feat(config): route SEvt output directory through app config

[plexoos]

This PR makes gphox::Config the authoritative source for runtime configuration parameters that are applied to supported binaries execution. The main focus is refactoring event runtime controls, especially event.mode and event.output_dir, so they are read from the app config JSON and then synchronized into SEventConfig from one place.

• Added --config support to the affected binaries so runtime behavior can be selected via config file.
• Updated tests/scripts to use the config-controlled output location instead of relying on fake USER/GEOM environment setup.
• Eliminated the need to set USER and GEOM environment variables for the supported binaries/tests that now route output through Config.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 967e5d6a97

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

This PR refactors event output directory selection so that application-level JSON config (via gphox::Config) can drive SEvt output folder placement by synchronizing settings into SEventConfig.

Changes:

• Add event.output_dir support to gphox::Config and apply it via SEventConfig::SetOutFold.
• Update SEvt::DefaultBase to honor explicit SEventConfig::OutFold() overrides when determining the base output path.
• Update example/test executables and scripts to use config-driven behavior (plus bump project C++ standard to C++20).

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
tests/test_simg4ox.sh	Stops forcing fake USER/GEOM, adds set -e, quotes paths.
tests/test_GPURaytrace.sh	Removes fake USER/GEOM env-prefix for execution.
tests/test_GPUPhotonSource_8x8SiPM.sh	Removes fake USER/GEOM env-prefix for execution.
tests/test_GPUPhotonFileSource.sh	Removes fake USER/GEOM env-prefix for execution.
tests/compare_ab.py	Adds --base to locate event folders under a configurable root.
sysrap/SEvt.cc	Routes DefaultBase() through explicit OutFold overrides when present.
src/GPURaytrace.cpp	Adds --config and constructs gphox::Config to apply config.
src/GPUPhotonFileSource.cpp	Adds --config and constructs gphox::Config to apply config.
src/GPUCerenkov.cpp	Adds --config and constructs gphox::Config to apply config.
src/config.h	Introduces EventMode, plus config fields/methods for applying event policy.
src/config.cpp	Implements event mode parsing, output dir reading, and Apply() syncing to SEventConfig.
config/dev.json	Adds event.output_dir (set to /tmp).
CMakeLists.txt	Raises project C++ standard to C++20; sets CUDA standard to 17.

---

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

[ggalgoczi]

1. Config::Apply() unconditionally overrides SEventConfig::OutFold. wipes any OPTICKS_OUT_FOLD env, bypasses the $TMP/GEOM/$GEOM/$Exec controlled-dir flow, dumps SEvt output in cwd whenever a config lacks event.output_dir.
   This is a problem for all 6 binaries.

Commands that break on PR #322 HEAD:

OPTICKS_OUT_FOLD=/data/run42 simg4ox -g foo.gdml -m run.mac    # env override silently wiped → cwd
GEOM=mygeom USER=alice simg4ox -g foo.gdml -m run.mac   # controlled-dir flow bypassed → cwd
simg4ox -g foo.gdml -m run.mac -c sphere_leak           # config has no output_dir → cwd
GPUPhotonSource -g 8x8.gdml -c 8x8SiPM_crystal -m run.mac            # same → cwd
cd /home/alice && simg4ox -c wls_test ...      # writes A000/B000/ into $HOME

2. storch torch left uninitialized when JSON omits the torch section, primitives are indeterminate, generate_photons(cfg.torch) reads UB. Commented on specific lines.
   Command that breaks:

GPHOX_CONFIG_DIR=/tmp simg4ox -c notorch -g foo.gdml -m run.mac      # generate_photons reads uninit struct

3. Generic catch (std::exception&) too broad, it swallows filesystem and any non-JSON throws under "Failed reading config parameters."

Some more non-blockers:

1. gphox::EventMode duplicates SEventConfig's mode strings. src/config.h:13-23 mirrors sysrap/SEventConfig.hh:387-394. Drift risk if sysrap adds a mode. Suggest driving EventModeInfos at src/config.cpp:37-46 off SEventConfig::Minimal etc. directly.
2. config/dev.json:28 ships "output_dir": "/tmp": concurrent users on one host collide at /tmp/ALL0_no_opticks_event_name/.

Questions:

• Is the OPTICKS_OUT_FOLD env-var override regression intentional?
• Why is dev.json shipping a host-wide /tmp instead of a per-user path?

[plexoos]

Commands that break on PR #322 HEAD:

That is exactly the motivation behind this PR — rely only on the config file to control the output dir and other runtime parameter and dont' use env vars

[plexoos]

• Is the OPTICKS_OUT_FOLD env-var override regression intentional?

yes, all parameters should be defined via json configs

• Why is dev.json shipping a host-wide /tmp instead of a per-user path?

Agree and switched to the current directory by default. The user can set it to any path via a config file.

[plexoos]

3. Generic catch (std::exception&) too broad, it swallows filesystem and any non-JSON throws under "Failed reading config parameters."

changed to std::invalid_argument

[plexoos]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c1a0a630e8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[ggalgoczi]

The direction (config-only over env vars) is a good one, but only if all env variables eventually move to JSON; otherwise developers have to keep two mental models (env + JSON) to predict a run, which is worse than either extreme. Reproducibility-affecting knobs like OPTICKS_MAX_BOUNCE, the four OPTICKS_PROPAGATE_EPSILON*, OPTICKS_EVENT_MODE, OPTICKS_HIT_MASK, and OPTICKS_MAX_RECORD all change physics outcomes and need to be in the config for the single-source-of-truth claim to hold. What's your plan for migrating the rest?

[plexoos]

What's your plan for migrating the rest?

The plan is to migrate the rest gradually over time. You are welcome to help of course.

[plexoos]

@ggalgoczi Do you have any other questions?