I have a problem with which I am struggling without success. I have two functions, ps1 and ps2, which are essentially the same:
function ps1(A::T) where {T <: PeriodicSymbolicMatrix}
return T
end
function ps2(A::T) where {T <: PeriodicSymbolicMatrix}
return T
end
The only difference is that ps1 is exported by the package PeriodicSystems, while ps2 is simply compiled within REPL. The methods associated to these functions are:
julia> methods(ps1)
# 1 method for generic function "ps1" from PeriodicSystems:
[1] ps1(A::T) where T<:PeriodicMatrices.PeriodicSymbolicMatrix
@ C:\Users\Andreas\Documents\software\Julia\PeriodicSystems.jl\src\ps.jl:127
julia> methods(ps2)
# 1 method for generic function "ps2" from Main:
[1] ps2(A::T) where T<:PeriodicSymbolicMatrix
@ REPL[13]:1
The type PeriodicSymbolicMatrix is defined in another package PeriodicMatrices and imported into PeriodicSystems. It satisfies
Both PeriodicSystems and PeriodicMatrices define their own PeriodicMatrix type, so probably that’s the problem.
julia> using PeriodicMatrices, PeriodicSystems
julia> a = PeriodicSymbolicMatrix(rand(10), 1.0; nperiod=3)
WARNING: both PeriodicSystems and PeriodicMatrices export "PeriodicSymbolicMatrix"; uses of it in module Main must be qualified
Probably one of the packages should be a dependency of the other.
It seems that you have moved PeriodicMatrices from an independent module inside PeriodicSystem to a separate package, and then you need to import the definitions from the new package and delete the, I think, deprecated definitions inside the PeriodicSystems package. That is, probably this file should not exist anymore, and be replaced by using the new PeriodicMatrices package.
Indeed, I extracted PeriodicMatrices from the PeriodicSystems, because of the general importance of periodic matrices in many domains (and not only in control problems). I am working on a new version of PeriodicSystems, without the PeriodicMatrices part. I removed all duplicities, (e.g., the definitions of types). I hope I did the splitting correctly, but I will check once again, since it is possible, that some remains still exist.
julia> function ps2(A::T) where {T <: PeriodicSymbolicMatrix}
return T
end
ps2 (generic function with 1 method)
julia> ps2(A)
PeriodicSymbolicMatrix{:c, Num, Matrix{Num}}
I wonder what is wrong with my code … or with Julia!
For me your MWE does not work so well. If I use your definition of the package in REPL, I have to do
using .PeriodicSystems
(otherwise it asks me to install your package). But since it does not export ps1, I then have to call
PeriodicSystems.ps1(A)
which just works fine with the local small module you provided.
To me the warning above gives the main indicator, you define the type in two classes, so those are two completely different things (like 2 different person with the same name).
So the PeriodicSystems.PeriodicSymbolicMatrix is a completely different type as PeriodicMatrices.PeriodicSymbolicMatrix. Lets for simplicity and shortage call these P1 and P2, respectively, to disinghuish them though they have the same name. They might even have the same functionality if you copied them over, but that would then still be like two persons having the same name and looking alike.
If you now for example export ps1 but not the first of these two types (P1), but PeriodicMatrices exports its type (P2), then you define ps2 for matrices of type P2. since P2 is exported but P1 is not, a is of type P2.
Hence it will work with ps2 but not the (exported) ps1.
You can see that also in your error message above where both PeriodicMatrices.PeriodicSymbolicMatrix and PeriodicSymbolicMatrix are used, it seems the first is not exported from that package, the second is (from somewhere else).
