Tarrasque is a library, build around Skadi, to allow the easy and straightforward analysis of Dota 2 replays. While Skadi provides only the raw data, Tarrasque allows you to deal in objects and relationships. A comparison will show this best.
This code uses Skadi to print out the names of the players in the replay, along with the name of the hero they are playing
import io
from skadi.engine import world as w
from skadi.replay import demo as rd
demo = rd.construct("./demo.dem")
for tick, string_tables, world in demo.stream(tick=5000):
ehandle, player_resource = world.find_by_dt(player_resource_dt)
for i in range(31):
player_name_key = ("DT_DOTA_PlayerResource", "m_iszPlayerNames.%40s" % i)
player_name = player_resource[player_name_key]
if not player_name:
break
hero_ehandle_key = ("DT_DOTA_PlayerResource", "m_hSelectedHero")
hero_ehandle = player_resource[hero_ehandle_key]
hero_dt = world.recv_tables[world.classes[hero_ehandle]].dt
hero_name = hero_dt.replace("DT_DOTA_Unit_Hero_", "").replace("_", " ")
print hero_name
break
Using Tarrasque, this could be written as
import tarrasque
replay = tarrasque.StreamBinding.from_file("demo.dem")
for player in replay.players:
print player.name
print player.hero.name
The code speaks for itself. Tarrasque makes it simple, easy and even fun to analyse Dota 2 replays.
Tarrasque is a mapper between Dota2 entities (DT classes) and Python classes. Every Tarrasque class that represents an entity has a dt_key property that specifies the DT class that it represents, and once instantiated, every Tarrasque class has a ehandle property that is used to get information from the world. The current world can be accessed via world, and the results of world.find(self.ehandle) via properties. All this and more is documented on DotaEntity.
Think of Tarrasque as an ORM for Dota2, except the models are already maintained, and you don’t have to worry about the database. You don’t have to mess about writing code to deal with the (disgusting) stuff that Dota2 does in its replays, as Tarrasque exposes the data to you in a manner that follows Python conventions; you’ll get a None object instead of -1, and the string "radiant" instead of the integer 2 (where appropriate. Tarrasque understands that values have special meanings only in specific contexts). This allows you to just use the data, and not need to worry about the stuff underneath.
The one major difference between a database ORM and Tarrasque is that while most ORM models are statefull (that is, when the database changes, the model stays the same until reloaded), Tarrasque models contain no state, other than that which is needed to uniquely identify the instance (and now you know what an ehandle is). This means that you never have to do hero.update(tick_number) or similar; all that is handled automatically via the StreamBinding/ DotaEntity abstraction.