Re: [julia-users] BinDeps: How to access include path?
Le dimanche 16 novembre 2014 à 15:41 -0500, Erik Schnetter a écrit : I see. I was afraid that the C structs may change in between different versions of the library. If the structs are re-analyzed when the wrapper is installed, this would not be an issue. Stable libraries shouldn't change their structs, except when releasing new major versions. In that case even classic C programs will break, so it's not really something to worry about specifically for Julia code. Ideally your package should download a version series known to work, so things don't break even a breaking version is released. And if a new major version is released and the structs change, other things are likely to change too, so that your code will break anyway. That's what SOVERSION is used for usually. My two cents On Sun, Nov 16, 2014 at 2:58 PM, Tim Holy tim.h...@gmail.com wrote: The only person who needs to use Clang.jl is the package developer. Clang writes *.jl files that save you the trouble of keyboarding all the julia equivalents of the C structs. It also generates the ccall wrappers. You then write your package on top of those generated files. Clang is therefore only a dependency for the developer, not for the user. That said, if you've already written all your types, Clang won't provide you with much benefit. Quite a few packages use Clang, but one I've been involved with is CUDArt.jl. --Tim On Sunday, November 16, 2014 01:01:42 PM Erik Schnetter wrote: Thanks for all the pointers! Things are now working fine without wrapper functions but with 120 lines of immutable type declarations. Clang.jl sounds interesting, but would probably make hwloc.jl too difficult to use if it is a prerequisite. Let's see how often the structs of hwloc change with future versions. hwloc is a portable library that determines the number of cores (and many other properties) of the local machine: see https://github.com/eschnett/hwloc.jl. -erik On Sun, Nov 16, 2014 at 12:41 PM, Jake Bolewski jakebolew...@gmail.com wrote: No need to flatten if everything is immutable. This file has some examples of wrapping structs of structs https://github.com/jakebolewski/LibGit2.jl/blob/cac78b5c03531b5afbcb0ae042 538dd351527752/src/types.jl#L19. On Sunday, November 16, 2014 12:33:39 PM UTC-5, Tim Holy wrote: AFAIK no need to flatten. Since Isaiah didn't advertise it himself, I'll mention his Clang.jl package, which I've found to be a big help in such situations. It's just possible you may not need any C glue code. --Tim On Sunday, November 16, 2014 12:15:50 PM Erik Schnetter wrote: Thanks for the pointers on immutable types. Is it possible to access structs inside structs this way? That a struct inside a struct, not a pointer inside a struct. Is an immutable type again a bits type? Or do I need to flatten the structs to make this work? -erik On Sun, Nov 16, 2014 at 11:16 AM, Isaiah Norton isaiah...@gmail.com wrote: The library defines some C structs that are part of the API. My current approach uses wrapper C functions to access struct elements. Is there a better way? It depends how complicated are the structs. Structs can often be reflected as Julia types, as long as isbits(TypeName) == true, and accessed with unsafe_load (see for example PyCall's definitions of PyObject_*). Some tricky spots include fixed-size arrays and unions (strictly speaking you can make an aggregate element having the maximal size in the union, and then do bitshifts manually. but there is no automatic support). There is also the StrPack.jl package, which calculates the appropriate memory layout for a given struct and provides serialization/deserialization. I thus I need to build this wrapper file as well. I created a SimpleBuild rule for this. However, I need to know the path where BinDeps installed (or found) the header files so that I can compile this file. How do I access this information? This is going to be package-manager and build-tool specific, and I don't know if there is any abstraction/helper for includes in BinDeps yet. On Sun, Nov 16, 2014 at 10:09 AM, Erik Schnetter schn...@gmail.com wrote: I want to wrap a library (hwloc) for Julia. This is working fine. The library defines some C structs that are part of the API. My current approach uses wrapper C functions to access struct elements. Is there a better way? I thus I need to build this wrapper file as well. I created a SimpleBuild rule for this. However, I need to know the path where BinDeps installed (or found) the header files so that I can compile this file. How do I access this information? -erik -- Erik Schnetter
Re: [julia-users] BinDeps: How to access include path?
On Mon, Nov 17, 2014 at 3:15 AM, Milan Bouchet-Valat nalimi...@club.fr wrote: Le dimanche 16 novembre 2014 à 15:41 -0500, Erik Schnetter a écrit : I see. I was afraid that the C structs may change in between different versions of the library. If the structs are re-analyzed when the wrapper is installed, this would not be an issue. Stable libraries shouldn't change their structs, except when releasing new major versions. In that case even classic C programs will break, so it's not really something to worry about specifically for Julia code. Ideally your package should download a version series known to work, so things don't break even a breaking version is released. In this context, most library users will build from source. I would not expect binary compatibility. So C programs would not break. It is not possible to choose a particular version when installing e.g. via apt-get or homebrew. And if a new major version is released and the structs change, other things are likely to change too, so that your code will break anyway. That's what SOVERSION is used for usually. I'm not choosing a particular soversion either. I don't think that is even supported with BinDeps? -erik My two cents On Sun, Nov 16, 2014 at 2:58 PM, Tim Holy tim.h...@gmail.com wrote: The only person who needs to use Clang.jl is the package developer. Clang writes *.jl files that save you the trouble of keyboarding all the julia equivalents of the C structs. It also generates the ccall wrappers. You then write your package on top of those generated files. Clang is therefore only a dependency for the developer, not for the user. That said, if you've already written all your types, Clang won't provide you with much benefit. Quite a few packages use Clang, but one I've been involved with is CUDArt.jl. --Tim On Sunday, November 16, 2014 01:01:42 PM Erik Schnetter wrote: Thanks for all the pointers! Things are now working fine without wrapper functions but with 120 lines of immutable type declarations. Clang.jl sounds interesting, but would probably make hwloc.jl too difficult to use if it is a prerequisite. Let's see how often the structs of hwloc change with future versions. hwloc is a portable library that determines the number of cores (and many other properties) of the local machine: see https://github.com/eschnett/hwloc.jl. -erik On Sun, Nov 16, 2014 at 12:41 PM, Jake Bolewski jakebolew...