Merge pull request #139 from JuliaStats/depr

Match Base syntax, remove type instability from gl

Author: garborg

benchmark/operators.jl

@@ -1,5 +1,5 @@
 module DataArraysBenchmark
-using DataArrays, Benchmark
+using DataArrays, Benchmark, Compat
 
 # seed rng for more consistent timings
 srand(1776)
@@ -26,7 +26,7 @@ function make_test_types(genfunc, sz)
     )
 end
 
-make_bools{N}(x::NTuple{N}) = convert(Array{Bool, N}, randbool(x...))
+make_bools{N}(x::NTuple{N}) = convert(Array{Bool, N}, bitrand(x...))
 make_bools(x::Integer...) = make_bools(x)
 
 macro perf(fn, replications, idx...)

src/broadcast.jl

@@ -9,13 +9,13 @@ function check_broadcast_shape(shape::Dims, As::Union(AbstractArray,Number)...)
     samesize = true
     for A in As
         if ndims(A) > length(shape)
-            error("cannot broadcast array to have fewer dimensions")
+            throw(DimensionMismatch("cannot broadcast array to have fewer dimensions"))
         end
         for k in 1:length(shape)
             n, nA = shape[k], size(A, k)
             samesize &= (n == nA)
             if n != nA != 1
-                error("array could not be broadcast to match destination")
+                throw(DimensionMismatch("array could not be broadcast to match destination"))
             end
         end
     end
@@ -183,9 +183,9 @@ for bsig in (DataArray, PooledDataArray), asig in (Union(Array,BitArray,Number),
     @eval let cache = Dict{Function,Dict{Uint64,Dict{Int,Function}}}()
         function Base.map!(f::Base.Callable, B::$bsig, As::$asig...)
             nd = ndims(B)
-            length(As) <= 8 || error("too many arguments")
+            length(As) <= 8 || throw(ArgumentError("too many arguments"))
             samesize = check_broadcast_shape(size(B), As...)
-            samesize || error("dimensions must match")
+            samesize || throw(DimensionMismatch("Argument dimensions must match"))
             arrtype = datype_int(As...)
 
             cache_f    = @get! cache      f        Dict{Uint64,Dict{Int,Function}}()
@@ -200,7 +200,7 @@ for bsig in (DataArray, PooledDataArray), asig in (Union(Array,BitArray,Number),
             invoke(Base.map!, (Base.Callable, $bsig, $asig), f, B, r)
         function Base.broadcast!(f::Function, B::$bsig, As::$asig...)
             nd = ndims(B)
-            length(As) <= 8 || error("too many arguments")
+            length(As) <= 8 || throw(ArgumentError("too many arguments"))
             samesize = check_broadcast_shape(size(B), As...)
             arrtype = datype_int(As...)
 
@@ -235,7 +235,7 @@ macro da_broadcast_vararg(func)
     if (func.head != :function && func.head != :(=)) ||
        func.args[1].head != :call || !isa(func.args[1].args[end], Expr) ||
        func.args[1].args[end].head != :...
-        error("@da_broadcast_vararg may only be applied to vararg functions")
+        throw(ArgumentError("@da_broadcast_vararg may only be applied to vararg functions"))
     end
 
     va = func.args[1].args[end]
@@ -261,8 +261,9 @@ end
 
 macro da_broadcast_binary(func)
     if (func.head != :function && func.head != :(=)) ||
-       func.args[1].head != :call || length(func.args[1].args) != 3
-        error("@da_broadcast_binary may only be applied to two-argument functions")
+       func.args[1].head != :call ||
+       length(func.args[1].args) != 3
+        throw(ArgumentError("@da_broadcast_binary may only be applied to two-argument functions"))
     end
     (f, A, B) = func.args[1].args
     body = func.args[2]
@@ -276,7 +277,7 @@ end
 # Broadcasting DataArrays returns a DataArray
 @da_broadcast_vararg Base.broadcast(f::Function, As...) = databroadcast(f, As...)
 
-# Definitions for operators, 
+# Definitions for operators,
 Base.(:(.*))(A::BitArray, B::Union(DataArray{Bool}, PooledDataArray{Bool})) = databroadcast(*, A, B)
 Base.(:(.*))(A::Union(DataArray{Bool}, PooledDataArray{Bool}), B::BitArray) = databroadcast(*, A, B)
 @da_broadcast_vararg Base.(:(.*))(As...) = databroadcast(*, As...)