Indeed, you may have additional stuff defined in your section that is confusing things. The MWE works as expected:
julia> using Symbolics, PeriodicMatrices
julia> module PeriodicSystems
using Symbolics
using PeriodicMatrices
function ps1(A::T) where {T <: PeriodicSymbolicMatrix}
return T
end
end
julia> using .PeriodicSystems
julia> A = PeriodicSymbolicMatrix(rand(10), 1.0)
PeriodicSymbolicMatrix{:c, Num, Matrix{Num}}(Num[0.8533861311322568; 0.9551845692903604; … ; 0.47612337713522335; 0.7758785937837308;;], 1.0, 1)
julia> PeriodicSystems.ps1(A)
PeriodicSymbolicMatrix{:c, Num, Matrix{Num}}
which is exported in the main file and you defined
which is now something completely different and a completely new type. and you also export it in the main project file. So if I now do
julia> using PeriodicSystems, PeriodicMatrices
julia> a = PeriodicSymbolicMatrix(rand(10), 1.0; nperiod=3)
ERROR: UndefVarError: `PeriodicSymbolicMatrix` not defined in `Main`
Hint: It looks like two or more modules export different bindings with this name, resulting in ambiguity. Try explicitly importing it from a particular module, or qualifying the name with the module it should come from.
Hint: a global variable of this name also exists in PeriodicSystems.
Hint: a global variable of this name also exists in PeriodicMatrices.
Stacktrace:
[1] top-level scope
@ REPL[4]:1
So the thing I am actually surprised about is that you are able to define ps2 in REPL, because in the just started REPL I can’t.
julia> function ps2(A::T) where {T <: PeriodicSymbolicMatrix}
return T
end
ERROR: UndefVarError: `PeriodicSymbolicMatrix` not defined in `Main`
Hint: It looks like two or more modules export different bindings with this name, resulting in ambiguity. Try explicitly importing it from a particular module, or qualifying the name with the module it should come from.
Hint: a global variable of this name also exists in PeriodicSystems.
Hint: a global variable of this name also exists in PeriodicMatrices.
Stacktrace:
[1] top-level scope
@ REPL[5]:1
This definition is not made in the new development version of PeriodicSystems, which is only locally available. The MWE example was intended to illustrate the failure of a function ps1, which according to my understanding of dispatch rules, should perform exactly in the same manner as ps2, entered via REPL. Sorry, not being clear in this respect.
The whole package PeriodicSystems has been reduced for the sake of testing to the few lines, which I reproduce again
module PeriodicSystems
using Symbolics
using PeriodicMatrices
function ps1(A::T) where {T <: PeriodicSymbolicMatrix}
return typeof(A)
end
export ps1
end
(this time ps1 is exported)
The MWE fails exactly in the same way as presented in my example.
You need to provide a full runnable example, because the issues you report are not reproducible:
(jl_6RRETd) pkg> st
Status `/tmp/jl_6RRETd/Project.toml`
[dd9cd634] PeriodicMatrices v0.1.1
⌃ [0c5d862f] Symbolics v5.36.0
Info Packages marked with ⌃ have new versions available and may be upgradable.
julia> using PeriodicMatrices
julia> module PeriodicSystems
using Symbolics
using PeriodicMatrices
function ps1(A::T) where {T <: PeriodicSymbolicMatrix}
return T
end
end
julia> using .PeriodicSystems
julia> PeriodicMatrices.PeriodicSymbolicMatrix == PeriodicSymbolicMatrix
true
julia> PeriodicSystems.PeriodicSymbolicMatrix == PeriodicSymbolicMatrix
true
julia> PeriodicSystems.PeriodicSymbolicMatrix
PeriodicSymbolicMatrix
julia> ps1(PeriodicSymbolicMatrix(rand(3), 1.0))
PeriodicSymbolicMatrix{:c, Symbolics.Num, Matrix{Symbolics.Num}}
You probably have other things loaded in your section that are causing the issues.
Hehe, well, that is hard to anticipate; and sure, I can not easily compare to your development version. However, it seems even with the export, I can not reproduce your problem then, because your (minimalistic, MWE) PeriodicSystems avoids exactly the double definition. I get exactly the same result as Leandro.
Note that we both do not have PeriodicSystems installed but just use your small example. But that also means we have to load that with using .PeriodicSystems – you miss the . so you might load your large dev-version or something older?
The first I would try would be start Julia and activate an empty temporary environment and load only the two packages required to reproduce your problem and see if it is still there.
So just to be clear: When you do using . PeriodicSystems all is fine and you get the same as we do, but when you do using PeriodicSystems then it does not?
Then you oversimplified your MWE and I think you still have the PeriodicSymbolicMatrix defined in both packages. But that is of course hard to tell because the development version that is loaded from a folder on your hard drive is nothing I can take a look at.