@gmail.com wrote: No need to flatten if everything is immutable. This file has some examples of wrapping structs of structs https://github.com/jakebolewski/LibGit2.jl/blob/cac78b5c03531b5afbcb0ae042 538dd351527752/src/types.jl#L19. On Sunday, November 16, 2014 12:33:39 PM UTC-5, Tim Holy wrote: AFAIK no need to flatten. Since Isaiah didn't advertise it himself, I'll mention his Clang.jl package, which I've found to be a big help in such situations. It's just possible you may not need any C glue code. --Tim On Sunday, November 16, 2014 12:15:50 PM Erik Schnetter wrote: Thanks for the pointers on immutable types. Is it possible to access structs inside structs this way? That a struct inside a struct, not a pointer inside a struct. Is an immutable type again a bits type? Or do I need to flatten the structs to make this work? -erik On Sun, Nov 16, 2014 at 11:16 AM, Isaiah Norton isaiah...@gmail.com wrote: The library defines some C structs that are part of the API. My current approach uses wrapper C functions to access struct elements. Is there a better way? It depends how complicated are the structs. Structs can often be reflected as Julia types, as long as isbits(TypeName) == true, and accessed with unsafe_load (see for example PyCall's definitions of PyObject_*). Some tricky spots include fixed-size arrays and unions (strictly speaking you can make an aggregate element having the maximal size in the union, and then do bitshifts manually. but there is no automatic support). There is also the StrPack.jl package, which calculates the appropriate memory layout for a given struct and provides serialization/deserialization. I thus I need to build this wrapper file as well. I created a SimpleBuild rule for this. However, I need to know the path where BinDeps installed (or found) the header files so that I can compile this file. How do I access this information? This is going to be package-manager and build-tool specific, and I don't know if there is any abstraction/helper for includes in BinDeps yet. On Sun, Nov 16, 2014 at 10:09 AM, Erik Schnetter schn...@gmail.com wrote: I want to wrap a library (hwloc) for Julia. This is working fine. The library defines some C
[julia-users] try/catch by exception type
Hello ! Is there a way to catch an exception by type in Julia ? Coming from python, I am very tempted to do this kind of things: ``` try # Access a dict here catch e if isa(KeyError, e) # Handle the KeyError, as I know what to do in that case else # Re-throw to upper level throw(e) end ``` But that is a lot of boilerplate when used many times. Maybe it is not the Julia way to express this kind of problem (handling one type of exception, while letting the others go up), but I could not find something about it. I can not just remove all the exceptions as some of them are thrown by Base. Thank for yours answers ! Guillaume
[julia-users] Re: How should I structure my code?
macro need_m() :(m * 20) end function algoX() m = 10 @need_m end algoX() # 200 I think you can create two similar macro to solve your problem. But i think this is bad practice. I think you should create some code without a scope, as I know, the macro can do such thing. Justas _於 2014年11月17日星期一UTC+8上午4時25分26秒寫道： Lets say I have this kind of code. How could I access variable M in compute() function? function algoX(data) M,N = size(data) param = 20 computeSomething() end export function computeSomething() a = compute() end function compute() # Needs variable M M * param end I do not want to nest all functions, because then I wont be able to export them for testing. Also I do not want to pass variable M to computeSomething() and then pass to compute(). What should I do?
Re: [julia-users] try/catch by exception type
Check out the somewhat lengthy, but informative thread on developing additional functionality with regards to exception handling. https://github.com/JuliaLang/julia/issues/7026 -Jacob On Mon, Nov 17, 2014 at 9:49 AM, Luthaf lut...@luthaf.fr wrote: Hello ! Is there a way to catch an exception by type in Julia ? Coming from python, I am very tempted to do this kind of things: ``` try # Access a dict here catch e if isa(KeyError, e) # Handle the KeyError, as I know what to do in that case else # Re-throw to upper level throw(e) end ``` But that is a lot of boilerplate when used many times. Maybe it is not the Julia way to express this kind of problem (handling one type of exception, while letting the others go up), but I could not find something about it. I can not just remove all the exceptions as some of them are thrown by Base. Thank for yours answers ! Guillaume
Re: [julia-users] try/catch by exception type
Ok, thank you ! So the way to go is better ask for permission than for forgiveness ! John Myles White a écrit : I don't believe this is possible in Julia right now. Which is ok in this case, since working with a KeyError is a very un-Julian way to check for key existence. You'll want to use haskey instead. -- John On Nov 17, 2014, at 2:49 PM, Luthaflut...@luthaf.fr wrote: Hello ! Is there a way to catch an exception by type in Julia ? Coming from python, I am very tempted to do this kind of things: ``` try # Access a dict here catch e if isa(KeyError, e) # Handle the KeyError, as I know what to do in that case else # Re-throw to upper level throw(e) end ``` But that is a lot of boilerplate when used many times. Maybe it is not the Julia way to express this kind of problem (handling one type of exception, while letting the others go up), but I could not find something about it. I can not just remove all the exceptions as some of them are thrown by Base. Thank for yours answers ! Guillaume
Re: [julia-users] try/catch by exception type
