Problem with app using Turing

Hi. I try to compile an app to predict bayesian model . I saved posterior chain, and model specification. My precomp_file is something like that.

**precomp_file.jl**

```julia
using Turing
using Distributions
using LinearAlgebra
using DataFrames
using CSV
using StatsBase
using Tracker
using MKL
using Bijectors

import Logging
Logging.disable_logging(Logging.Warn)

# El objetivo es precompilar las funciones para
# que la app vaya rápido

## Leer cadena y especificación modelo
chain_path = "/home/jose/Julia_projects/bayesian_app/resources/cadena_refactor.jls"
# al hacer include dentro de la app mira por defecto en src
esp_modelo_path = "/home/jose/Julia_projects/bayesian_app/src/especificacion_modelo_refactor.jl"
df_path = "/home/jose/Julia_projects/bayesian_app/resources/Advertising.csv"
chain = read(chain_path, Chains)

df = CSV.read(df_path, DataFrame, ntasks=10)

include(esp_modelo_path)

## predecir
predicciones = predict(model_turing(df ), chain)

pred_df = DataFrame(predicciones)
pred_df = pred_df[!, Not([:iteration, :chain])]
medias = mean.(eachcol(pred_df))
mediana = quantile.(eachcol(pred_df), 0.5)
desv_ti = std.(eachcol(pred_df))
q01 = quantile.(eachcol(pred_df), 0.1)
q90 = quantile.(eachcol(pred_df), 0.9)

pred_interval = DataFrame(hcat(medias, mediana, desv_ti, q01, q90), :auto)

rename!(pred_interval,[:means,:median , :std, :q01, :q90])

CSV.write("data/predicciones.csv", pred_interval)

```

where **especificacion_modelo_refactor.jl i**s a julia script only with model turing specification

```julia
@model function model_turing(df)
TV = df.TV
radio = df.radio
newspaper = df.newspaper
# Prior coeficientes
a ~ Normal(0, 0.5)
tv_coef ~ Normal(0, 0.5)
radio_coef ~ Normal(0, 0.5)
newspaper_coef ~ Normal(0, 0.5)
σ₁ ~ Gamma(1, 1)

mu = a .+ tv_coef .* TV .+ radio_coef .* radio .+ newspaper_coef .* newspaper
sales ~ MvNormal(mu, σ₁^2 * I)
end

```
My app file is something like that.

```julia
module bayesian_app

using Turing
using Distributions
using LinearAlgebra
using DataFrames
using CSV
using StatsBase
using Tracker
using MKL

function julia_main()::Cint
try
real_main()
catch
Base.invokelatest(Base.display_error, Base.catch_stack())
return 1
end
return 0
end

function real_main()
if length(ARGS) == 0
error("pass arguments")
end

# Read chain
chain = read(ARGS[1], Chains)
# Read model_espec in julia script
include(ARGS[2])

# read data
df = CSV.read(ARGS[3], DataFrame, ntasks=10)
# Predict
predicciones = predict(model_turing(df), chain )

pred_df = DataFrame(predicciones)
pred_df = pred_df[!, Not([:iteration, :chain])]
medias = mean.(eachcol(pred_df))
mediana = quantile.(eachcol(pred_df), 0.5)
desv_ti = std.(eachcol(pred_df))
q01 = quantile.(eachcol(pred_df), 0.1)
q90 = quantile.(eachcol(pred_df), 0.9)

pred_interval = DataFrame(hcat(medias, mediana, desv_ti, q01, q90), :auto)

rename!(pred_interval,[:means,:median , :std, :q01, :q90])
CSV.write("pred_interval_80.csv", pred_interval)
CSV.write("pred_posterior.csv", pred_df)

end

end # module

```

And **I try to compile using**
```julia
create_app("../bayesian_app", "../bayesian_bin_predict", precompile_execution_file="../bayesian_app/src/precomp_file.jl", force=true, filter_stdlibs = false)
```

And get the compile, once I install some dpendencies like Tracker.jl an others.

I try to test my app passing a mcmc chain, a julia file where I wrote the model especification , but I get weird error message

```bash
../bayesian_bin_predict/bin/bayesian_app cadena_refactor.jls especificacion_modelo_refactor.jl Advertising.csv
┌ Warning: Package Bijectors does not have ChainRulesCore in its dependencies:
│ - If you have Bijectors checked out for development and have
│ added ChainRulesCore as a dependency but haven't updated your primary
│ environment's manifest file, try `Pkg.resolve()`.
│ - Otherwise you may need to report an issue with Bijectors
└ Loading ChainRulesCore into Bijectors from project dependency, future warnings for Bijectors are suppressed.

┌ Warning: Error requiring `Tracker` from `Bijectors`
```
But when I see Manifest and Project I view Tracker.jl and Bijectors.

Any idea?

Thanks

Start by reproducing the reported create_app invocation with precomp_file.jl, then compare its dependency setup with the bayesian_app module and the generated application's Project and Manifest files. Trace the Bijectors warning and the failed Tracker requirement; done means the compiled app runs with the three supplied arguments without those dependency warnings or errors.

Written by the indexing model from the issue text.