src/extras.jl

@@ -65,7 +65,7 @@ cut(x::AbstractVector, ngroups::Integer) = cut(x, quantile(x, [1 : ngroups - 1]
 
 function rep{T <: Integer}(x::AbstractVector, lengths::AbstractVector{T})
     if length(x) != length(lengths)
-        error("vector lengths must match")
+        throw(DimensionMismatch("vector lengths must match"))
     end
     res = similar(x, sum(lengths))
     i = 1

src/indexing.jl

@@ -82,15 +82,15 @@ end
 
 # Indexing with NA throws an error
 function Base.to_index(A::DataArray)
-    any(A.na) && error("cannot index an array with a DataArray containing NA values")
+    any(A.na) && throw(NAException("cannot index an array with a DataArray containing NA values"))
     Base.to_index(A.data)
 end
 
 # Fast implementation of checkbounds for DataArray input
 Base.checkbounds(sz::Int, I::AbstractDataVector{Bool}) =
     length(I) == sz || throw(BoundsError())
 function Base.checkbounds{T<:Real}(sz::Int, I::AbstractDataArray{T})
-    anyna(I) && error("cannot index into an array with a DataArray containing NAs")
+    anyna(I) && throw(NAException("cannot index into an array with a DataArray containing NAs"))
     extr = daextract(I)
     for i = 1:length(I)
         @inbounds v = unsafe_getindex_notna(I, extr, i)
@@ -100,9 +100,9 @@ end
 
 # Fallbacks to avoid ambiguity
 setindex!(t::AbstractDataArray, x, i::Real) =
-    error("setindex! not defined for ",typeof(t))
+    throw(MethodError(setindex!, typeof(t), typeof(x), typeof(i)))
 getindex(t::AbstractDataArray, i::Real) =
-    error("indexing not defined for ", typeof(t))
+    throw(MethodError(getindex, typeof(t), typeof(i)))
 
 ## getindex: DataArray
 

src/operators.jl

@@ -227,9 +227,9 @@ end
 macro swappable(func, syms...)
     if (func.head != :function && func.head != :(=)) ||
        func.args[1].head != :call || length(func.args[1].args) != 3
-        error("@swappable may only be applied to functions of two arguments")
+        throw(ArgumentError("@swappable may only be applied to functions of two arguments"))
     end
-    
+
     func2 = deepcopy(func)
     fname = func2.args[1].args[1]
     if isa(fname, Expr) && fname.head == :curly
@@ -238,7 +238,7 @@ macro swappable(func, syms...)
 
     for s in unique([fname, syms...])
         if swapargs(func2, s) < 1
-            error("No argument swapped")
+            throw(ErrorException("No argument swapped"))
         end
     end
     esc(Expr(:block, func, func2))
@@ -796,7 +796,7 @@ end
 Base.(:/){T,N}(b::AbstractArray{T,N}, ::NAtype) =
     DataArray(Array(T, size(b)), trues(size(b)))
 @dataarray_binary_scalar Base.(:/) Base.(:/) eltype(a) <: FloatingPoint || typeof(b) <: FloatingPoint ?
-                                      promote_type(eltype(a), typeof(b)) : Float64 false                   
+                                      promote_type(eltype(a), typeof(b)) : Float64 false
 @swappable Base.(:./){T,N}(::NAtype, b::AbstractArray{T,N}) =
     DataArray(Array(T, size(b)), trues(size(b)))
 @dataarray_binary_scalar Base.(:./) Base.(:/) eltype(a) <: FloatingPoint || typeof(b) <: FloatingPoint ?

src/pooleddataarray.jl

@@ -27,7 +27,7 @@ type PooledDataArray{T, R<:Integer, N} <: AbstractDataArray{T, N}
                              p::Vector{T})
         # refs mustn't overflow pool
         if length(rs.a) > 0 && maximum(rs.a) > prod(size(p))
-            error("Reference array points beyond the end of the pool")
+            throw(ArgumentError("Reference array points beyond the end of the pool"))
         end
         new(rs.a,p)
     end
@@ -67,7 +67,7 @@ function PooledDataArray{T,R<:Integer,N}(d::AbstractArray{T, N},
                                          m::AbstractArray{Bool, N},
                                          r::Type{R} = DEFAULT_POOLED_REF_TYPE)
     if length(pool) > typemax(R)
-        error("Cannot construct a PooledDataVector with type $R with a pool of size $(length(pool))")
+        throw(ArgumentError("Cannot construct a PooledDataVector with type $R with a pool of size $(length(pool))"))
     end
 