Yes, we are civilized after all. -- John On Nov 17, 2014, at 2:53 PM, Luthaf lut...@luthaf.fr wrote: Ok, thank you ! So the way to go is better ask for permission than for forgiveness ! John Myles White a écrit : I don't believe this is possible in Julia right now. Which is ok in this case, since working with a KeyError is a very un-Julian way to check for key existence. You'll want to use haskey instead. -- John On Nov 17, 2014, at 2:49 PM, Luthaf lut...@luthaf.fr wrote: Hello ! Is there a way to catch an exception by type in Julia ? Coming from python, I am very tempted to do this kind of things: ``` try # Access a dict here catch e if isa(KeyError, e) # Handle the KeyError, as I know what to do in that case else # Re-throw to upper level throw(e) end ``` But that is a lot of boilerplate when used many times. Maybe it is not the Julia way to express this kind of problem (handling one type of exception, while letting the others go up), but I could not find something about it. I can not just remove all the exceptions as some of them are thrown by Base. Thank for yours answers ! Guillaume
Re: [julia-users] help() doesn't work in REPL in julia-0.4.0-1439 from julianightlies PPA
Well, as of julia-0.4.0-1454~ubuntu14.04.1 It still happens. By the way, I was wrong about which package to blame, it's not julia-doc but julia itself (julia-doc contains the manual and other documentation, but the basic help strings the REPL help() functions prints are actually in the julia package). I'm pretty sure this is a .deb package problem and not an issue with Julia itself, so I'll tell the PPA maintainer. On Sunday, November 16, 2014 10:21:01 PM UTC-5, Isaiah wrote: If this is not fixed now, please file an issue. On Thu, Nov 13, 2014 at 5:30 PM, Omar Antolín Camarena omar.a...@gmail.com javascript: wrote: The julianightlies PPA at some point moved the documentation into a separate package, julia-doc, that seems to have a slight problem: julia help(help) ERROR: could not open file /usr/share/doc/julia/helpdb.jl in include at ./boot.jl:242 in include_from_node1 at ./loading.jl:128 in evalfile at loading.jl:168 (repeats 2 times) in init_help at help.jl:39 in help at help.jl:89 in help at help.jl:166 in help at help.jl:181 My guess is that the problem is that, as is common in Debian packages, the files in the julia-doc package are gzipped and the help() function doesn't expect that. Indeed, after running sudo gunzip /usr/share/doc/julia/helpdb.jl.gz the help() function seems to work in the REPL again.
Re: [julia-users] SymTridiagonal
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack wrote:
The input should be two Vectors, but your first argument is a Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com javascript:
:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching
convert(::Type{SymTridiagonal{T}}, ::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
Re: [julia-users] SymTridiagonal
Semantically, ones(n,1) creates a vector and not a matrix.
I'd rather say that in MATLAB ones(n,1) creates a vector.
This has been discussed many times on the list and in issues. In
particular, see the famous https://github.com/JuliaLang/julia/issues/4774.
In Julia, Vector{T} and Matrix{T} are aliases for Array{T,1} and Array{T,2}
which I think is reasonable. The questions are to what extend a nx1 Matrix
should work similarly to a Vector and a Vector should work similarly to a
nx1 Matrix. That is the discussion in the issue mentioned, and it is
actually more subtle than one would expect.
2014-11-17 12:04 GMT-05:00 Eka Palamadai ekanat...@gmail.com:
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack wrote:
The input should be two Vectors, but your first argument is a Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching convert(::Type{SymTridiagonal{T}},
::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
[julia-users] Re: Arrayfire and Julia
Happy to see thus reaction from a core julia developer. Hope julia makes parallel programming on CPUs and GPUs easier.
Re: [julia-users] SymTridiagonal
which I think is reasonable is a subjective argument.
It would be helpful if the type system is intuitive and non-confusing to
programmers.
On Monday, November 17, 2014 12:24:58 PM UTC-5, Andreas Noack wrote:
Semantically, ones(n,1) creates a vector and not a matrix.
I'd rather say that in MATLAB ones(n,1) creates a vector.
This has been discussed many times on the list and in issues. In
particular, see the famous https://github.com/JuliaLang/julia/issues/4774
.
In Julia, Vector{T} and Matrix{T} are aliases for Array{T,1} and
Array{T,2} which I think is reasonable. The questions are to what extend a
nx1 Matrix should work similarly to a Vector and a Vector should work
similarly to a nx1 Matrix. That is the discussion in the issue mentioned,
and it is actually more subtle than one would expect.
2014-11-17 12:04 GMT-05:00 Eka Palamadai ekan...@gmail.com javascript:
:
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack wrote:
The input should be two Vectors, but your first argument is a Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching convert(::Type{SymTridiagonal{T}},
::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
Re: [julia-users] SymTridiagonal
What's intuitive is very dependent upon your background. If you're coming from
Matlab, for example, everything is a matrix and Matlab does this
extraordinarily-confusing thing:
ones(3,3,3) gives me a 3d array;
ones(3,3) gives me a 2d array;
but
ones(3)
ans =
1 1 1
1 1 1
1 1 1
Why the heck did it give me a 2d matrix when I asked for a 1-dimensional
vector of 1s?
Julia is much more consistent: the dimensionality of the created object is
equal to the number of indices you supply. If you ask for something that's
3x1, that's the size you'll get out; perforce, that is a 2d array.
--Tim
On Monday, November 17, 2014 09:41:10 AM Eka Palamadai wrote:
which I think is reasonable is a subjective argument.
It would be helpful if the type system is intuitive and non-confusing to
programmers.
On Monday, November 17, 2014 12:24:58 PM UTC-5, Andreas Noack wrote:
Semantically, ones(n,1) creates a vector and not a matrix.
I'd rather say that in MATLAB ones(n,1) creates a vector.
This has been discussed many times on the list and in issues. In
particular, see the famous https://github.com/JuliaLang/julia/issues/4774
.
In Julia, Vector{T} and Matrix{T} are aliases for Array{T,1} and
Array{T,2} which I think is reasonable. The questions are to what extend a
nx1 Matrix should work similarly to a Vector and a Vector should work
similarly to a nx1 Matrix. That is the discussion in the issue mentioned,
and it is actually more subtle than one would expect.
2014-11-17 12:04 GMT-05:00 Eka Palamadai ekan...@gmail.com javascript:
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack wrote:
The input should be two Vectors, but your first argument is a Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching
convert(::Type{SymTridiagonal{T}},
::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
Re: [julia-users] SymTridiagonal
I don't know what matlab does.
