An Alternate Larsoft Build Process Patrick Gartung FNAL - PowerPoint PPT Presentation

Larsoft librarians biweekly meeting Nov 18, 2014 An Alternate Larsoft Build Process Patrick Gartung FNAL gartung@fnal.gov

Some Context ● Ben Morgan (University of Warwick) created geant4 cmake files ● Created cmake files for Art (fnal-art) and FNALcore (cpp0x, cetlib, fhicl-cpp and MessageFacility) – LBNE/FNALCore git repo – LBNE/fnal-art git repo ● Exports FNALCore and Art library interface using only cmake native macros ● Can be used in any cmake file by adding FNALCore and Art install directories by adding to cmake prefix path and calling find_package(FNALCore 1.0.0) and find_package(art 1.11.03) – FNALCore talk – Art build talk – Art build update talk ● Motivated by desire to build on OSX with system install of compiler and externals. ● Makes it possible to build using lbne-build which is based on worch.

Objectives ● Rewrite the cmake files for the larsoft package to only use cmake macros available directly from the cmake package. ● Use the changes that Ben Morgan made to the art package as a guide. ● Record the options used to invoke cmake in a worch config

Getting larsoft into github ● Checkout all of the larsoft repos from FNAL redmine ( aka upstream ) ● Create an empty repo for each in LBNE account on github – https://github.com/LBNE ● Push upstream develop and master branches to github ● Github rejected because of 120MB file that was added in initial svn commit to larsoft and then removed. ● Followed directions that github gave for removing large files – https://help.github.com/articles/working-with-large-files/

Setting up build environment ● Started a build from scratch using lbne-build package ( lbne specific config for worch ) – https://github.com/LBNE/lbne-build ● Spent some time debugging why “make test” was failing for FNALCore and fnalart packages ● Updated config for the latest version of FNALCore and fnalart ● Updated sqlite3 version and attempted to use the FindSQLite3.cmake macros which should work without passing extra -D options to cmake. ● Figured out that the cmake macros work for a system install of sqlite3 but not in worch environment. Fell back to passing -D options to cmake. ● Updated clhep and root versions needed by fnalart.

Build first larsoft package – larcore ● Started with larcore because of dependency chain Larcore<-Lardata<-Larevt<-Larsim<-Larreco<-Larana ● Quickly added rules to build regular libraries and add header files in install area ● Needed to create cmake modules directory and copy over Find*.cmake from fnalart. ● Found a bug in the way root was configured while generating dictionary library - DGCCXML_EXECUTABLE={gccxml_install_dir}/bin/gccxml

Comparing it to upstream ● Library sizes were much smaller than upstream ● Location of lib and bin directories different – OK because lib location exported in cmake files ● All header files in same location ● Needed to add .gdml and .pl and .C files in gdml dir, .fcl files in job dir, and test scripts in bin

Making it equivalent to upstream ● After adding cxx flag -g sizes were comparable ● Dug through cetbuildtools macros and ups config for larcore to reverse engineer install_gdml(),install_fhicl() and install_scripts() macros. ● Cmake globbing patterns, ups set variables, etc. ● Didn't implement install_source() to install the library source files following the install pattern of fnalart.

Side effect of resizing larcore repo ● While comparing gdml dir found file was missing larcore/Geometry/gdml/icarus.gdml ● In the repo I found icarus.gdml.REMOVED.git-id ● I had passed the wrong flag to the “BFG” utility recommended by github – BFG removed top N files by size not files over github 100MB limit ● Both “BFG” and “git filter-branch” removed the one file over 100MB but all commits after that are rewritten changing their commit id. ● Asked github for an exception to the 100MB file size limit for the larsoft repos. They graciously raised the size limit to 200MB. ● All larsoft repos now cloned to github with the upstream commit ids. ● Cloned the lbnecode repo on github just to be sure there were no size limit problems.

Building lardata ● Added lines to larcore cmake files to export library interfaces following fnalart example ● Added larcoreConfig.cmake.in to be used by other package to import larcore interfaces ● Added GENIE and LHAPDF to worch config and built them. ● Figured out where the get the sources for nutools package. ● Modified nutools cmake files to not use cetbuildtools macros and stored changes as a patch to upstream in worch. Only built libSimulationBase.so and copied associated header files. All other cmake files commented out for now. ● Turns out nutools was not needed for lardata because source code using it had been moved to uboonecode repo.

Building larevt and larsim ● Nutools header needed to build larevt Filters/FinalStateParticleFilter_module.cc:28:39: fatal error: SimulationBase/MCNeutrino.h: – No such file or directory ● More nutools libraries need to be built for larsim – EventGeneratorBaseCRY – EventGeneratorBaseGENIE ● Need to modify nutools cmake files so they work without cetbuildtools and ups. Then they can be built with worch.