TCMC optimisation fails to kick in when in the presence of multi-specz

[jimbxb]

As the title says.

E.g., consider the following:

pub def list(ar:array(T)):list(T) = ls where {
    if { [?x | ?ar] = ar ::
        ?ls = list(ar)
        [x | ls] = ?ls 
       | else :: 
        ?ls = []
    }
}

The compiler generates the following:

list > public (1 calls)
0: opt.list<0>
list(ar##0:wybe.array(T), ?#result##0:wybe.list(T))<{}; {}>:
  AliasPairs: []
  InterestingCallProperties: [InterestingUnaliased 0]
  MultiSpeczDepInfo: [(0,(wybe.array.[|]<0>,fromList [NonAliasedParamCond 2 [0]])),(1,(opt.list<0>,fromList [NonAliasedParamCond 0 []]))]
    wybe.array.[|]<0>(?x##0:T, ?ar##1:wybe.array(T), ~ar##0:wybe.array(T), ?tmp#3##0:wybe.bool) #0 @opt:16:10
    case ~tmp#3##0:wybe.bool of
    0:
        foreign llvm move(0:wybe.list(T), ?#result##0:wybe.list(T)) @opt:15:5

    1:
        opt.list<0>[410bae77d3](~ar##1:wybe.array(T), ?tmp#0##0:wybe.list(T)) #1 @opt:17:15
        foreign lpvm alloc(16:wybe.int, ?tmp#6##0:wybe.list(T)) @list:9:27
        foreign lpvm mutate(~tmp#6##0:wybe.list(T), ?tmp#7##0:wybe.list(T), 0:wybe.int, 1:wybe.int, 16:wybe.int, 0:wybe.int, ~x##0:T) @list:9:27
        foreign lpvm mutate(~tmp#7##0:wybe.list(T), ?#result##0:wybe.list(T), 8:wybe.int, 1:wybe.int, 16:wybe.int, 0:wybe.int, ~tmp#0##0:wybe.list(T)) @list:9:27

However with -xno-multi-specz, the compiler can use the TCMC optimisation. The only difference is the multi-specz call in the body.

list > public (1 calls)
0: opt.list<0>
list(ar##0:wybe.array(T), outByReference #result##0:wybe.list(T))<{}; {}>:
  AliasPairs: []
  InterestingCallProperties: [InterestingUnaliased 0]
  MultiSpeczDepInfo: [(0,(wybe.array.[|]<0>,fromList [NonAliasedParamCond 2 [0]])),(1,(opt.list<0>,fromList [NonAliasedParamCond 0 []]))]
    wybe.array.[|]<0>(?x##0:T, ?ar##1:wybe.array(T), ~ar##0:wybe.array(T), ?tmp#3##0:wybe.bool) #0 @opt:16:10
    case ~tmp#3##0:wybe.bool of
    0:
        foreign llvm move(0:wybe.list(T), ?#result##0:wybe.list(T)) @opt:15:5

    1:
        foreign lpvm alloc(16:wybe.int, ?tmp#6##0:wybe.list(T)) @list:9:27
        foreign lpvm mutate(~tmp#6##0:wybe.list(T), ?tmp#7##0:wybe.list(T), 0:wybe.int, 1:wybe.int, 16:wybe.int, 0:wybe.int, ~x##0:T) @list:9:27
        opt.list<0>(~ar##1:wybe.array(T), outByReference tmp#0##0:wybe.list(T)) #1 @opt:17:15
        foreign lpvm mutate(~tmp#7##0:wybe.list(T), ?#result##0:wybe.list(T), 8:wybe.int, 1:wybe.int, 16:wybe.int, 0:wybe.int, ~takeReference tmp#0##0:wybe.list(T)) @list:9:27

[neon64]

Hey, good pick up! I believe this behavior should be fixed in #316 (comment), I've done the rest of my uni exams now, so hopefully can look at polishing those changes