These are my planned cleanups, in three steps:

1. Cleanup make system (dev branch)

• Remove globbing and the non-deterministic build failures caused by this
• Remove fake nvcc and similar kludges
• Use a central Makefile for examples
• Simplify switch to OMP to only adding OMP_TARGET=1
• Automate running (and timing) of all examples as a test system
• Add separate run lines for each test in int_testing.py
• Fix 4 remaining examples that currently (also in master?) fail
  • dalitz: std::logic_error, basic_string::_S_construct null not valid: Missing parameter (now has a default)
  • zachFit: Was missing a file (nice error message now)
  • pipipi0DPFit: Missing parameter (now gives nice error message)
  • chisquare: Missing parameter (now has a default)
• let_testing.py skips tests that are missing files

Done! See PR #23 .

2. Move files to more traditional structure (new folders branch)

• New structure, including:

include/goofit
include/goofit/PDFs
include/goofit/rootstuff
src/goofit
src/PDFs
src/rootstuff

• Including goofit files (in goofit and examples) would be:

#include "goofit/MyGoofitHeader.h"

• File extension normalized (one ext for cpp, one for cuda, and one for headers)
• Added script to automate this for easy transitions
• Clean up formatting with astyle, include style file .astylerc

Done! Now in the dev branch.

3. Move to CMake build system (better builds, make it easier to integrate Hydra ?

• More compute capability choices, auto enables 3.5 extra features if possible, Auto option for GPU detection. Uses latest version of FindCUDA, supplied for CMake < 3.7
• Optional separable compilation: cleaned up PDFs, added as option (on one system, multithreaded builds went from ~5 minutes to under 2 minutes)
• Add auto Cuda/OMP option
• Add Intel build
• Add support for GooFit Packages
• Allow packages/examples to add PDFs

Done! Now in the dev branch.

4. Further questions:

• Add documentation about building to readme
• Remove unneeded warnings

Future work

• Look at including Hydra
• Look at setting up Travis CI (with low compute power tests!)
• Add TBB - too hard for now, due to use of THREADIDX, BLOCKIDX
• Improve separation of headers and source in PDFs
• Look at removing the separate docs branch, and use docs in master (now supported in GitHub)
• Make datafiles more readily available
• Use some sort of parser in the examples to improve option descriptions