Fix interval type inference problem (#78)

* Fix interval type inference problem

* bump version

Project.toml

@@ -1,7 +1,7 @@
 name = "Ripserer"
 uuid = "aa79e827-bd0b-42a8-9f10-2b302677a641"
 authors = ["mtsch <matijacufar@gmail.com>"]
-version = "0.14.1"
+version = "0.14.2"
 
 [deps]
 Compat = "34da2185-b29b-5c13-b0c7-acf172513d20"

src/abstractfiltration.jl

@@ -157,17 +157,20 @@ computed. Defaults to sorting the diagram.
 """
 postprocess_diagram(::AbstractFiltration, diagram) = sort!(diagram)
 
-function interval_meta_type(flt::AbstractFiltration, dim, reps, field)
+# Vals everywhere so compiler computes this at compile time.
+@inline function interval_type(
+    flt::AbstractFiltration, ::Val{dim}, ::Val{reps}, ::Type{F}
+) where {dim, reps, F}
     if reps
-        return @NamedTuple begin
+        return PersistenceInterval{@NamedTuple begin
             birth_simplex::simplex_type(flt, dim)
             death_simplex::Union{simplex_type(flt, dim + 1), Nothing}
-            representative::Vector{chain_element_type(simplex_type(flt, dim), field)}
-        end
+            representative::Vector{chain_element_type(simplex_type(flt, dim), F)}
+        end}
     else
-        return @NamedTuple begin
+        return PersistenceInterval{@NamedTuple begin
             birth_simplex::simplex_type(flt, dim)
             death_simplex::Union{simplex_type(flt, dim + 1), Nothing}
-        end
+        end}
     end
 end

src/reductionmatrix.jl

@@ -309,8 +309,8 @@ function birth_death(::ReductionMatrix{false}, column, pivot)
 end
 
 function add_interval!(
-    intervals, matrix::ReductionMatrix, column, pivot, cutoff, reps
-)
+    intervals, matrix::ReductionMatrix, column, pivot, cutoff, ::Val{reps}
+) where reps
     birth_time, birth_sx, death_time, death_sx = birth_death(matrix, column, pivot)
     if death_time - birth_time > cutoff
         if reps && is_cohomology(matrix)
@@ -325,13 +325,14 @@ function add_interval!(
         else
             rep = NamedTuple()
         end
-        int = PersistenceInterval(
-            birth_time, death_time;
+        meta = (;
             birth_simplex=birth_sx,
             death_simplex=death_sx,
             rep...,
         )
-        !isnothing(int) && push!(intervals, int)
+        push!(intervals, PersistenceInterval(
+            birth_time, death_time, meta
+        ))
     end
 end
 
@@ -344,13 +345,13 @@ function compute_intervals!(
             desc="Computing $(dim(matrix))d intervals... ",
         )
     end
-    intervals = PersistenceInterval{
-        interval_meta_type(matrix.filtration, dim(matrix), reps, field_type(matrix))
-    }[]
+    intervals = interval_type(
+        matrix.filtration, Val(dim(matrix)), Val(reps), field_type(matrix)
+    )[]
 
     for column in matrix.columns_to_reduce
         pivot = reduce_column!(matrix, column)
-        add_interval!(intervals, matrix, column, pivot, cutoff, reps)
+        add_interval!(intervals, matrix, column, pivot, cutoff, Val(reps))
         progress && next!(progbar; showvalues=((:n_intervals, length(intervals)),))
     end
     thresh=Float64(threshold(matrix.filtration))

src/zerodimensional.jl

@@ -78,8 +78,8 @@ end
 birth(dset::DisjointSetsWithBirth, i) = dset.births[i]
 
 function add_interval!(
-    intervals, dset::DisjointSetsWithBirth, filtration, vertex, edge, cutoff, reps
-)
+    intervals, dset::DisjointSetsWithBirth, filtration, vertex, edge, cutoff, ::Val{reps}
+) where reps
     birth_time, birth_vertex = birth(dset, vertex)
     death_time = isnothing(edge) ? Inf : birth(edge)
     if death_time - birth_time > cutoff
@@ -116,7 +116,7 @@ function zeroth_intervals(
     CE = chain_element_type(V, F)
     dset = DisjointSetsWithBirth(vertices(filtration), birth(filtration))
 
-    intervals = PersistenceInterval{interval_meta_type(filtration, 0, reps, F)}[]
+    intervals = interval_type(filtration, Val(0), Val(reps), F)[]
 
     to_skip = edge_type(filtration)[]
     to_reduce = edge_type(filtration)[]
@@ -134,7 +134,7 @@ function zeroth_intervals(
         if i ≠ j
             # According to the elder rule, the vertex with the higer birth will die first.
             last_vertex = birth(dset, i) > birth(dset, j) ? i : j
-            add_interval!(intervals, dset, filtration, last_vertex, edge, cutoff, reps)
+            add_interval!(intervals, dset, filtration, last_vertex, edge, cutoff, Val(reps))
 
             union!(dset, i, j)
             push!(to_skip, edge)
@@ -145,7 +145,7 @@ function zeroth_intervals(
     end
     for v in vertices(filtration)
         if find_root!(dset, v) == v && !isnothing(simplex(filtration, Val(0), (v,), 1))
-            add_interval!(intervals, dset, filtration, v, nothing, cutoff, reps)
+            add_interval!(intervals, dset, filtration, v, nothing, cutoff, Val(reps))
         end
         progress && next!(progbar; showvalues=((:n_intervals, length(intervals)),))
     end