Julia for-loop variable scoping

[andreas-ernst]

Compare the following nearly identical bits of code:

for i=1:3
    needX = false
    try
        X[1] = X[1] + 1
    catch err
        print(err)     
        needX = true
    end
    if needX
        X = [0]
    end
    println("  X=",X)
end

Output:

UndefVarError(:X)  X=[0]
UndefVarError(:X)  X=[0]
UndefVarError(:X)  X=[0]

Now a statement that should do nothing (X=X) to get:

for i=1:3
    needX = false
    try
        X = X # without this statement we create an UndefVarError every time
        X[1] = X[1] + 1
    catch err
        print(err)     
        needX = true
    end
    if needX
        X = [0]
    end
    println("  X=",X)
end

to get the output

UndefVarError(:X)  X=[0]
  X=[1]
  X=[2]

Another interesting variant on this theme:

for i=1:3
    needX,needY = false,false
    try
       x[1] = x[1] + 1
    catch err
        print(err)
        needX = true
    end
    if needX
        x = [0]
    end
    try
        y = y + 1
    catch err
        print(err)     
        needY = true
    end
    if needY
        y = 0
    end
    println("  x=",x,"  y=",y)
end

for i=1:3
    needY = false
    try
        Y = Y + 1
    catch err
        print(err)     
        needY = true
    end
    if needY
        Y = 0
    end
    println("  Y=",Y)
end

output

UndefVarError(:x)UndefVarError(:y)  x=[0]  y=0
UndefVarError(:x)UndefVarError(:y)  x=[0]  y=0
UndefVarError(:x)UndefVarError(:y)  x=[0]  y=0
UndefVarError(:Y)  Y=0
  Y=1
  Y=2

Adding the x=x or y=y statements retains the x variables in the first loop but not the y variables.

Perhaps an indication that the soft scoping rules of for-loops are a bit too baroque.

Version used for testing:

Julia Version 0.6.2
Commit d386e40c17 (2017-12-13 18:08 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, haswell)

[JeffBezanson]

Perhaps an indication that the soft scoping rules of for-loops are a bit too baroque.

No, I suspect it's a bug. "Soft scope" (which no longer exists in v0.7) only has to do with writing to global variables.