Pkg.add("Distributions") ERROR: key not found: "Sampling"

[floswald]

I get this error when trying to install on a new installation. I suspect the error comes from here rather than GLM where I first registered this: JuliaStats/GLM.jl#81

[lindahua]

This is weird. Neither GLM nor Distributions depend on Sampling (which was removed from METADATA). Can you do Pkg.update() first, and redo the add operation again?

[floswald]

I did that just now - nothing changed. I still get

julia> Pkg.add("Distributions")
ERROR: key not found: "Sampling"

it is very strange because my installation is brand new, so there shouldn't be any weird dependencies dangling around. what would you suggest next?

[floswald]

maybe you can spot anything here?

julia> Pkg.status()
6 required packages:
 - Cairo                         0.2.14
 - DataFrames                    0.5.7
 - Debug                         0.0.3
 - FactCheck                     0.1.2
 - PyPlot                        1.2.9
 - Winston                       0.11.2
14 additional packages:
 - ArrayViews                    0.4.6
 - BSplines                      0.0.3              master
 - BinDeps                       0.2.14
 - Color                         0.2.10
 - DataArrays                    0.2.0
 - GZip                          0.2.13
 - Homebrew                      0.0.6
 - IniFile                       0.2.2
 - PyCall                        0.4.8
 - Reexport                      0.0.1
 - SortingAlgorithms             0.0.1
 - StatsBase                     0.6.1
 - Tk                            0.2.13
 - URIParser                     0.0.2

[lindahua]

Let me look into this.

[lindahua]

I can confirm this -- by using a brand new .julia directory.

[lindahua]

The issue looks like as follows:

Distributions did depend on Sampling temporarily a few months ago (version 0.4.8 and 0.4.9) , and no longer depends on it afterwards. Actually, no package is depending on Sampling now -- that's why I remove it from METADATA.

However, the dependency analysis done within Pkg.add wants to retrieve the information for every dependency ever included in any previous version, thus causing this breakage.

I will reinstate Sampling in METADATA for now to workaround this issue.

However, I think Julia base should allow an old dependency to be removed from METADATA without causing the failure of adding newer version of the package.

cc: @StefanKarpinski

[lindahua]

@floswald I have reinstated Sampling in METADATA. Please see whether it works now (do Pkg.update first).

[nihonjinrxs]

Thanks @lindahua. @floswald thanks for creating the issue.

[IainNZ]

Oh man its Stats/StatsBase all over again. The golden rule is "Never remove a package from METADATA, only deprecate". We need to put this somewhere so people who push direct to METADATA can see it.

[floswald]

Good job @lindahua. All back online.

[johnmyleswhite]

That Golden Rule is kind of a bummer. We should see if we can make things a little more robust to removal of packages.

[StefanKarpinski]

If you remove a package, you have to remove all versions of other packages that depend on it – otherwise those versions will obviously be broken and uninstallable. I'm not sure how else things could possibly work. What you can do if you want to remove a package entirely is simultaneously remove all the versions of other packages that depend on it, or if they didn't really depend on it, remove the dependency from their requirements. Then everything should work.

[kmsquire]

Probably better to continue the discussion in JuliaLang/julia#6669.