Your function isn't really using a pre-allocated output , because you're creating I. To really use the preallocated output (indexes), you'll need to set values in that array. resize! might be useful here, or you could use push!.
Regarding the warnings, I have no idea, but it looks like you've broken something deeper (notice it's putting "find" in front of filenames in base). Try developing your code outside base (that way you don't have to build Julia each time you want to test it), get it working, and only then add it into base. --Tim On Sunday, July 13, 2014 03:52:39 PM J Luis wrote: > Hmm, tried my luck ... but > > If I add this to bitarray.jl > > find(B::BitArray) = find(Int[], B::BitArray) > > function find(indexes, B::BitArray) > l = length(B) > nnzB = countnz(B) > # I = Array(Int, nnzB) > I = similar(indexes, nnzB) > > it works but I also get these strange warnings when rebuilding. And I say > strange because I don't even find any *findbitarray.jl* in my Julia > instalation > > Warning: New definition > findbitarray.jl > could not show value of type Tupleintset.jl > at bitarray.jl:1330dict.jl > > is ambiguous with: > findset.jl > could not show value of type Tuplehashing.jl > at array.jl:1052iterator.jl > . > To fix, define > findinference.jl > could not show value of type Tupleosutils.jl > > before the new definition. > char.jl > Warning: New definition > findascii.jl > could not show value of type Tupleutf8.jl > at bitarray.jl:1330utf16.jl > > is ambiguous with: > findutf32.jl > could not show value of type Tupleiobuffer.jl > at array.jl:1077string.jl > . > To fix, define > findutf8proc.jl > could not show value of type Tupleregex.jl > > Domingo, 13 de Julho de 2014 19:52:48 UTC+1, Tim Holy escreveu: > > But actually your point is a good one, and is one more argument for > > https://github.com/JuliaLang/julia/issues/1115 > > Over time more and more functions have been modified to have this type of > > interface, so certainly there is room for a > > > > find!(indexes, testf::Function, x) > > > > function. `find` could then be written as > > > > find(testf::Function, x) = find!(Int[], testf, x) > > > > Since this matters to you, care to prepare a pull request for this and any > > others that annoy you? > > > > --Tim > > > > On Sunday, July 13, 2014 11:38:46 AM J Luis wrote: > > > Thanks. That's what I intend to do but I confess that I find the default > > > > to > > > > > Int64 a bit annoying (for example when writing wrappers to C functions > > > > whre > > > > > the Int(s) arguments are almost never Int64) > > > > > > Domingo, 13 de Julho de 2014 10:50:12 UTC+1, Tim Holy escreveu: > > > > If this really matters to you, check out the implementations of find > > > > in > > > > > > base/array.jl. It's so short, you can trivially implement whatever > > > > behavior > > > > you want. For example, you could pass in an empty output array and > > > > have it > > > > > > push! the indexes into it. > > > > > > > > --Tim > > > > > > > > On Saturday, July 12, 2014 12:43:29 PM J Luis wrote: > > > > > > julia> find(x->x>5,a) > > > > > > > > > > > > 5-element Array{Int64,1}: > > > > > > 1 > > > > > > 2 > > > > > > 7 > > > > > > 8 > > > > > > > > > > > > 10 > > > > > > > > > > which very very sadly are Int64. When dealing with large matrices > > > > this > > > > > > may > > > > > > > > > lead to a large memory waste. These almost mandatory 64 bits issue > > > > is > > > > > > one > > > > > > > > > the things that annoyed me more in Matlab for many times it was the > > > > > difference between having something work ... or not
