{- |

Module : $Header$

Description : heterogeneous signatures colimits approximations

Copyright : (c) Mihai Codescu, and Uni Bremen 2002-2006

License : GPLv2 or higher, see LICENSE.txt

Maintainer : mcodescu@informatik.uni-bremen.de

Stability : provisional

Portability : non-portable

Heterogeneous version of weakly_amalgamable_cocones.

Needs some improvements (see TO DO).

-}

module Static.WACocone(isConnected,

isAcyclic,

isThin,

removeIdentities,

hetWeakAmalgCocone,

initDescList,

dijkstra,

buildStrMorphisms,

weakly_amalgamable_colimit

) where

import Control.Monad

import Data.List(nub)

import qualified Data.Map as Map

import qualified Data.Set as Set

import Data.Graph.Inductive.Graph as Graph

import Common.Lib.Graph as Tree

import Common.ExtSign

import Common.Result

import Common.LogicT

import Logic.Logic

import Logic.Comorphism

import Logic.Modification

import Logic.Grothendieck

import Logic.Coerce

import Static.GTheory

import Comorphisms.LogicGraph

weakly_amalgamable_colimit :: StaticAnalysis lid

basic_spec sentence symb_items symb_map_items

sign morphism symbol raw_symbol

=> lid -> Tree.Gr sign (Int, morphism)

-> Result (sign, Map.Map Int morphism)

weakly_amalgamable_colimit l diag = do

(sig, sink) <- signature_colimit l diag

return (sig, sink)

-- until amalgamability check is fixed, just return a colimit

-- get (commented out) code from rev:11881

-- | checks whether a graph is connected

isConnected :: Gr a b -> Bool

isConnected graph = let

nodeList = nodes graph

root = head nodeList

availNodes = Map.fromList $ zip nodeList (repeat True)

bfs queue avail = case queue of

[] -> avail

n:ns -> let

avail1 = Map.insert n False avail

nbs = filter (\x -> Map.findWithDefault (error "isconnected") x avail)

$ neighbors graph n

in bfs (ns++nbs) avail1

in filter (\x -> Map.findWithDefault (error "iscon 2") x

(bfs [root] availNodes)) nodeList == []

-- | checks whether the graph is thin

isThin :: Gr a b -> Bool

isThin graph = checkThinness (Map.empty) $ edges graph

checkThinness :: Map.Map Edge Int -> [Edge] -> Bool

checkThinness paths eList =

case eList of

[] -> True

(sn,tn):eList' ->

if (sn,tn) `elem` (Map.keys paths) then

False -- multiple paths between (sn, tn)

else

let pathsToS = filter (\(_,y) -> y == sn) $ Map.keys paths

updatePaths pathF dest pList =

case pList of

[] -> Just pathF

(x,_):pList' ->

if (x,dest) `elem` (Map.keys pathF) then Nothing

else updatePaths (Map.insert (x,dest) 1 pathF) dest pList'

in case updatePaths paths tn pathsToS of

Nothing -> False

Just paths' -> checkThinness (Map.insert (sn,tn) 1 paths') eList'

-- | checks whether a graph is acyclic

isAcyclic :: (Eq b) => Gr a b -> Bool

isAcyclic graph = let

filterIns gr = filter (\ x -> indeg gr x == 0)

queue = filterIns graph $ nodes graph

topologicalSort q gr = case q of

[] -> null $ edges gr

n : ns -> let

oEdges = lsuc gr n

graph1 = foldl (flip Graph.delLEdge) gr

$ map (\ (y, label) -> (n, y, label)) oEdges

succs = filterIns graph1 $ suc gr n

in topologicalSort (ns ++ succs) graph1

in topologicalSort queue graph

-- | auxiliary for removing the identity edges from a graph

removeIdentities :: Gr a b -> Gr a b

removeIdentities graph = let

addEdges gr eList = case eList of

[] -> gr

(sn, tn, label):eList1 -> if sn == tn then addEdges gr eList1

else addEdges (insEdge (sn, tn, label) gr) eList1

in (addEdges $ insNodes (labNodes graph) Graph.empty)

$ labEdges graph

-- assigns to a node all proper descendents

initDescList :: (Eq a, Eq b) => Gr a b -> Map.Map Node [(Node, a)]

initDescList graph = let

descsOf n = let

nodeList = filter (\x -> x /= n) $ pre graph n

f = Map.fromList $ zip nodeList (repeat False)

precs nList nList' avail =

case nList of

[] -> nList'