As a user, ones(n,1) and ones(n) both return me a vector, and it is
confusing to find that ones(n,1) != ones(n).
On Monday, November 17, 2014 12:53:25 PM UTC-5, Tim Holy wrote:
What's intuitive is very dependent upon your background. If you're coming
from
Matlab, for example, everything is a matrix and Matlab does this
extraordinarily-confusing thing:
ones(3,3,3) gives me a 3d array;
ones(3,3) gives me a 2d array;
but
ones(3)
ans =
1 1 1
1 1 1
1 1 1
Why the heck did it give me a 2d matrix when I asked for a 1-dimensional
vector of 1s?
Julia is much more consistent: the dimensionality of the created object is
equal to the number of indices you supply. If you ask for something that's
3x1, that's the size you'll get out; perforce, that is a 2d array.
--Tim
On Monday, November 17, 2014 09:41:10 AM Eka Palamadai wrote:
which I think is reasonable is a subjective argument.
It would be helpful if the type system is intuitive and non-confusing to
programmers.
On Monday, November 17, 2014 12:24:58 PM UTC-5, Andreas Noack wrote:
Semantically, ones(n,1) creates a vector and not a matrix.
I'd rather say that in MATLAB ones(n,1) creates a vector.
This has been discussed many times on the list and in issues. In
particular, see the famous
https://github.com/JuliaLang/julia/issues/4774
.
In Julia, Vector{T} and Matrix{T} are aliases for Array{T,1} and
Array{T,2} which I think is reasonable. The questions are to what
extend a
nx1 Matrix should work similarly to a Vector and a Vector should work
similarly to a nx1 Matrix. That is the discussion in the issue
mentioned,
and it is actually more subtle than one would expect.
2014-11-17 12:04 GMT-05:00 Eka Palamadai ekan...@gmail.com
javascript:
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack wrote:
The input should be two Vectors, but your first argument is a Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching
convert(::Type{SymTridiagonal{T}},
::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
Re: [julia-users] SymTridiagonal
Your best bet, then, is to decide as quickly as possible whether you want to
use Julia. If you start reading here:
http://docs.julialang.org/en/latest/manual/faq/#what-does-type-stable-mean
you'll maximize your chances of quickly discovering other things that will
likely annoy you :-). While only the section I directly linked to is necessary
to understand why `ones(n,1)` can't return a Vector, you should also be sure
to read the next 2 sections on DomainErrors and machine arithmetic, just to
make sure you've drunk the full cup's worth of annoyance.
Then you'll be in a good position to make an informed judgment about whether
you want to accept the hassles in exchange for the benefits the type system
provides.
Best,
--Tim
On Monday, November 17, 2014 10:02:07 AM Eka Palamadai wrote:
I don't know what matlab does.
As a user, ones(n,1) and ones(n) both return me a vector, and it is
confusing to find that ones(n,1) != ones(n).
On Monday, November 17, 2014 12:53:25 PM UTC-5, Tim Holy wrote:
What's intuitive is very dependent upon your background. If you're coming
from
Matlab, for example, everything is a matrix and Matlab does this
extraordinarily-confusing thing:
ones(3,3,3) gives me a 3d array;
ones(3,3) gives me a 2d array;
but
ones(3)
ans =
1 1 1
1 1 1
1 1 1
Why the heck did it give me a 2d matrix when I asked for a 1-dimensional
vector of 1s?
Julia is much more consistent: the dimensionality of the created object is
equal to the number of indices you supply. If you ask for something that's
3x1, that's the size you'll get out; perforce, that is a 2d array.
--Tim
On Monday, November 17, 2014 09:41:10 AM Eka Palamadai wrote:
which I think is reasonable is a subjective argument.
It would be helpful if the type system is intuitive and non-confusing to
programmers.
On Monday, November 17, 2014 12:24:58 PM UTC-5, Andreas Noack wrote:
Semantically, ones(n,1) creates a vector and not a matrix.
I'd rather say that in MATLAB ones(n,1) creates a vector.
This has been discussed many times on the list and in issues. In
particular, see the famous
https://github.com/JuliaLang/julia/issues/4774
.
In Julia, Vector{T} and Matrix{T} are aliases for Array{T,1} and
Array{T,2} which I think is reasonable. The questions are to what
extend a
nx1 Matrix should work similarly to a Vector and a Vector should work
similarly to a nx1 Matrix. That is the discussion in the issue
mentioned,
and it is actually more subtle than one would expect.
2014-11-17 12:04 GMT-05:00 Eka Palamadai ekan...@gmail.com
javascript:
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack wrote:
The input should be two Vectors, but your first argument is a Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching
convert(::Type{SymTridiagonal{T}},
::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
[julia-users] Re: try/catch by exception type
What about using multiple dispatch? handle(::ExceptionType1, ::MyObject) = ... handle(::ExceptionType2, ::MyObject) = ... handle(::ExceptionType3, ::MyObject) = ... handle(e::Exception, ::MyObject) = rethrow(e) try # Access a dict here catch e handle(e, dict) end Am Montag, 17. November 2014 15:49:24 UTC+1 schrieb Luthaf: Hello ! Is there a way to catch an exception by type in Julia ? Coming from python, I am very tempted to do this kind of things: ``` try # Access a dict here catch e if isa(KeyError, e) # Handle the KeyError, as I know what to do in that case else # Re-throw to upper level throw(e) end ``` But that is a lot of boilerplate when used many times. Maybe it is not the Julia way to express this kind of problem (handling one type of exception, while letting the others go up), but I could not find something about it. I can not just remove all the exceptions as some of them are thrown by Base. Thank for yours answers ! Guillaume
Re: [julia-users] SymTridiagonal
On Monday, November 17, 2014 12:02:07 PM UTC-6, Eka Palamadai wrote:
I don't know what matlab does.
As a user, ones(n,1) and ones(n) both return me a vector, and it is
confusing to find that ones(n,1) != ones(n).
