[freeciv-ai] (PR#10216) AI Builds Too Many Transports

["Gregory Berkolaiko" <Gregory.Berkolaiko@xxxxxxxxxxxxx>]

<URL: http://rt.freeciv.org/Ticket/Display.html?id=10216 >

> [badamson@xxxxxxxxxxx - Mon Oct 04 01:19:11 2004]:
> 
> I've done some further investigation of this problem. Initially I'd 
> thought that the city code for emigrating settlers (function 
> domestic_advisor_choose_build) was building transports for colonists too 
> often. However, debugging with additional  assertions and logging code 
> suggests that the military advisor is making the decision to build the 
> numerous transports.
> 
> What seems to happen is this:
> 
> 1) The military advisor (function military_advisor_choose_build) 
> considers building a naval attacker (for example, a battleship) by 
> calling ai_choose_attacker(pcity, SEA_MOVING).
> 
> 2) When the military adviser examines whether it can use that attacker 
> (by calling function kill_something_with), the code can select a land 
> unit as the target. That is, it can choose a naval bombardment.
> 
> 3) That code (kill_something_with)) then checks whether there is a path 
> from the city (were the attacker will begin) and the target (the enemy 
> land unit to be bombarded), using function goto_is_sane. That call is 
> used to decide whether a land attacker needs a ferry; a land attacker 
> will need a ferry if there is no valid (land) path from the city to the 
> target. However, the function is also used for sea attackers. For those 
> units, the kill_something_with function incorrectly concludes that the 
> (naval!) attack unit will need a ferry (go_by_boat becomes TRUE). I 
> guess it is not handling the 'overlap' needed for shore bombardment 
> properly.
> 
> 4) Consequently, the military adviser decides to build a transport for 
> some (many?) cases when it should build a naval attacker.
> 
> Note that there is no feed back loop here: the completion of an 
> additional transport will not diminish the desire to build more
transports.
> 
> Does this analysis seem plausible? I'm investigating further, and 
> testing a candidate fix.

Well done, Benedict!  Just to confirm your analysis, here is a small
patch which fails my autogame pretty soon.

I will now investigate why goto_is_sane is failing us.

G.

? core.10423
? core.10618
Index: ai/advmilitary.c
===================================================================
RCS file: /home/freeciv/CVS/freeciv/ai/advmilitary.c,v
retrieving revision 1.174
diff -u -r1.174 advmilitary.c
--- ai/advmilitary.c    29 Sep 2004 02:24:18 -0000      1.174
+++ ai/advmilitary.c    4 Oct 2004 03:44:15 -0000
@@ -1110,6 +1110,10 @@
                           &best_choice, NULL, U_LAST);
   } else { 
     /* Attract a boat to our city or retain the one that's already here */
+    if (!is_ground_unit(myunit)) {
+      UNIT_LOG(LOG_ERROR, myunit, "Wants a boat, ridiculous!!");
+    }
+    assert(is_ground_unit(myunit));
     best_choice.need_boat = TRUE;
     process_attacker_want(pcity, benefit, def_type, def_vet, ptile, 
                           &best_choice, ferryboat, boattype);

auto.rc
Description: Binary data