_ -> let

nList'' = concat $

map (\y -> filter (\x -> if x `elem` Map.keys avail

then

Map.findWithDefault (error "iDL")

x avail

else True ) $

filter (\x -> x /= y) $

pre graph y) nList

avail' = Map.union avail $

Map.fromList $ zip nList'' (repeat False)

in precs (nub nList'') (nub $ nList' ++ nList'') avail'

in precs nodeList nodeList f

in Map.fromList$ map (\node -> (node, filter (\x->(fst x) `elem`

descsOf node)

$ labNodes graph )) $ nodes graph

commonBounds :: (Eq a) => Map.Map Node [(Node, a)] -> Node -> Node -> [(Node,a)]

commonBounds funDesc n1 n2 = filter

(\x -> x `elem` ((Map.!) funDesc n1) && x `elem` ((Map.!) funDesc n2) )

$ nub $ (Map.!) funDesc n1 ++ (Map.!) funDesc n2

-- returns the greatest lower bound of two maximal nodes,if it exists

glb :: (Eq a) => Map.Map Node [(Node, a)] -> Node -> Node -> Maybe (Node,a)

glb funDesc n1 n2 = let

cDescs = commonBounds funDesc n1 n2

subList [] _ = True

subList (x:xs) l2 = x `elem` l2 && subList xs l2

glbList = filter (\(n, x) -> subList

(filter (\(n0,x0) -> (n,x)/= (n0,x0)) cDescs) (funDesc Map.! n)

) cDescs

-- a node n is glb of n1 and n2 iff

-- all common bounds of n1 and n2 are also descendants of n

in case glbList of

[] -> Nothing

x:_ -> Just x -- because if it exists, there can be only one

-- if no greatest lower bound exists, compute all maximal bounds of the nodes

maxBounds :: (Eq a) => Map.Map Node [(Node, a)] -> Node -> Node -> [(Node, a)]

maxBounds funDesc n1 n2 = let

cDescs = commonBounds funDesc n1 n2

isDesc n0 (n,y) = (n,y) `elem` funDesc Map.! n0

noDescs (n,y) = filter (\(n0, _) -> isDesc n0 (n,y)) cDescs == []

in filter noDescs cDescs

-- dijsktra algorithm for finding the the shortest path between two nodes

dijkstra :: GDiagram -> Node -> Node -> Result GMorphism

dijkstra graph source target = do

let

dist = Map.insert source 0 $ Map.fromList $

zip (nodes graph) $ repeat $ 2 * (length $ edges graph)

prev = if source == target then Map.insert source source Map.empty

else Map.empty

q = nodes graph

com = case lab graph source of

Nothing -> Map.empty --shouldnt be the case

Just gt -> Map.insert source (ide $ signOf gt) Map.empty

extractMin queue dMap = let

u = head $

filter (\x -> Map.findWithDefault (error "dijkstra") x dMap ==

(minimum $

map (\x1 -> Map.findWithDefault (error "dijkstra") x1 dMap)

queue))

queue

in ( Set.toList $ Set.difference (Set.fromList queue) (Set.fromList [u]) , u)

updateNeighbors d p c u gr = let

outEdges = out gr u

upNeighbor dMap pMap cMap uNode edgeList = case edgeList of

[] -> (dMap, pMap, cMap)

(_, v, (_, gmor)):edgeL -> let

alt = ( Map.findWithDefault (error "dijkstra") uNode dMap) + 1

in

if (alt >= Map.findWithDefault (error "dijsktra") v dMap) then

upNeighbor dMap pMap cMap uNode edgeL

else let

d1 = Map.insert v alt dMap

p1 = Map.insert v uNode pMap

c1 = Map.insert v gmor cMap

in upNeighbor d1 p1 c1 uNode edgeL

in upNeighbor d p c u outEdges

-- for each neighbor of u, if d(u)+1 < d(v), modify p(v) = u, d(v) = d(u)+1

mainloop gr sn tn qL d p c = let

(q1, u) = extractMin qL d

(d1, p1, c1) = updateNeighbors d p c u gr

in if (u == tn) then shortPath sn p1 c1 [] tn

else mainloop gr sn tn q1 d1 p1 c1

shortPath sn p1 c s u =

if not $ u `elem` Map.keys p1 then fail "path not found"

else let

x = Map.findWithDefault (error $ show u) u p1 in

if x == sn then return (u:s, c)

else shortPath sn p1 c (u:s) x

(nodeList, com1) <- mainloop graph source target q dist prev com

foldM comp ((Map.!) com1 source) . map ((Map.!) com1) $ nodeList

-- builds the arrows from the nodes of the original graph

-- to the unique maximal node of the obtained graph

buildStrMorphisms :: GDiagram -> GDiagram

->Result (G_theory, Map.Map Node GMorphism)

buildStrMorphisms initGraph newGraph = do

let (maxNode, sigma) = head $ filter (\(node,_) -> outdeg newGraph node == 0) $

labNodes newGraph

buildMor pairList solList = do

case pairList of

(n, _):pairs -> do nMor <- dijkstra newGraph n maxNode

buildMor pairs (solList ++ [(n,nMor)])

