At present, this PR is so EFV can oversee proposed changes to #205. It configures, but I haven't tried so much as building the compiler.

Added 19 Jan

• switched boost target from generic Boost::boost to h-o Boost::headers allowed in CMake 3.15
• removed some cmake min version boilerplate left over from external_project structure
• forced add_library(libint-libcompiler STATIC after build failed with BUILD_SHARED_LIBS=ON
• updated find_project(Python and result vars since CMake Python detection improved much around 3.15
• change DATADIR so that it's relative path in Config $<INSTALL_INTERFACE:DATADIR="\$\{_IMPORT_PREFIX\}/${CMAKE_INSTALL_DATADIR}">
• rearrange the six exported libraries into defines, features, properties, includes, link_libraries
• switches ${PN} and ${NS} abbreviation variables to ${L2} as the former generic names can get overwritten by a call (e.g., find_package() where another project assigns them)
• return to 3-char orderings in Config (e.g., gss for gaussian-standard-standard) since I think the 2-char ss were in anticipation of runtime-selectable solid harmonics. (less a priority for psi now)

WIP c. 19 Jan

• bumped CMake to 3.18 for list sorting in int_am.cmake. will revert to 3.16 at the end unless new requirements surface.
• start upgrade guide and options description in INSTALL.md
• detect Eigen if needed in Config file
• FindTargetEigen3
• int_am finalization (need to show where new cmake harness #205 not behaving. need to handle G12 properly)
• organize exports to targets files better
• handle library targets as components in Config better

Added 20 Jan

• moved configure.ac and config.h.in to a archived_libool/ dir so they can't participate unbeknownst to us
• have only the 3 definitive version lines (version, buildid, soversion) in top-level CM file. leave derived version vars and handoff to library target to cmake modules. define versions in project() (top-level and library export) so that usual _MAJOR, PROJECT_TWEAK, etc. that ppl may expect are defined
• install basis/ into versioned location again. should it really be share/libint/version/basis, not libint2?

Added 21 Jan

• check user can overwrite all the install locations and they get passed from generator correctly and all captured in the final library install to prefix.
• gave up defaulting CMAKE_INSTALL_LIBDIR to lib. probably more natural to stick with standard GNUInstallDirs. renamed (back) the custom install vars LIBINT2_INSTALL_BASISDIR and LIBINT2_INSTALL_CMAKEDIR so as to not invent new vars starting CMAKE_
• address pkgconfig paths issue mentioned at 205
• expand Libint2:: aliases so that available internally. use these for tests and drop the int-library, etc. series.

Added 24 Jan

• simplify enable_testing and build_testing sequences for new cmake and test them.
• add onebody, header-only, and pure_am components for psi's benefit. document/copy-over program-specific parameters.
• make -> cmake for doxygen. it generates files, but I haven't examined them.
• take adv of newer cmake to skip DESTINATION where can. Make sure each install() for library goes to Libint2_Runtime or Libint2_Development

Added 28 Jan

• rename FindMPFR.cmake to FindMultiprecision.cmake to handle gmp, gmpxx, and mpfr targets. Based on offical cmake module FindGSL.cmake to hopefully handle Windows, too. Seek correct targets in build_libint and library.
• remove old-style cmake variables like Libint2_LIBRARIES from config. added back Libint2_EXT_VERSION. Rearrange config so no targets created before "found" assured.
• make sure CMAKE_PREFIX_PATH passes to ExternalProject even when list by placing in Cache. Only install bundled Boost if CXX activated
• try to standardize cmake formatting (on lines I've touched for worthier reason) to 4 spaces normal and 2 spaces w/i command.
• Fortran work. Setting both CMAKE_Fortran_MODULE_DIRECTORY and property Fortran_MODULE_DIRECTORY seemed redundant. Switched to only the latter and added the path to include directories on the target. Looking on the internet, where to install module files looks like it has no guidelines (esp. since compiler-version-dependent) so added an install knob. Added fpic when necessary so tests work with shared lib.

Added 12 Feb

• renamed GHA workflow to "CI". split it into two jobs, the first of which build_repo, as before, builds and runs the compiler, builds the library, tests the library, tests a library consumer for Lin/Mac Debug/Release, all static libs. To this was added artifacting the built tarball and building on Windows (Release only). The Windows build is an expected fail -- the generated source is faulty. The second and new job build_export takes the linux tarball artifact and builds it on Lin/Mac/Win, shared libs where possible, with some additional dependency quirks testing. Changed build matrix from include format to cfg format, as I think it's easier to read. Switched Debug/Release EP/FetchContent pairs as Windows only works with 1/4 combinations.
• Reverted minimal CMake to 3.16 with a fn in int_am.cmake . Test exactly 3.16 in GHA.
• Bump CMake version to 2.7.1 from 2.7.0-beta.6 to match configure.ac
• Added an option REQUIRE_CXX_API_COMPILED to turn on/off the compiled C++11 interface apart from the header-only library. It's on by default, so no change.
• Set up ENABLE_T1G12_SUPPORT missing before.
• Added G12 ints. new macro in int_am.cmake to handle non-list, etc. and components
• Same for G12DKH but these throw error at compiler exe or lib build stage.
• renamed int-cxx-{static,shared} to int-cxx-compiled-{static,shared} for clarity
• reworked Eigen3 detection. copied a FindEigen3.cmake from latest Eigen release. This is installed with L2 and the local path is prepended so this FindEigen3 is always used. The FindEigen already handles (1) finding installations with a Eigen3Config.cmake, (2) seeking if the Eigen3_INCLUDE_DIR already in cache, (3) searching for header like a traditional FindPkg.cmake. It always returns a target, Eigen3::Eigen. This has been modified to address some of your concerns. (1) turned off build package registry to avoid the ephemeral build trees. (2) If an option LIBINT_LOCAL_Eigen3_FIND is set, the FindEigen instead looks to import a local L2-defined target (see below) to ensure ABI consistency. This is off by default. The FindEigen3 detection is shared by the library build and libint2-config.cmake
• Eigen3 handling on the src/lib/libint/CM.export side is outsourced to FindEigen3. If LIBINT_LOCAL_Eigen3_INSTALL is set, the detected eigen3 is installed as a target in a separate file. This is off by default, but I don't care if it's on by default, so long as it's turn-off-able and the _FIND option is off by default.
• fixed some locale problems for Windows in basis.h. Added some ifdefs for gnu compilers that I hit when using the wrong Windows compiler. Left them, but not really needed.
• removed a seemingly redundant basisset file. no trouble so far.
• Fixed Mac shared libraries that weren't linking
• Half fixed Win shared libraries that aren't exposing fn symbols (fixed with windows_export_all_symbols) or libint2_build_* data symbols (not fixed; see issue)
• half install target for fortran interface. (target installed, export written but not installed.) I don't know if it's wanted or generally redistributable enough.
• consolidated the main install(target and install(export according to my latest reading of best cmake practices.

WIP

• Still working on INSTALL.md and general tidying
• Do you want ENABLE_XHOST to autoselect optimization for current instruction set? It's gotten more complicated (2 -> 20 lines)
• Should I mark the name int2-cxx as provisional? I find the name confusing, and the library isn't testable.
• The components names C, CXX, CXX_ho, I keep getting wrong, which probably means they're not intuitive.
• Merge README.md and INSTALL.md?