smarter BoundsError reporting

[vtjnash]

BoundsError() made me sad (so I fixed it)

this adds argument information to intrinsic bounds checking operations, optimized to have no added impact on the runtime cost, in the common case of not having an error

lots of examples (because there's a lot of cases this handles):

julia> a,b,c = 1,2
ERROR: BoundsError(
  of (1,2)
  at index [3]
  )
 in indexed_next at /Volumes/Lion/Users/jameson/Documents/julia/usr/lib/julia/sys.dylib

julia> f(::Any, ::Any, ::Any, args...) = Any
f (generic function with 1 method)

julia> invoke(f, (Any, Any, Any), 1, 1, 1)
ERROR: BoundsError(
  of (Int64,Int64...)
  at index [3]
  )
 in typeinf at ./inference.jl:1393
 in typeinf_ext at ./inference.jl:1242

julia> [][1,2]
ERROR: BoundsError(
  of 0-element Array{Any,1}
  at index [1,2]
  )
 in getindex at no file

julia> [1,2][10]
ERROR: BoundsError(
  of 2-element Array{Int64,1}:
 1
 2
  at index [10]
  )
 in getindex at /Volumes/Lion/Users/jameson/Documents/julia/usr/lib/julia/sys.dylib

julia> [1,2][10,10]
ERROR: BoundsError(
  of 2-element Array{Int64,1}:
 1
 2
  at index [10,10]
  )
 in getindex at no file

julia> f(x) = (a=1+2im; a.(x))
f (generic function with 2 methods)

julia> f(4)
ERROR: BoundsError(
  of 1 + 2im
  at index [4]
  )
 in f at no file

julia> f(x) = (a=(1,2.,""); a[x])
f (generic function with 2 methods)

julia> f(4)
ERROR: BoundsError(
  of (1,2.0,"")
  at index [4]
  )
 in f at no file

julia> f(x) = (a=(1,2.); a[x])
f (generic function with 2 methods)

julia> f(4)
ERROR: BoundsError(
  of (1,2.0)
  at index [4]
  )
 in f at no file

julia> f(x) = (a=(1,2,4,6,7); a[x])
f (generic function with 2 methods)

julia> f(30)
ERROR: BoundsError(
  of (1,2,4,6,7)
  at index [30]
  )
 in f at no file

julia> f(x) = (a=IOBuffer(); a.(x) = 3)
f (generic function with 2 methods)

julia> f(30)
ERROR: BoundsError(
  of IOBuffer
  at index [30]
  )
 in f at no file

julia> f(x) = (a=IOBuffer(); a.(x))
f (generic function with 2 methods)

julia> f(30)
ERROR: BoundsError(
  of IOBuffer(data=UInt8[...], readable=true, writable=true, seekable=true, append=false, size=0, maxsize=Inf, ptr=1, mark=-1)
  at index [30]
  )
 in f at no file

julia> f() = (a=IOBuffer(); a.(100))
f (generic function with 3 methods)

julia> f()
ERROR: BoundsError(
  of IOBuffer(data=UInt8[...], readable=true, writable=true, seekable=true, append=false, size=0, maxsize=Inf, ptr=1, mark=-1)
  at index [100]
  )
 in f at no file

julia> f() = (a=1+2im; a.(100))
f (generic function with 3 methods)

julia> f()
ERROR: BoundsError(
  of 1 + 2im
  at index [100]
  )
 in f at no file

julia> f() = (a=(1,2,3); a[100])
f (generic function with 3 methods)

julia> f()
ERROR: BoundsError(
  of (1,2,3)
  at index [100]
  )
 in f at no file

julia> f() = (a=[1,2,3]; a[100])
f (generic function with 3 methods)

julia> f()
ERROR: BoundsError(
  of 3-element Array{Int64,1}:
 1
 2
 3
  at index [100]
  )
 in f at no file

julia> [][]
ERROR: BoundsError(
  of 0-element Array{Any,1}
  at index [1]
  )
 in getindex at no file

julia> (1,2,3)[100]
ERROR: BoundsError(
  of (1,2,3)
  at index [100]
  )
 in getindex at /Volumes/Lion/Users/jameson/Documents/julia/usr/lib/julia/sys.dylib

julia> (1,2,3.)[100]
ERROR: BoundsError(
  of (1,2,3.0)
  at index [100]
  )
 in getindex at /Volumes/Lion/Users/jameson/Documents/julia/usr/lib/julia/sys.dylib