[] -> return solList

morList <- buildMor (labNodes initGraph) []

return $ (sigma, Map.fromList morList)

-- computes the colimit and inserts it into the graph

addNodeToGraph :: GDiagram -> G_theory -> G_theory -> G_theory -> Int -> Int

-> Int -> GMorphism -> GMorphism

-> Map.Map Node [(Node, G_theory)] -> [(Int, G_theory)]

-> Result (GDiagram, Map.Map Node [(Node, G_theory)])

addNodeToGraph oldGraph

(G_theory lid extSign _ _ _)

gt1@(G_theory lid1 extSign1 idx1 _ _)

gt2@(G_theory lid2 extSign2 idx2 _ _)

n

n1

n2

(GMorphism cid1 ss1 _ mor1 _)

(GMorphism cid2 ss2 _ mor2 _)

funDesc maxNodes = do

let newNode = 1 + (maximum $ nodes oldGraph) --get a new node

s1 <- coerceSign lid1 lid "addToNodeGraph" extSign1

s2 <- coerceSign lid2 lid "addToNodeGraph" extSign2

m1 <- coerceMorphism (targetLogic cid1) lid "addToNodeGraph" mor1

m2 <- coerceMorphism (targetLogic cid2) lid "addToNodeGraph" mor2

let spanGr = Graph.mkGraph

[(n, plainSign extSign), (n1, plainSign s1), (n2, plainSign s2)]

[(n, n1, (1, m1)), (n, n2, (1, m2))]

(sig,morMap) <- weakly_amalgamable_colimit lid spanGr

-- must coerce here

m11 <- coerceMorphism lid (targetLogic cid1) "addToNodeGraph" $

morMap Map.! n1

m22 <- coerceMorphism lid (targetLogic cid2) "addToNodeGraph" $

morMap Map.! n2

let gth = noSensGTheory lid (mkExtSign sig) startSigId

gmor1 = GMorphism cid1 ss1 idx1 m11 startMorId

gmor2 = GMorphism cid2 ss2 idx2 m22 startMorId

case maxNodes of

[] -> do

let newGraph = insEdges [(n1, newNode,(1, gmor1)),(n2, newNode,(1,gmor2))] $

insNode (newNode, gth) oldGraph

funDesc1 = Map.insert newNode

(nub $ (Map.!)funDesc n1 ++ (Map.!) funDesc n2 ) funDesc

return (newGraph, funDesc1)

_ -> computeCoeqs oldGraph funDesc (n1, gt1) (n2, gt2)

(newNode, gth) gmor1 gmor2 maxNodes

-- for each node in the list, check whether the coequalizer can be computed

-- if so, modify the maximal node of graph and the edges to it from n1 and n2

computeCoeqs :: GDiagram -> Map.Map Node [(Node, G_theory)]

-> (Node,G_theory) -> (Node,G_theory) -> (Node, G_theory)

-> GMorphism -> GMorphism -> [(Node, G_theory)]->

Result (GDiagram, Map.Map Node [(Node, G_theory)])

computeCoeqs oldGraph funDesc (n1,_) (n2,_) (newN, newGt) gmor1 gmor2 [] = do

let newGraph = insEdges [(n1, newN, (1, gmor1)),(n2, newN, (1, gmor2))] $

insNode (newN, newGt) oldGraph

descFun1 = Map.insert newN

(nub $ (Map.!)funDesc n1 ++ (Map.!) funDesc n2 ) funDesc

return $ (newGraph, descFun1)

computeCoeqs graph funDesc (n1,gt1) (n2,gt2)

(newN, _newGt@(G_theory tlid tsign _ _ _))

_gmor1@(GMorphism cid1 sig1 idx1 mor1 _ )

