto& pair : warehouse_distances) {
+			if (pair.first.first == current) {
+				current_cluster.push_back(pair.first.second);


Mirrored from Codeberg
On Wed Jun 12 01:12:10 CEST 2024, Tóth András (tothxa) wrote:

A long chain would be clustered too, but IMHO a really long warehouse chain is an indication that the player is doing something highly unusual and unexpected and we can't really guess what he might wish to accomplish with this, except that such an economy seems to be tightly chain-linked and therefore may be meant to be treated as one unit.

Or maybe it's just a player who took the advice to build warehouses a bit too overzealously. :) They shouldn't be punished for it. But I really don't know what we could do that's simple enough.

I also have a doubt about how we can decide what's an appropriate cluster threshold.

Noordfrees and others added 10 commits February 1, 2024 17:53

Divide economies into districts

c37e880

Base production needs and request/supply matching around districts

1dcd1b5

Cluster nearby warehouses

8179a90

Merge branch 'master' of codeberg.org:wl/widelands into districts

438a33e

Merge branch 'master' of codeberg.org:wl/widelands into districts

9cb7d42

Baseplate for import swapping

fe41e1c

Fix import recalculating for non-open requests

5bd2c53

Merge branch 'master' of codeberg.org:wl/widelands into districts

71be2fb

Bugfixes

f984a65

4 files were automatically formatted.

9089d28

bunnybot added this to the v1.3 milestone Feb 6, 2024

bunnybot changed the title ~~WIP: Economic Districts~~ Economic Districts Feb 6, 2024

bunnybot self-assigned this Feb 6, 2024

Noordfrees and others added 6 commits February 6, 2024 20:48

Typo

0d5086f

delete unused function

d1d2428

Economy splitting fixes

357e1d5

3 files were automatically formatted.

557a4fd

Merge branch 'master' of codeberg.org:wl/widelands into districts

5398aa3

'src/economy/economy.cc' was automatically formatted.

12d64e6

bunnybot changed the title ~~Economic Districts~~ WIP: Economic Districts Feb 11, 2024

Merge branch 'master' into districts

2a7e4c8

Merge branch 'master' of codeberg.org:wl/widelands into districts

901ef88

bunnybot added ci:fail CI checks failed and removed ci:fail CI checks failed labels May 17, 2024

Merge branch 'master' of codeberg.org:wl/widelands into districts

6c3226d

bunnybot added ci:success CI checks succeeded and removed ci:fail CI checks failed labels May 20, 2024

Merge branch 'master' into districts

3db23f2

bunnybot added ci:success CI checks succeeded and removed ci:success CI checks succeeded labels May 25, 2024

Merge branch 'master' into districts

8733454

bunnybot added ci:success CI checks succeeded and removed ci:success CI checks succeeded labels Jun 1, 2024

bunnybot commented Jun 11, 2024

View reviewed changes

Noordfrees and others added 3 commits June 11, 2024 19:00

Merge branch 'master' of codeberg.org:wl/widelands into districts

8afff36

Address review

9295e43

'src/economy/flag.cc' was automatically formatted.

880771e

bunnybot commented Jun 11, 2024

View reviewed changes

bunnybot removed the ci:success CI checks succeeded label Jun 11, 2024

bunnybot commented Jun 11, 2024

View reviewed changes

bunnybot added the ci:success CI checks succeeded label Jun 11, 2024

bunnybot commented Jun 11, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Economic Districts #6345

Economic Districts #6345

bunnybot commented Feb 6, 2024 •

edited

bunnybot commented Feb 6, 2024

bunnybot commented Feb 20, 2024

bunnybot commented Feb 20, 2024 •

edited

bunnybot left a comment

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

bunnybot Jun 11, 2024

		@@ -1015,6 +1018,19 @@ void Flag::log_general_info(const Widelands::EditorGameBase& egbase) const {

		Widelands::PlayerImmovable::log_general_info(egbase);

		if (district_center_[0] == nullptr) {

Economic Districts #6345

Are you sure you want to change the base?

Economic Districts #6345

Conversation

bunnybot commented Feb 6, 2024 • edited

bunnybot commented Feb 6, 2024

bunnybot commented Feb 20, 2024

bunnybot commented Feb 20, 2024 • edited

bunnybot left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bunnybot commented Feb 6, 2024 •

edited

bunnybot commented Feb 20, 2024 •

edited