Add --trim-mode (compile-time) preference #31 (fixed)

[vishal2005025]

Title: Add compile-time _trim_enabled preference during --trim precompilation

Summary

When building with --trim, some packages need to know at precompile time that trimming is enabled so they can register / specialize code accordingly (see SciML/LinearSolve.jl#778). This PR ensures a compile-time preference _trim_enabled is visible to packages during the Pkg.instantiate(); Pkg.precompile() step invoked by JuliaC when trimming is requested.

What I changed

• Implemented temporary depot-based LocalPreferences injection for precompilation in:
  • src/compiling.jl

Behavior

• If an ImageRecipe has trimming enabled (i.e. recipe.trim_mode !== nothing && recipe.trim_mode != "no"), the precompile subprocess will:
  1. Create a temporary depot directory.
  2. Write config/LocalPreferences.toml containing:

     [Preferences]
     _trim_enabled = true
     

  3. Invoke Pkg.instantiate(); Pkg.precompile() with JULIA_DEPOT_PATH pointed to the temporary depot so Preferences.load_preference(Preferences, "_trim_enabled") will return true for packages during precompilation.
  4. Clean up the temporary depot after the precompile subprocess finishes.

Why this approach

• The repo discussion suggested using a LocalPreferences.toml hack so packages can call:
  Preferences.load_preference(Preferences, "_trim_enabled")
  and see a compile-time flag during precompilation.
• Using a temporary depot avoids touching the user's global depot or permanently modifying configuration.
• This design is minimal, low-risk, and limited only to the precompile subprocess invoked by JuliaC.

Files edited

• src/compiling.jl — add logic to create temporary depot, write LocalPreferences.toml, set JULIA_DEPOT_PATH for the precompile subprocess, and cleanup afterward.

Testing & verification

• Ran the package test suite: julia --project -e "using Pkg; Pkg.test()" — all tests passed locally on Windows.
• The existing test suite includes trimming-related tests (CLI and programmatic flows) and completed successfully.
• Suggested additional test (optional): add a small integration test that asserts a package reading the preference during precompile returns true, e.g. a tiny test project that calls Preferences.load_preference(Preferences, "_trim_enabled") in a top-level file and checks the behavior during the Pkg.precompile() invoked by the compile pipeline.

Caveats & notes

• This relies on packages using Preferences.jl and reading LocalPreferences.toml via Preferences.load_preference. Packages that obtain preferences from other sources will not be affected.
• The temporary depot is only used during Pkg.instantiate(); Pkg.precompile() subprocess and is removed afterwards; user's depot is not modified.
• The change affects only the precompile subprocess for trimming-enabled runs. Runtime behavior is unchanged.
• If you want to expose the preference differently (e.g., via an exported Base function or environment variable), we can iterate — this approach follows the repo discussion and avoids invalidating the global pkgimage cache.

How to try locally

Run the existing test-suite (this was used to validate the change):

julia --project -e "using Pkg; Pkg.test()"

[vishal2005025]

please review!

[vishal2005025]

@topolarity please take a look!

[gbaraldi]

Does this cause Julia to re-precompile the whole environment for every run? Or just the packages that have read that preference

[topolarity]

I assume this is working around JuliaPackaging/Preferences.jl#86

We should probably fix that bug instead

[gbaraldi]

Sorry to have hijacked your PR slightly

[Copilot]

Pull request overview

This PR adds compile-time preference support for trim mode, enabling packages to detect during precompilation whether trimming is enabled. The implementation creates a temporary Julia environment with a LocalPreferences.toml file that sets trim_enabled = true, making this preference available to packages during the precompilation subprocess.

Key Changes:

• Modified precompilation flow to inject trim preferences via a temporary environment in JULIA_LOAD_PATH
• Added test project TrimPrefsProject that reads the preference and validates propagation
• Added integration test to verify the preference affects compiled executable behavior

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
src/compiling.jl	Adds temporary environment creation with LocalPreferences.toml for trim mode preference injection during precompilation
test/programatic.jl	Adds integration test verifying trim preference propagation to compiled executables
test/runtests.jl	Adds constant for new test project path
test/TrimPrefsProject/src/TrimPrefsProject.jl	New test module that reads trim_enabled preference at compile time
test/TrimPrefsProject/Project.toml	Project definition for trim preference test
.gitignore	Updates pattern to ignore additional Manifest file variants

Critical Issues Found:

1. Undefined variables in cleanup code - The cleanup block (lines 148-160) references local_prefs_file and local_prefs_backup which are never defined, causing this code to fail at runtime
2. Windows compatibility issue - JULIA_LOAD_PATH uses Unix path separator : which won't work on Windows (needs ;)
3. Misleading test assertion - Test checks for cleanup of a file that's never created in that location
4. Naming inconsistency - PR description mentions _trim_enabled but implementation uses trim_enabled

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[vishal2005025]

Sorry to have hijacked your PR slightly

@gbaraldi No problem at all - thanks for taking a look! I’m happy to update anything if needed.

[topolarity]

Suggested change

	env_overrides["JULIA_LOAD_PATH"] = join([tmp_prefs_env, "@", "@stdlib"], load_path_sep)
	env_overrides["JULIA_LOAD_PATH"] = tmp_prefs_env * load_path_sep

since JULIA_LOAD_PATH=":foo" expands to: ["@", "@v#.#", "@stdlib", "foo"]

[topolarity]

mktempdepot(f, trim_mode) = trim_mode ? mktempdir(f) : f(nothing)
mktempdepot(trim_mode) do tmp_depot
   ...
end

might be a more standard pattern for this

[vishal2005025]

@topolarity @gbaraldi i have address all the reviewed changes.
please take a look and let me know if changes are required.