As Andreas and Tim have tried to say, your claim that ones(n,1) and
ones(n) both return me a vector is incorrect.
julia ones(3,1)
3x1 Array{Float64,2}:
1.0
1.0
1.0
julia ones(3)
3-element Array{Float64,1}:
1.0
1.0
1.0
The first is a matrix with 3 rows and 1 column. The second is a vector.
That's why they are not equal.
On Monday, November 17, 2014 12:53:25 PM UTC-5, Tim Holy wrote:
What's intuitive is very dependent upon your background. If you're coming
from
Matlab, for example, everything is a matrix and Matlab does this
extraordinarily-confusing thing:
ones(3,3,3) gives me a 3d array;
ones(3,3) gives me a 2d array;
but
ones(3)
ans =
1 1 1
1 1 1
1 1 1
Why the heck did it give me a 2d matrix when I asked for a 1-dimensional
vector of 1s?
Julia is much more consistent: the dimensionality of the created object
is
equal to the number of indices you supply. If you ask for something
that's
3x1, that's the size you'll get out; perforce, that is a 2d array.
--Tim
On Monday, November 17, 2014 09:41:10 AM Eka Palamadai wrote:
which I think is reasonable is a subjective argument.
It would be helpful if the type system is intuitive and non-confusing
to
programmers.
On Monday, November 17, 2014 12:24:58 PM UTC-5, Andreas Noack wrote:
Semantically, ones(n,1) creates a vector and not a matrix.
I'd rather say that in MATLAB ones(n,1) creates a vector.
This has been discussed many times on the list and in issues. In
particular, see the famous
https://github.com/JuliaLang/julia/issues/4774
.
In Julia, Vector{T} and Matrix{T} are aliases for Array{T,1} and
Array{T,2} which I think is reasonable. The questions are to what
extend a
nx1 Matrix should work similarly to a Vector and a Vector should work
similarly to a nx1 Matrix. That is the discussion in the issue
mentioned,
and it is actually more subtle than one would expect.
2014-11-17 12:04 GMT-05:00 Eka Palamadai ekan...@gmail.com
javascript:
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack wrote:
The input should be two Vectors, but your first argument is a
Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching
convert(::Type{SymTridiagonal{T}},
::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
Re: [julia-users] BinDeps: How to access include path?
Le lundi 17 novembre 2014 à 08:42 -0500, Erik Schnetter a écrit : On Mon, Nov 17, 2014 at 3:15 AM, Milan Bouchet-Valat nalimi...@club.fr wrote: Le dimanche 16 novembre 2014 à 15:41 -0500, Erik Schnetter a écrit : I see. I was afraid that the C structs may change in between different versions of the library. If the structs are re-analyzed when the wrapper is installed, this would not be an issue. Stable libraries shouldn't change their structs, except when releasing new major versions. In that case even classic C programs will break, so it's not really something to worry about specifically for Julia code. Ideally your package should download a version series known to work, so things don't break even a breaking version is released. In this context, most library users will build from source. I would not expect binary compatibility. So C programs would not break. Well, if they access the fields of the structs they may well break. But anyway, you will get more or less as much stability when building from source as when using binaries, as AFAIK libraries usually break API and ABI at the same time. It is not possible to choose a particular version when installing e.g. via apt-get or homebrew. As regards apt-get at least usually only one version is available for a given distro, so if it's not the correct one you need to find another method. I don't know about homebrew. And if a new major version is released and the structs change, other things are likely to change too, so that your code will break anyway. That's what SOVERSION is used for usually. I'm not choosing a particular soversion either. I don't think that is even supported with BinDeps? Indeed, I don't think it is, but I'd argue in the long term some version checking should be implemented. Regards -erik My two cents On Sun, Nov 16, 2014 at 2:58 PM, Tim Holy tim.h...@gmail.com wrote: The only person who needs to use Clang.jl is the package developer. Clang writes *.jl files that save you the trouble of keyboarding all the julia equivalents of the C structs. It also generates the ccall wrappers. You then write your package on top of those generated files. Clang is therefore only a dependency for the developer, not for the user. That said, if you've already written all your types, Clang won't provide you with much benefit. Quite a few packages use Clang, but one I've been involved with is CUDArt.jl. --Tim On Sunday, November 16, 2014 01:01:42 PM Erik Schnetter wrote: Thanks for all the pointers! Things are now working fine without wrapper functions but with 120 lines of immutable type declarations. Clang.jl sounds interesting, but would probably make hwloc.jl too difficult to use if it is a prerequisite. Let's see how often the structs of hwloc change with future versions. hwloc is a portable library that determines the number of cores (and many other properties) of the local machine: see https://github.com/eschnett/hwloc.jl. -erik On Sun, Nov 16, 2014 at 12:41 PM, Jake Bolewski jakebolew...@gmail.com wrote: No need to flatten if everything is immutable. This file has some examples of wrapping structs of structs https://github.com/jakebolewski/LibGit2.jl/blob/cac78b5c03531b5afbcb0ae042 538dd351527752/src/types.jl#L19. On Sunday, November 16, 2014 12:33:39 PM UTC-5, Tim Holy wrote: AFAIK no need to flatten. Since Isaiah didn't advertise it himself, I'll mention his Clang.jl package, which I've found to be a big help in such situations. It's just possible you may not need any C glue code. --Tim On Sunday, November 16, 2014 12:15:50 PM Erik Schnetter wrote: Thanks for the pointers on immutable types. Is it possible to access structs inside structs this way? That a struct inside a struct, not a pointer inside a struct. Is an immutable type again a bits type? Or do I need to flatten the structs to make this work? -erik On Sun, Nov 16, 2014 at 11:16 AM, Isaiah Norton isaiah...@gmail.com wrote: The library defines some C structs that are part of the API. My current approach uses wrapper C functions to access struct elements. Is there a better way? It depends how complicated are the structs. Structs can often be reflected as Julia types, as long as isbits(TypeName) == true, and accessed with unsafe_load (see for example PyCall's definitions of PyObject_*). Some tricky spots include fixed-size arrays and unions (strictly speaking you can make an aggregate element having the maximal size in the union, and then do bitshifts manually. but there is no automatic support). There is also the StrPack.jl package, which calculates the appropriate
[julia-users] capture STDIN character by character, turning off REPL.