_gmor2@(GMorphism cid2 sig2 idx2 mor2 _ ) ((n,gt):descs)= do

_rho1@(GMorphism cid3 _ _ mor3 _)<- dijkstra graph n n1

_rho2@(GMorphism cid4 _ _ mor4 _)<- dijkstra graph n n2

com1 <- compComorphism (Comorphism cid1) (Comorphism cid3)

com2 <- compComorphism (Comorphism cid1) (Comorphism cid3)

if com1 /= com2 then fail "Unable to compute coequalizer" else do

_gtM@(G_theory lidM signM _idxM _ _)<- mapG_theory com1 gt

s1 <- coerceSign lidM tlid "coequalizers" signM

mor3' <- coerceMorphism (targetLogic cid3) (sourceLogic cid1) "coeqs" mor3

mor4' <- coerceMorphism (targetLogic cid4) (sourceLogic cid2) "coeqs" mor4

m1 <- map_morphism cid1 mor3'

m2 <- map_morphism cid2 mor4'

phi1' <- comp m1 mor1

phi2' <- comp m2 mor2

phi1 <- coerceMorphism (targetLogic cid1) tlid "coeqs" phi1'

phi2 <- coerceMorphism (targetLogic cid2) tlid "coeqs" phi2'

-- build the double arrow for computing the coequalizers

let doubleArrow = Graph.mkGraph

[(n, plainSign s1), (newN, plainSign tsign)]

[(n, newN, (1, phi1)), (n, newN, (1, phi2))]

(colS, colM) <- weakly_amalgamable_colimit tlid doubleArrow

let newGt1 = noSensGTheory tlid (mkExtSign colS) startSigId

mor11' <- coerceMorphism tlid (targetLogic cid1) "coeqs" $ (Map.!) colM newN

mor11 <- comp mor1 mor11'

mor22' <- coerceMorphism tlid (targetLogic cid2) "coeqs" $ (Map.!) colM newN

mor22 <- comp mor2 mor22'

let gMor11 = GMorphism cid1 sig1 idx1 mor11 startMorId

let gMor22 = GMorphism cid2 sig2 idx2 mor22 startMorId

computeCoeqs graph funDesc (n1, gt1) (n2,gt2) (newN, newGt1)

gMor11 gMor22 descs

-- returns a maximal node available

pickMaxNode :: (MonadPlus t) => Gr a b -> t (Node,a)

pickMaxNode graph = msum $ map return $

filter (\(node,_) -> outdeg graph node == 0) $

labNodes graph

-- returns a list of common descendants of two maximal nodes:

-- one node if a glb exists, or all maximal descendants otherwise

commonDesc :: Map.Map Node [(Node,G_theory)] -> Node -> Node

-> [(Node, G_theory)]

commonDesc funDesc n1 n2 = case glb funDesc n1 n2 of

Just x -> [x]

Nothing -> maxBounds funDesc n1 n2

-- returns a weakly amalgamable square of lax triangles

pickSquare :: (MonadPlus t) => Result GMorphism -> Result GMorphism -> t Square

pickSquare (Result _ (Just phi1@(GMorphism cid1 _ _ _ _)))

(Result _ (Just phi2@(GMorphism cid2 _ _ _ _))) =

if (isHomogeneous phi1 && isHomogeneous phi2) then

return $ mkIdSquare $ Logic $ sourceLogic cid1

--since they have the same target, both homogeneous implies same logic

else do

-- if one of them is homogeneous, build the square

-- with identity modification of the other comorphism

let defaultSquare = if isHomogeneous phi1 then

[mkDefSquare $ Comorphism cid2]

else if isHomogeneous phi2 then

[mirrorSquare $ mkDefSquare $ Comorphism cid1]

else []

case maybeResult $ lookupSquare_in_LG (Comorphism cid1)(Comorphism cid2) of

Nothing -> msum $ map return $ defaultSquare

Just sqList -> msum $ map return $ sqList ++ defaultSquare

pickSquare (Result _ Nothing) _ = fail "Error computing comorphisms"

pickSquare _ (Result _ Nothing) = fail "Error computing comorphisms"

-- builds the span for which the colimit is computed

buildSpan :: GDiagram ->

Map.Map Node [(Node, G_theory)] ->

AnyComorphism ->

AnyComorphism ->

AnyComorphism ->

AnyComorphism ->

AnyComorphism ->

AnyModification ->

AnyModification ->

G_theory ->

G_theory ->

G_theory ->

GMorphism ->

GMorphism ->

Int -> Int -> Int ->

[(Int, G_theory)]->

Result (GDiagram, Map.Map Node [(Node,G_theory)])

buildSpan graph

funDesc

d@(Comorphism _cidD)

e1@(Comorphism cidE1)

e2@(Comorphism cidE2)