     newrefs = Array(R, size(d))
@@ -474,9 +474,11 @@ function getpoolidx{T,R<:Union(Uint8, Uint16, Int8, Int16)}(pda::PooledDataArray
         push!(pda.pool, val)
         pool_idx = length(pda.pool)
         if pool_idx > typemax(R)
-            error("You're using a PooledDataArray with ref type $R, which can only hold $(int(typemax(R))) values,\n",
-                  "and you just tried to add the $(typemax(R)+1)th reference.  Please change the ref type\n",
-                  "to a larger int type, or use the default ref type ($DEFAULT_POOLED_REF_TYPE).")
+            throw(ErrorException(
+                "You're using a PooledDataArray with ref type $R, which can only hold $(int(typemax(R))) values,\n",
+                "and you just tried to add the $(typemax(R)+1)th reference.  Please change the ref type\n",
+                "to a larger int type, or use the default ref type ($DEFAULT_POOLED_REF_TYPE)."
+            ))
         end
     end
     return pool_idx
@@ -529,7 +531,7 @@ end
 function replace!{S, T}(x::PooledDataArray{S}, fromval::T, toval::NAtype)
     fromidx = findfirst(x.pool, fromval)
     if fromidx == 0
-        error("can't replace a value not in the pool in a PooledDataVector!")
+        throw(ErrorException("can't replace a value not in the pool in a PooledDataVector!"))
     end
 
     x.refs[x.refs .== fromidx] = 0
@@ -553,7 +555,7 @@ function replace!{R, S, T}(x::PooledDataArray{R}, fromval::S, toval::T)
     # throw error if fromval isn't in the pool
     fromidx = findfirst(x.pool, fromval)
     if fromidx == 0
-        error("can't replace a value not in the pool in a PooledDataArray!")
+        throw(ErrorException("can't replace a value not in the pool in a PooledDataArray!"))
     end
 
     # if toval is in the pool too, use that and remove fromval from the pool
@@ -767,7 +769,7 @@ function array{T, R}(da::PooledDataArray{T, R})
     res = Array(T, size(da))
     for i in 1:n
         if da.refs[i] == zero(R)
-            error(NAException())
+            throw(NAException())
         else
             res[i] = da.pool[da.refs[i]]
         end

src/reduce.jl

@@ -23,7 +23,7 @@ function mapreduce_seq_impl_skipna(f, op, T, A::DataArray, ifirst::Int, ilast::I
     data = A.data
     na = A.na
     chunks = na.chunks
-    
+
     v, i = skipna_init(f, op, na, data, ifirst, ilast)
 
     while i < ilast
@@ -64,7 +64,7 @@ end
 
 mapreduce_impl_skipna{T}(f, op, A::DataArray{T}) =
     mapreduce_seq_impl_skipna(f, op, T, A, 1, length(A.data))
-mapreduce_impl_skipna(f, op::Base.AddFun, A::DataArray) = 
+mapreduce_impl_skipna(f, op::Base.AddFun, A::DataArray) =
     mapreduce_pairwise_impl_skipna(f, op, A, 1, length(A.na.chunks),
                                    length(A.na)-countnz(A.na),
                                    max(128, Base.sum_pairwise_blocksize(f)))
@@ -144,9 +144,9 @@ function Base.varm{T}(A::DataArray{T}, m::Number; corrected::Bool=true, skipna::
         nna == n && return convert(Base.momenttype(T), NaN)
         nna == n-1 && return convert(Base.momenttype(T),
                                      abs2(A.data[Base.findnextnot(na, 1)] - m)/(1 - int(corrected)))
-        
+
         /(nna == 0 ? Base.centralize_sumabs2(A.data, m, 1, n) :
-                     mapreduce_impl_skipna(Base.CentralizedAbs2Fun(m), Base.AddFun(), A), 
+                     mapreduce_impl_skipna(Base.CentralizedAbs2Fun(m), Base.AddFun(), A),
           n - nna - int(corrected))
     else
         any(A.na) && return NA
@@ -165,14 +165,15 @@ end
 function Base.var(A::DataArray; corrected::Bool=true, mean=nothing, skipna::Bool=false)
     mean == 0 ? Base.varzm(A; corrected=corrected, skipna=skipna) :
     mean == nothing ? varm(A, Base.mean(A; skipna=skipna); corrected=corrected, skipna=skipna) :
-    isa(mean, Union(Number, NAtype)) ? varm(A, mean; corrected=corrected, skipna=skipna) :
-    error("Invalid value of mean.")
+    isa(mean, Union(Number, NAtype)) ?
+        varm(A, mean; corrected=corrected, skipna=skipna) :
+        throw(ErrorException("Invalid value of mean."))
 end
 
-Base.stdm(A::DataArray, m::Number; corrected::Bool=true, skipna::Bool=false) = 
+Base.stdm(A::DataArray, m::Number; corrected::Bool=true, skipna::Bool=false) =
     sqrt(varm(A, m; corrected=corrected, skipna=skipna))
 
-Base.std(A::DataArray; corrected::Bool=true, mean=nothing, skipna::Bool=false) = 
+Base.std(A::DataArray; corrected::Bool=true, mean=nothing, skipna::Bool=false) =
     sqrt(var(A; corrected=corrected, mean=mean, skipna=skipna))
 
 ## weighted mean