I would like to capture STDIN character by character, without any characters going to the REPL. Right now if I have function stdio_producer() while true c = read(STDIN,Char) if c == '\e' println(Done) break else print(:); print(c) produce(c) end end end it only captures every other character. Thanks, Gustavo
Re: [julia-users] SymTridiagonal
As a user, ones(n,1) and ones(n) both return me a vector, and it is confusing to find that ones(n,1) != ones(n) I was where you are now a few months ago. It's a learning cure thing, I think, because now I don't make that mistake anymore or I'm like, oh yea, of course and change it 2 seconds later. But to a new user it can be uninviting and not easily solved with just more documentation. The question to me is what is the tradeoff? For a semi-experienced user like me, it looks like Julia is trying to pick a spot of convenience while trying to retain access to optimization. The convenience part is the REPL, no requirement for variable type declaration, no function type return declaration, and on the other hand types and an options for variable type declaration to allow the JIT to better optimize. Your experience sits right where those to conflicting things are fighting it out right now, and this wall-of-text doesn't help you out any. I think this might be helped by having more verbose error messages (optionally).
[julia-users] Dimension Independent Array Access
I would appreciate advise how to best implement the following:
I have an N^d array A, where the dimension d depends on the
application (d \in \{1, 2, 3\}). Somewhere else I have a list L which is a
d x M array or integers corresponding to points/elements in this array,
e.g. if
L[:, 3] == [5, 2, 8]
then A would be N^3 and
M[L[1,3], L[2,3], L[3,3]]
would read
A[5, 2, 8]
The difficulty is that I would like to read from A in a *fast*
dimension-independent way. My only idea was to define a type and overload
getindex, setindex! ; see below. If I tested this correctly, then I only
loose a factor of two in terms of both speed and memory.
I am slightly worried that it is a non-standard access to a standard array.
+ I'd love to get that factor 2 back. Hence, I'd like to know whether
there are faster / more elegant alternatives, or just other ways of doing
this?
module damod
export darray
immutable type darray
data
dim
end
function darray(A)
return darray(A, length(size(A)))
end
export getindex
function getindex(A::darray, ii)
if A.dim == 2
return A.data[ii[1], ii[2]]
elseif A.dim == 3
return A.data[ii[1], ii[2], ii[3]]
end
end
end
using damod
N = 100
a = darray(rand(N, N, N))
@timefor n = 1:N, m=1:N, k=1:N; dummy = a[[n,m,k]] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a.data[n,m,k] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a[[n,m,k]] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a.data[n,m,k] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a[[n,m,k]] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a.data[n,m,k] end
elapsed time: 0.470031905 seconds (128633244 bytes allocated, 11.90% gc time)
elapsed time: 0.209358484 seconds (48565624 bytes allocated, 8.98% gc time)
elapsed time: 0.42617227 seconds (128641396 bytes allocated, 16.12% gc time)
elapsed time: 0.21529278 seconds (48565624 bytes allocated, 10.65% gc time)
elapsed time: 0.457094657 seconds (128565624 bytes allocated, 14.37% gc time)
elapsed time: 0.201969234 seconds (48565624 bytes allocated, 10.24% gc time)
[julia-users] Re: Help with metaprogramming
Thanks Simon and Tim.
Re: [julia-users] Dimension Independent Array Access
You can do a lot better by dispatching on the dimension of A. Something like
this:
mygetindex{T}(A::AbstractArray{T,2}, L, Lcol) = A[L[1,Lcol], L[2,Lcol]]
mygetindex{T}(A::AbstractArray{T,3}, L, Lcol) = A[L[1,Lcol], L[2,Lcol],
L[3,Lcol]]
Put your timing loops in a function so you take advantage of inference. For
example:
function accesseach(A, L)
dummy = 0.0
for Lcol = 1:size(L, 2)
dummy = mygetindex(A, L, Lcol)
end
dummy
end
julia L = Array(Int, 3, N^3);
julia idx = 1; for k = 1:N, j = 1:N, i = 1:N
L[1,idx] = i
L[2,idx] = j
L[3,idx] = k
idx += 1
end
julia accesseach(A, L)
0.4109242495731802
julia @time accesseach(A, L)
elapsed time: 0.006461321 seconds (13824 bytes allocated)
0.4109242495731802
As you can see, quite a bit faster.
--Tim
On Monday, November 17, 2014 12:01:08 PM Christoph Ortner wrote:
I would appreciate advise how to best implement the following:
I have an N^d array A, where the dimension d depends on the
application (d \in \{1, 2, 3\}). Somewhere else I have a list L which is a
d x M array or integers corresponding to points/elements in this array,
e.g. if
L[:, 3] == [5, 2, 8]
then A would be N^3 and
M[L[1,3], L[2,3], L[3,3]]
would read
A[5, 2, 8]
The difficulty is that I would like to read from A in a *fast*
dimension-independent way. My only idea was to define a type and overload
getindex, setindex! ; see below. If I tested this correctly, then I only
loose a factor of two in terms of both speed and memory.
I am slightly worried that it is a non-standard access to a standard array.
