Merge pull request #47 from tk3369/tk/migrate0.7

Julia 0.7 support

Author: tk3369

.travis.yml

@@ -5,6 +5,7 @@ os:
   - osx
 julia:
   - 0.6
+  - 0.7
   - nightly
 notifications:
   email: false

deps/build.jl

@@ -35,9 +35,14 @@ Base.haskey(d::CIDict, s::String) = haskey(d.dct, lcase(s))
 Base.keys(d::CIDict) = keys(d.dct)
 Base.values(d::CIDict) = values(d.dct)
 
-Base.start(d::CIDict) = start(d.dct)
-Base.next(d::CIDict, i::Int) = next(d.dct, i)
-Base.done(d::CIDict, i::Int) = done(d.dct, i)
+@static if VERSION.minor < 7
+    Base.start(d::CIDict) = start(d.dct)
+    Base.next(d::CIDict, i::Int) = next(d.dct, i)
+    Base.done(d::CIDict, i::Int) = done(d.dct, i)
+else
+    Base.iterate(d::CIDict) = iterate(d.dct)
+    Base.iterate(d::CIDict, i) = iterate(d.dct, i)
+end
 
 Base.length(d::CIDict) = length(d.dct)
 

src/CIDict.jl

@@ -74,7 +74,8 @@ read the entire file content.  When called again, fetch the next `nrows` rows.
 function read(handler::Handler, nrows=0) 
     # println("Reading $(handler.config.filename)")
     elapsed = @elapsed result = read_chunk(handler, nrows)
-    elapsed = round(elapsed, 5)
+    # TODO base keyword arg should not be needed due to Compat.jl issue #567
+    elapsed = Compat.round(elapsed; digits = 5, base = 10)
     println1(handler, "Read $(handler.config.filename) with size $(size(result, 1)) x $(size(result, 2)) in $elapsed seconds")
     return result
 end
@@ -983,7 +984,7 @@ end
 
 # convert Float64 value into Date object 
 function date_from_float(x::Vector{Float64})
-    @compat v = Vector{Union{Date, Missing}}(uninitialized, length(x))
+    @compat v = Vector{Union{Date, Missing}}(undef, length(x))
     for i in 1:length(x)
         v[i] = isnan(x[i]) ? missing : (sas_date_origin + Dates.Day(round(Int64, x[i])))
     end
@@ -992,7 +993,7 @@ end
 
 # convert Float64 value into DateTime object 
 function datetime_from_float(x::Vector{Float64})
-    @compat v = Vector{Union{DateTime, Missing}}(uninitialized, length(x))
+    @compat v = Vector{Union{DateTime, Missing}}(undef, length(x))
     for i in 1:length(x)
         v[i] = isnan(x[i]) ? missing : (sas_datetime_origin + Dates.Second(round(Int64, x[i])))
     end

src/SASLib.jl

@@ -213,6 +213,6 @@ const VENDOR_SAS           = 0x01
 const FALLBACK_ENCODING = "UTF-8"
 const ENCODINGS_OK_WITH_BASE_TRANSCODER = [ "UTF-8" , "US-ASCII" ]
 
-@compat const REGULAR_STR_ARRAY(n) = Array{String}(uninitialized, n)
+@compat const REGULAR_STR_ARRAY(n) = Array{String}(undef, n)
 const EMPTY_STRING = ""
 

