Can't build local package

General Usage
1

Here is the error message:

(@v1.11) pkg> build ConvLayer
Precompiling project...
  ✗ ConvLayer
  0 dependencies successfully precompiled in 1 seconds. 423 already precompiled.
  1 dependency errored.
  For a report of the errors see `julia> err`. To retry use `pkg> precompile`

julia> err
PkgPrecompileError: The following 1 direct dependency failed to precompile:

ConvLayer 

Failed to precompile ConvLayer [36137998-1e00-46a9-86ff-ef6b0df5841b] to "/Users/lewis/.julia/compiled/v1.11/ConvLayer/jl_utMxkI".
ERROR: LoadError: UndefVarError: `preptrain` not defined in `ConvLayer`
Stacktrace:
 [1] top-level scope
   @ ~/code/ConvLayer/chatgpt_conv_code/src/ConvLayer.jl:54
 [2] include
   @ ./Base.jl:557 [inlined]
 [3] include_package_for_output(pkg::Base.PkgId, input::String, depot_path::Vector{String}, dl_load_path::Vector{String}, load_path::Vector{String}, concrete_deps::Vector{Pair{Base.PkgId, UInt128}}, source::Nothing)
   @ Base ./loading.jl:2881
 [4] top-level scope
   @ stdin:6
in expression starting at /Users/lewis/code/ConvLayer/chatgpt_conv_code/src/ConvLayer.jl:1
in expression starting at stdin:

Here is line are the lines that include line 54 from the module definition named ConvLayer.jl:

export 
    preptrain,

export is on line 53, preptrain is on line 54.

There is only one source file in this local package and it is included in the module definition:

include("sample_code.jl")

In that source file there certainly is a function called preptrain:

function preptrain(modelspecs::Function, batch_size, mini_batch_size)
    trainset = MNIST(:train)
    testset = MNIST(:test)
<more code follows--not shown>

Sort of hard to solve things like this that seem not to make sense. I’ve compared to another one of my much more complex module/packages. Nothing I can see that I’ve done any differently. This source file worked when I simply included it. Making it part of a module seems to have broken it.

2

In order to help you, I think more information would be necessary (at least I can’t see any obvious errors in what you have posted here). If this is not sensitive code, you could make it publicly available on Github or similar and post a link to it. If you cannot do that, then try removing sections of code from (a copy of) the package and try to pinpoint which small change actually triggers the error. If you do eventually get this reduced to a minimal example that triggers the error then you can post it here.

3

This was a stupid mistake on my part with an error message that completely missed the problem.

The package and module system is very sensitive to naming, which is a bit buried in the extensive documentation.

The package name must be <name>.jl, and the module must be <name> in that file, and the name entry in project.toml must be “<name>”. It makes sense but it might be nice if one of the name specifiers dominated all others. But, it would be perhaps unnecessarily confusing to allow the three specifiers to disagree. For sure it would be. That was all ok.

Further, no name in the module file or in any included source files can also have <name>. Oops. I did that. So, I renamed the module and package everywhere because I liked using my original name as the name of a struct in the code.

My bad. But, the error message certainly provided no clue but that often happens when one error masks another–or prevents parsing so a lesser error is reported. Hard to prevent this…

1 Like