julia> (1,2,3.,"4")[100]
ERROR: BoundsError(
  of (1,2,3.0,"4")
  at index [100]
  )
 in getindex at /Volumes/Lion/Users/jameson/Documents/julia/usr/lib/julia/sys.dylib

julia> (1+2im).(100)
ERROR: BoundsError(
  of 1 + 2im
  at index [100]
  )

julia> error(BoundsError())
ERROR: BoundsError()
 in error at no file

julia> error(BoundsError(Int))
ERROR: BoundsError(
  of Int64
  )
 in error at no file

julia> error(BoundsError(Int, 10))
ERROR: BoundsError(
  of Int64
  at index [10]
  )
 in error at no file

julia> error(BoundsError(Int, (:a,)))
ERROR: BoundsError(
  of Int64
  at index [a]
  )
 in error at no file

julia> f(a,b,c...) = c[a]
f (generic function with 1 method)

julia> f(1000,2,3,4,5,6)
ERROR: BoundsError(
  of (3,4,5,6)
  at index [1000]
  )

ref #9520
ref #6297
ref #7978
ref #4744

[nalimilan]

Wow, this is a really incredible start for the new year! This is a major improvement IMHO.

One small suggestion: maybe to make the error even more explicit, say

ERROR: BoundsError(
  attempt to access X
  at index [Y]
  )

[timholy]

💯

[dhoegh]

+1 for the wording suggested by @nalimilan, I find it more explicit and easy to understand.

[Ismael-VC]

+1 for explicit and readable, cool!

On Thu, Jan 1, 2015 at 8:16 AM, Daniel Høegh [email protected]
wrote:

+1 for the wording suggested by @nalimilan https://github.com/nalimilan,
I find it more explicit and easy to understand.

—
Reply to this email directly or view it on GitHub
#9534 (comment).

[chrisvoncsefalvay]

<3 this.

[tknopp]

+10000

This can only be beaten by package precompilation Jameson ;-)

[timholy]

We should all get together and buy Jameson his own tropical island for Christmas (as long as it has really good WiFi).

[joehuchette]

This is a fantastic change, but the printing maybe should be tweaked a bit:

a = rand(100); a[101]

gives

[timholy]

@joehuchette, the printing could be tackled by anyone, that's something that doesn't need Jameson's wizard skills. Care to take a crack yourself? See the implementation of show(io::IO, be::BoundsError) in base/replutil.c.

[JeffBezanson]

Yes the printing needs to be fixed. There is no point in showing syntax-style BoundsError( ... ) if there is free-form english inside the parentheses.

[yuyichao]

I see getfield(1 + 2im, 3) in the test. Is there a reason to leave getindex(1 + 2im, 3) (as well as other use of BoundError() in the base module)?

Also given that the help string is

help?> getfield
Base.getfield(value, name::Symbol)

Why is a non-symbol name even allowed?

IMHO, it is super confusing right now if for the following error is raised

julia> getfield([1], 1)
ERROR: BoundsError: attempt to access 1-element Array{Int64,1}:
 1
  at index [1]

Which looks like (just from the error) julia raise a error on [1][1].....

Edit: Ok. I see that number index is used to access field in a composite type. Nevertheless, the error message above is still confusing. It's probably better to change the doc string of getfield to include this and have a slightly better error message as well.

[vtjnash]

since getfield and getindex throw exactly the same error, there wasn't much I could do about the error message. hopefully, that will usually be available from context when reading the source code.

getfield (x[n]) and getindex (x.(n)) are quite different in how they operate:
getindex(x::Numeric, n) returns the number x (for n = 1), while
getfield(x, n) returns the n-th field of the object x (numbered in order of their definition in the object)

[yuyichao]

Maybe adding a flag to indicate the context? (getindex/getfield, maybe also another one for iterator?)

[nalimilan]

Just wondering: wouldn't it be better to print the index first, and the array after? When arrays are long, the index is hard to find. For example (from #9607):

julia-0.4> b[4] = true
ERROR: BoundsError: attempt to access 3-element BitArray{1}:
  true
  true
 false
  at index [4]

vs. something like

julia-0.4> b[4] = true
ERROR: BoundsError: attempt to access
  index [4]
of 3-element BitArray{1}:
  true
  true
 false

What do you think?

[timholy]

When the underlying array is 100x100x2500, the index (if it comes first) will disappear from the console's history. But in any event we probably want to switch to printing a summary when the array gets large.

[jakebolewski]

I don't see how printing out the value is useful. The Type of the value, it's size / length and the out-of-bounds index requested should be all you need.