StackOverflowError during import of DataFrames

[Arkoniak]

It looks like the package uses too broad assumptions. It seems that if package exports already extended function everything go sideways.

When trying to do @usingmerge DataFrames I encounter the following error

using UsingMerge
@usingmerge verbose=3 DataFrames

# skipping lines

# DataFrames.convert conflicts with Base.convert --- adding methods#= /home/skoffer/.julia/
packages/UsingMerge/jdqiP/src/UsingMerge.jl:86 =# @doc #= /home/skoffer/.julia/packages/Usi
ngMerge/jdqiP/src/UsingMerge.jl:86 =# @doc(DataFrames.convert) Base.convert
 and docs
(Base.convert(::Type{T}, x::Ptr) where T <: Integer) = DataFrames.convert(T, x)
(Base.convert(::Type{Pair{A, B}}, x::Pair{A, B}) where {A, B}) = DataFrames.convert(Pair{A,
 B}, x)
(Base.convert(::Type{Pair{A, B}}, x::Pair) where {A, B}) = DataFrames.convert(Pair{A, B}, x
)
Base.convert(::Type{IOContext}, io::IO) = DataFrames.convert(IOContext, io)
(Base.convert(::Type{Some{T}}, x::Some{T}) where T) = DataFrames.convert(Some{T}, x)
(Base.convert(::Type{Some{T}}, x::Some) where T) = DataFrames.convert(Some{T}, x)
(Base.convert(::Type{LinearAlgebra.Adjoint{T, S}}, A::LinearAlgebra.Adjoint) where {T, S})
= DataFrames.convert(LinearAlgebra.Adjoint{T, S}, A)
ERROR:
SYSTEM (REPL): showing an error caused an error
ERROR:
SYSTEM (REPL): caught exception of type StackOverflowError while trying to handle a nested
exception; giving up

[jmichel7]

On Wed, Apr 21, 2021 at 10:04:00AM -0700, Andrey Oskin wrote: It looks like the package uses too broad assumptions. When trying to do ***@***.***` `DataFrames` I encounter the following error

It is a bad idea to usingmerge Dataframes, or any package which was not designed to be used with usingmerge or such that usingmerge that package would overwrite a Base method. You probably want rather to: using DataFrames @usingmerge YourPackage since it is YourPackage that you control. Now, it would be an enhancement of my package to usemerge only some methods in a package and not all. If I am motivated by having some users I might try to program that... ------------------------------------------------------------------------------ Jean MICHEL, Groupes et representations, IMJ-PRG UMR7586 tel.(33)157279144 Bureau 639 Bat. Sophie Germain Case 7012 - 75205 PARIS Cedex 13

[Arkoniak]

Ah, it's a pity, really. Because most of the packages were written without the notion of using @usingmerge :-) It restricts it's usage significantly.

Now, it would be an enhancement of my package to usemerge only some methods
in a package and not all. If I am motivated by having some users I might
try to program that...

I do not know how to motivate you, but I think such an enhancement would be very useful. Also, just a wild idea, maybe some aspects/methodology of SnoopCompiler.jl can be utilized? It seems that they know how to get all methods instances, definitions , arguments and backedges. Maybe this information can be useful for automatical determination what can and can not be reexported.

[Arkoniak]

And anyway, thank you for your package. I've read that long discourse discussion and was not convinced with the argumentation of the current design choice. It's good that someone can try to overcome these restrictions.

[jmichel7]

On Wed, Apr 21, 2021 at 11:53:45AM -0700, Andrey Oskin wrote: And anyway, thank you for your package. I've read that long discourse discussion and was not convinced with the argumentation of the current design choice. It's good that someone can try to overcome these restrictions.

Thank you for your interest. I will let you know when I make a new version of my package. ------------------------------------------------------------------------------ Jean MICHEL, Groupes et representations, IMJ-PRG UMR7586 tel.(33)157279144 Bureau 639 Bat. Sophie Germain Case 7012 - 75205 PARIS Cedex 13

[jmichel7]

Hello,

On Wed, Apr 21, 2021 at 11:52:30AM -0700, Andrey Oskin wrote: I do not know how to motivate you, but I think such an enhancement would be very useful.

I put on GitHub a new version which allows to specify as in `using` a list of names to use/merge: @usingmerge BenchmarkTools: invariants, ratio I tried also to make more checks to work with more packages. For the particular case of `DataFrames`, there is something strange going on: `DataFrames` exports the names `coalesce` and `skipmissing` but does not seem to add methods to them. This produces a recursivity bug if trying to ***@***.***` it. Anyway, if you have any trouble with a package `xxx` I recommend that you try @usingmerge verbose=3 xxx The `verbose=3` will print before executing any command executed by ***@***.***`, enabling you (and me) to better see what happens. Best regards, ------------------------------------------------------------------------------ Jean MICHEL, Groupes et representations, IMJ-PRG UMR7586 tel.(33)157279144 Bureau 639 Bat. Sophie Germain Case 7012 - 75205 PARIS Cedex 13

[Arkoniak]

This is awesome, thank you!
I'll write if I encounter any new issues.