+ I'd love to get that factor 2 back. Hence, I'd like to know whether
there are faster / more elegant alternatives, or just other ways of doing
this?
module damod
export darray
immutable type darray
data
dim
end
function darray(A)
return darray(A, length(size(A)))
end
export getindex
function getindex(A::darray, ii)
if A.dim == 2
return A.data[ii[1], ii[2]]
elseif A.dim == 3
return A.data[ii[1], ii[2], ii[3]]
end
end
end
using damod
N = 100
a = darray(rand(N, N, N))
@timefor n = 1:N, m=1:N, k=1:N; dummy = a[[n,m,k]] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a.data[n,m,k] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a[[n,m,k]] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a.data[n,m,k] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a[[n,m,k]] end
@timefor n = 1:N, m=1:N, k=1:N; dummy = a.data[n,m,k] end
elapsed time: 0.470031905 seconds (128633244 bytes allocated, 11.90% gc
time) elapsed time: 0.209358484 seconds (48565624 bytes allocated, 8.98% gc
time) elapsed time: 0.42617227 seconds (128641396 bytes allocated, 16.12%
gc time) elapsed time: 0.21529278 seconds (48565624 bytes allocated, 10.65%
gc time) elapsed time: 0.457094657 seconds (128565624 bytes allocated,
14.37% gc time) elapsed time: 0.201969234 seconds (48565624 bytes
allocated, 10.24% gc time)
Re: [julia-users] SymTridiagonal
Thanks.
Fortunately (or unfortunately) i have to use julia, and will have to make
noise
where something is confusing.
On Monday, November 17, 2014 1:26:09 PM UTC-5, Tim Holy wrote:
Your best bet, then, is to decide as quickly as possible whether you want
to
use Julia. If you start reading here:
http://docs.julialang.org/en/latest/manual/faq/#what-does-type-stable-mean
you'll maximize your chances of quickly discovering other things that will
likely annoy you :-). While only the section I directly linked to is
necessary
to understand why `ones(n,1)` can't return a Vector, you should also be
sure
to read the next 2 sections on DomainErrors and machine arithmetic, just
to
make sure you've drunk the full cup's worth of annoyance.
Then you'll be in a good position to make an informed judgment about
whether
you want to accept the hassles in exchange for the benefits the type
system
provides.
Best,
--Tim
On Monday, November 17, 2014 10:02:07 AM Eka Palamadai wrote:
I don't know what matlab does.
As a user, ones(n,1) and ones(n) both return me a vector, and it is
confusing to find that ones(n,1) != ones(n).
On Monday, November 17, 2014 12:53:25 PM UTC-5, Tim Holy wrote:
What's intuitive is very dependent upon your background. If you're
coming
from
Matlab, for example, everything is a matrix and Matlab does this
extraordinarily-confusing thing:
ones(3,3,3) gives me a 3d array;
ones(3,3) gives me a 2d array;
but
ones(3)
ans =
1 1 1
1 1 1
1 1 1
Why the heck did it give me a 2d matrix when I asked for a
1-dimensional
vector of 1s?
Julia is much more consistent: the dimensionality of the created
object is
equal to the number of indices you supply. If you ask for something
that's
3x1, that's the size you'll get out; perforce, that is a 2d array.
--Tim
On Monday, November 17, 2014 09:41:10 AM Eka Palamadai wrote:
which I think is reasonable is a subjective argument.
It would be helpful if the type system is intuitive and
non-confusing to
programmers.
On Monday, November 17, 2014 12:24:58 PM UTC-5, Andreas Noack wrote:
Semantically, ones(n,1) creates a vector and not a matrix.
I'd rather say that in MATLAB ones(n,1) creates a vector.
This has been discussed many times on the list and in issues. In
particular, see the famous
https://github.com/JuliaLang/julia/issues/4774
.
In Julia, Vector{T} and Matrix{T} are aliases for Array{T,1} and
Array{T,2} which I think is reasonable. The questions are to what
extend a
nx1 Matrix should work similarly to a Vector and a Vector should
work
similarly to a nx1 Matrix. That is the discussion in the issue
mentioned,
and it is actually more subtle than one would expect.
2014-11-17 12:04 GMT-05:00 Eka Palamadai ekan...@gmail.com
javascript:
Semantically, ones(n,1) creates a vector and not a matrix.
Why is ones(n,1) different from ones(n)?
The type system is very confusing and non-intuitive.
On Sunday, November 16, 2014 7:28:28 PM UTC-5, Andreas Noack
wrote:
The input should be two Vectors, but your first argument is a
Matrix
2014-11-16 19:25 GMT-05:00 Eka Palamadai ekan...@gmail.com:
SymTridiagonal does not seem to work properly.
For e.g, the following snippet fails.
julia n=10 ;
A=SymTridiagonal(2*ones(n,1), -1*ones(n-1));
ERROR: `convert` has no method matching
convert(::Type{SymTridiagonal{T}},
::Array{Float64,2}, ::Array{Float64,1})
in call at base.jl:34
Any thoughts?
Re: [julia-users] Concrete parametric types
Yeah realised that just after I posted my message :S
On Monday, 17 November 2014 13:07:42 UTC+11, Jameson wrote:
: is the subtype operator
:: is the isa operator
You can't (currently) express conditions of the form you want (except as
assertions in the constructor)
On Sun, Nov 16, 2014 at 8:03 PM i.cos...@me.com javascript: wrote:
I've got a parametric type:
immutable SplineInterpolator{N:Integer} : Interpolator1D end
NB:
abstract Interpolator
abstract Interpolator1D : Interpolator
Now, when I try to create the concrete type...
SplineInterpolator{3}
...I get the following error message:
ERROR: type: SplineInterpolator: in N, expected N:Integer, got Int64
NB that typeof(3) : Integer returns true
Why am I get this error message? Using Julia 0.3.2 on OS X 10.10
Thanks
[julia-users] Metaprogramming: what's going on here/is there a better way?