src/constants.jl

@@ -0,0 +1,73 @@
+# Performance Comparison (0.7.0-beta vs 0.6.3)
+
+## Small Data Set
+
+Reading data_pandas/productsales.sas7bdat (140K) is 12% faster in v0.7.
+
+v0.7
+```
+julia> @benchmark readsas("data_pandas/productsales.sas7bdat", verbose_level = 0) 
+BenchmarkTools.Trial: 
+  memory estimate:  1.01 MiB
+  allocs estimate:  14727
+  --------------
+  minimum time:     1.748 ms (0.00% GC)
+  median time:      1.843 ms (0.00% GC)
+  mean time:        2.027 ms (5.61% GC)
+  maximum time:     58.967 ms (96.56% GC)
+  --------------
+  samples:          2458
+  evals/sample:     1
+```
+
+v0.6.3
+```
+julia> @benchmark readsas("data_pandas/productsales.sas7bdat", verbose_level = 0)
+BenchmarkTools.Trial: 
+  memory estimate:  1.07 MiB
+  allocs estimate:  18505
+  --------------
+  minimum time:     1.987 ms (0.00% GC)
+  median time:      2.150 ms (0.00% GC)
+  mean time:        2.367 ms (5.56% GC)
+  maximum time:     10.130 ms (70.77% GC)
+  --------------
+  samples:          2108
+  evals/sample:     1
+```
+
+## Larger Data Set
+
+Reading data_AHS2013/topical.sas7bdat (14 MB) is 18% faster in v0.7.
+
+v0.7
+```
+julia> @benchmark readsas("data_AHS2013/topical.sas7bdat", verbose_level = 0) seconds=60
+BenchmarkTools.Trial: 
+  memory estimate:  649.63 MiB
+  allocs estimate:  19011924
+  --------------
+  minimum time:     1.959 s (10.46% GC)
+  median time:      2.042 s (12.78% GC)
+  mean time:        2.061 s (12.59% GC)
+  maximum time:     2.348 s (12.17% GC)
+  --------------
+  samples:          30
+  evals/sample:     1
+```
+
+v0.6.3
+```
+julia> @benchmark readsas("data_AHS2013/topical.sas7bdat", verbose_level = 0) seconds=60
+BenchmarkTools.Trial: 
+  memory estimate:  632.36 MiB
+  allocs estimate:  18653427
+  --------------
+  minimum time:     2.391 s (10.82% GC)
+  median time:      2.520 s (13.07% GC)
+  mean time:        2.524 s (12.84% GC)
+  maximum time:     2.638 s (12.87% GC)
+  --------------
+  samples:          24
+  evals/sample:     1
+```

test/perf_results_0.5.2/performance_0.7.0_vs_0.6.3.md

@@ -1,4 +1,6 @@
 using BenchmarkTools
+using InteractiveUtils
+using Printf
 
 if length(ARGS) != 2
     println("Usage: $PROGRAM_FILE <filename> <count>")
@@ -8,9 +10,8 @@ end
 versioninfo()
 println()
 
-tic()
-using SASLib
-@printf "Loaded library in %.3f seconds\n" toq()
+load_time = @elapsed using SASLib
+@printf "Loaded library in %.3f seconds\n" load_time
 
 b = @benchmark readsas($ARGS[1], verbose_level=0) samples=parse(Int, ARGS[2])
 display(b)

test/perf_test1.jl

@@ -24,7 +24,7 @@ Base.convert(::Type{YearStr}, v::Float64) = YearStr(string(round(Int, v)))
         x = SASLib.ObjectPool{String, UInt8}(default, 5)
         @test length(x) == 5
         @test size(x) == (5, )
-        @test endof(x) == 5
+        @test @compat lastindex(x) == 5
         @test count(v -> v == default, x) == 5
         @test count(v -> v === default, x) == 5
         @test map(v -> "x$v", x) == [ "x", "x", "x", "x", "x" ]
@@ -239,7 +239,7 @@ Base.convert(::Type{YearStr}, v::Float64) = YearStr(string(round(Int, v)))
         # Deal with v0.6/v0.7 difference
         # v0.6 shows Missings.Missing
         # v0.7 shows Missing
-        ty(x) = replace(x, "Missings.", "")   
+        ty(x) = replace(x, "Missings." => "")   
 
         # convenient comparison routine since v0.6/v0.7 displays different order
         same(x,y) = sort(ty.(string.(collect(x)))) == sort(ty.(string.(collect(y))))