_d1@(Comorphism _cidD1)

_d2@(Comorphism _cidD2)

_m1@(Modification cidM1)

_m2@(Modification cidM2)

gt@(G_theory lid sign _ _ _)

gt1@(G_theory lid1 sign1 _ _ _)

gt2@(G_theory lid2 sign2 _ _ _)

_phi1@(GMorphism cid1 _ _ mor1 _)

_phi2@(GMorphism cid2 _ _ mor2 _)

n n1 n2

maxNodes

= do

sig@(G_theory _lid0 _sign0 _ _ _) <- mapG_theory d gt -- phi^d(Sigma)

sig1 <- mapG_theory e1 gt1 -- phi^e1(Sigma1)

sig2 <- mapG_theory e2 gt2 -- phi^e2(Sigma2)

mor1' <- coerceMorphism (targetLogic cid1) (sourceLogic cidE1) "buildSpan" mor1

eps1 <- map_morphism cidE1 mor1' -- phi^e1(sigma1)

sign' <- coerceSign lid (sourceLogic$ sourceComorphism cidM1) "buildSpan" sign

tau1 <- tauSigma cidM1 (plainSign sign') -- I^u1_Sigma

tau1' <- coerceMorphism (targetLogic$ sourceComorphism cidM1)

(targetLogic cidE1) "buildSpan" tau1

rho1 <- comp tau1' eps1

mor2' <- coerceMorphism (targetLogic cid2) (sourceLogic cidE2) "buildSpan" mor2

eps2 <- map_morphism cidE2 mor2' --phi^e2(sigma2)

sign'' <- coerceSign lid (sourceLogic$ sourceComorphism cidM2) "buildSpan" sign

tau2 <- tauSigma cidM2 (plainSign sign'') -- I^u2_Sigma

tau2' <- coerceMorphism (targetLogic$ sourceComorphism cidM2)

(targetLogic cidE2) "buildSpan" tau2

rho2 <- comp tau2' eps2

signE1 <- coerceSign lid1 (sourceLogic cidE1) " " sign1

signE2 <- coerceSign lid2 (sourceLogic cidE2) " " sign2

(graph1, funDesc1) <- addNodeToGraph graph sig sig1 sig2 n n1 n2

(GMorphism cidE1 signE1 startSigId rho1 startMorId)

(GMorphism cidE2 signE2 startSigId rho2 startMorId)

funDesc maxNodes

return (graph1, funDesc1)

pickMaximalDesc :: (MonadPlus t) => [(Node, G_theory)] -> t (Node, G_theory)

pickMaximalDesc descList = msum$ map return descList

nrMaxNodes :: Gr a b -> Int

nrMaxNodes graph = length $ filter (\n -> outdeg graph n == 0) $ nodes graph

-- | backtracking function for heterogeneous weak amalgamable cocones

hetWeakAmalgCocone :: (Monad m, LogicT t, MonadPlus (t m)) =>

GDiagram -> Map.Map Int [(Int, G_theory)] -> t m GDiagram

hetWeakAmalgCocone graph funDesc =

if nrMaxNodes graph == 1 then return graph

else once $ do

(n1,gt1) <- pickMaxNode graph

(n2,gt2) <- pickMaxNode graph

guard (n1 < n2) -- to consider each pair of maximal nodes only once

let descList = commonDesc funDesc n1 n2

case length descList of

0 -> mzero -- no common descendants for n1 and n2

_ -> do -- just one common descendant implies greatest lower bound

-- for several, the tail is not empty and we compute coequalizers

(n,gt) <- pickMaximalDesc descList

let phi1 = dijkstra graph n n1

phi2 = dijkstra graph n n2

square <- pickSquare phi1 phi2

let d = laxTarget $ leftTriangle square

e1 = laxFst $ leftTriangle square

d1 = laxSnd $ leftTriangle square

e2 = laxFst $ rightTriangle square

d2 = laxSnd $ rightTriangle square

m1 = laxModif $ leftTriangle square

m2 = laxModif $ rightTriangle square

case maybeResult phi1 of

Nothing -> mzero

Just phi1' -> case maybeResult phi2 of

Nothing -> mzero

Just phi2' -> do

let mGraph = buildSpan graph funDesc d e1 e2 d1 d2 m1 m2 gt gt1 gt2

phi1' phi2' n n1 n2 $ filter (\(nx,_) -> nx /=n) descList

case maybeResult mGraph of

Nothing -> mzero

Just (graph1, funDesc1) -> hetWeakAmalgCocone graph1 funDesc1