I've seen this type of function generation in other packages, and wanted to
try it for myself. This file in Twitter.jl has 5 functions with the same
overall structure:
https://github.com/randyzwitch/Twitter.jl/blob/master/src/help.jl
Here's what I ended up doing, which works, but I've got no idea why I had
to write the string interpolation the way I did. I went through many
permutations here, so if someone could explain why the
`:($(string(https://api.twitter.com/1.1/;,
$endp))` line needs to be written that way for string interpolation, it
would be appreciated. And any improvements in general are welcome.
funcname = (:get_help_configuration, :get_help_languages,
:get_help_privacy, :get_help_tos, :get_application_rate_limit_status)
endpoint = (help/configuration.json, help/languages.json,
help/privacy.json, help/tos.json, application/rate_limit_status.json)
for (func, endp) in Dict(funcname, endpoint)
@eval begin function ($func)(; options=Dict{String, String}())
r = get_oauth(:($(string(https://api.twitter.com/1.1/;, $endp))),
options)
return r.status == 200 ? JSON.parse(r.data) : r
end
end
end
Thanks!
Re: [julia-users] Metaprogramming: what's going on here/is there a better way?
It would appear so...I swear that I tried that, but I guess I didn't try
that permutation!
So what's it about the @eval macro that doesn't allow for regular string
interpolation, that I have to use the string() function instead of an
inline $?
On Monday, November 17, 2014 9:50:55 PM UTC-5, Jacob Quinn wrote:
Can you just do
funcname = (:get_help_configuration, :get_help_languages,
:get_help_privacy, :get_help_tos, :get_application_rate_limit_status)
endpoint = (help/configuration.json, help/languages.json,
help/privacy.json, help/tos.json, application/rate_limit_status.json)
for (func, endp) in Dict(funcname, endpoint)
@eval begin function ($func)(; options=Dict{String, String}())
r = get_oauth(string(https://api.twitter.com/1.1/;, $endp),
options)
return r.status == 200 ? JSON.parse(r.data) : r
end
end
end
?
On Mon, Nov 17, 2014 at 9:35 PM, Randy Zwitch randy@fuqua.duke.edu
javascript: wrote:
I've seen this type of function generation in other packages, and wanted
to try it for myself. This file in Twitter.jl has 5 functions with the same
overall structure:
https://github.com/randyzwitch/Twitter.jl/blob/master/src/help.jl
Here's what I ended up doing, which works, but I've got no idea why I had
to write the string interpolation the way I did. I went through many
permutations here, so if someone could explain why the `:($(string(
https://api.twitter.com/1.1/;, $endp))` line needs to be written that
way for string interpolation, it would be appreciated. And any improvements
in general are welcome.
funcname = (:get_help_configuration, :get_help_languages,
:get_help_privacy, :get_help_tos, :get_application_rate_limit_status)
endpoint = (help/configuration.json, help/languages.json,
help/privacy.json, help/tos.json, application/rate_limit_status.json)
for (func, endp) in Dict(funcname, endpoint)
@eval begin function ($func)(; options=Dict{String, String}())
r = get_oauth(:($(string(https://api.twitter.com/1.1/;,
$endp))), options)
return r.status == 200 ? JSON.parse(r.data) : r
end
end
end
Thanks!
[julia-users] Re: Arrayfire and Julia
May be somehow we will be able to integrate this with Julia so well that we will always have the first-mover advantage ;) -zahir On Monday, 17 November 2014 00:35:02 UTC-5, Viral Shah wrote: Wow, I did not know about this. We certainly should leverage this. The API looks easy to call too from C, from the examples on their github page, and that is good news for us. -viral On Saturday, November 15, 2014 5:22:17 AM UTC+5:30, Zahirul ALAM wrote: thanks. I have been using Julia for five months or so. I have not used the ccall function ever. I will try it out as soon as I am done my current project. On Friday, 14 November 2014 18:29:09 UTC-5, Tony Kelman wrote: I had the same response, great that they open-sourced the library, and wrapping it in a Julia package would be interesting and not that hard to do. It just needs someone to put a little bit of time into it. How long have you been using Julia? Try compiling the arrayfire library and writing a few ccall's into it from Julia, it should be a pretty direct translation of their C API. On Friday, November 14, 2014 2:59:01 PM UTC-8, Zahirul ALAM wrote: Arrayfire has a large library for GPU computing. They have decided to make their codes open source. I am hoping that someone much more able than me will write some sort of wrapper to use the Arrayfire library in Julia. Here is the github link: https://github.com/arrayfire/arrayfire
Re: [julia-users] Metaprogramming: what's going on here/is there a better way?
On Tuesday, November 18, 2014 12:59:50 PM UTC+10, Randy Zwitch wrote: It would appear so...I swear that I tried that, but I guess I didn't try that permutation! So what's it about the @eval macro that doesn't allow for regular string interpolation, that I have to use the string() function instead of an inline $? The @eval macro quotes its arguments. Within quoted code substitution is by Expr objects, not text, so it won't substitute into a string. Cheers Lex
[julia-users] Re: Dimension Independent Array Access
Dear Tim, Many thanks - this was illuminating and helped me understand the bottleneck. But it seems I did not give the correct amount of information in my question. In fact, I also want to access A[ idx ] for general index vectors idx that do not necessarily belong to the L array. For example, I may be given a set of vectors in an array R, d x K and want to access A[ L[:, n] + R[:, m] ], then loop over n, m and do some calculations. These sort of calculations are at the core of the code and I want the syntax as simple and intuitive as possible to avoid making mistakes. Can I confirm something: is it true that when I do A[ L[:, Lcol] ] then L[:, Lcol] creates a new d-dimensional array and copies that data into that array? Can this at all be avoided? (I believe this is the bottlenbeck in what I do?) I've now tried this again in the form module damod export darray3 immutable darray3 data end export getindex getindex(a::darray3, idx) = a.data[idx[1], idx[2],idx[3]] end This makes only a small difference. elapsed time: 0.163860853 seconds (10480 bytes allocated, 28.46% gc time) vs some variant of your code: elapsed time: 0.003819349 seconds (96 bytes allocated) What strikes me as odd here is that *104,000,080 bytes* are being allocated. If L is an Int16 array, then it should be no more than 100^3 x 3 x 8 = 6,000,000 ? Any further thoughts will be much appreciated. Christoph
