Libraries layout
CPT May 2018
Scope
Establish a standard method for laying out the common libraries to be used by components in a workflow. The complexity lies in the fact that components may be compiled with different library versions and even different compilers and versions of them, so the library layout must enable dynamically loading the relevant library for each workflow component.
Background and decision status
The previous method used by the “ITM” is judged too complicate to maintain in the long term, so some adaptation should be made at the occasion of the transition to IMAS. So far, IO thinks even the solution proposed here remains too complicate and foresees to limit the freedom to use different compilers and/or library versions within a given workflow. Therefore this method is to be applied by WPCD only in the short term, while IO will take its decision later on (see IMAS-1484 tracker for a recent solution proposed by Louwrens Van Dellen and IMAS-1072).
Libraries layout
Examples are provided for a single library (XMLLIB) for the sake of clarity, to be generalised for all libraries.
In case the library depends on the Data Dictionary (DD) version, this DD version information shall be included in the <LIB_VERSION> string.
Modules
The module name explicitly identifies the library’s name and version and the compiler’s name and version.
...
Each module is conflicting with the others (of the same library). Meta-modules allowing loading at once all modules in a coherent environment for a given workflow/application (if not conflicting) will be available. In case a complex workflow/application contains components using conflicting libraries (i.e. using different versions/compilers for the same library), the relevant module must be loaded prior to the component compilation, and then unloaded.
Libraries
The library layout explicitly identifies the library’s name and version and the compiler’s name and version and puts the various files of the library as follows:
...
| Code Block |
|---|
<ROOT_LIB_PATH>/<LIB_NAME>/lib/lib<LIB_NAME>-<LIB_VERSION>-<COMPILER_NAME>-<COMPILER_VERSION>.so -> ../<LIB_VERSION>-<COMPILER_NAME>-<COMPILER_VERSION>/lib/lib<LIB_NAME>.s |
Example
| Code Block |
|---|
<ROOT_LIB_PATH>/xmllib/1.0-ifort-17/include/*.h *.mod <ROOT_LIB_PATH>/xmllib/1.0-ifort-17/lib/*.a *.so <ROOT_LIB_PATH>/xmllib/1.0-ifort-17/lib/pkgconfig/xmllib.pc <ROOT_LIB_PATH>/xmllib/1.0-ifort-17/src <ROOT_LIB_PATH>/xmllib/1.0-ifort-17/example <ROOT_LIB_PATH>/xmllib/1.0-ifort-17/man <ROOT_LIB_PATH>/xmllib/lib/libxmllib-1.0-ifort-17.so -> ../1.0-ifort-17/lib/libxmllib.so |
Suggested naming convention for modules that depend on compiler
| Code Block |
|---|
// C code GNU C -> gnu Intel -> intel // Fortran Intel -> intel GNU Fortran -> gnu G95 -> g95 |
Example of XMLLIB installation based on IMAS based release
| Code Block |
|---|
swimas
|-- etc
| `-- modulefiles
| `-- xmllib
| |-- 4.0-gnu-6.1.0
| `-- 4.0-intel-17.0.4
`-- extra
`-- xmllib
|-- 4.0-gnu-4.8.5
| |-- include
| | |-- f90_file_reader.mod
| | |-- mod_f90_kind.mod
| | |-- string_manipulation_tools.mod
| | |-- xml2eg_mdl.mod
| | |-- xml_file_reader.mod
| | |-- xmllib_parser.mod
| | |-- xmllib_pathquery.mod
| | `-- xmllib_types.mod
| `-- lib
| |-- libxmllib.a
| `-- pkgconfig
| `-- xmllib.pc
|-- 4.0-intel-17.0.4
| |-- include
| | |-- f90_file_reader.mod
| | |-- mod_f90_kind.mod
| | |-- string_manipulation_tools.mod
| | |-- xml2eg_mdl.mod
| | |-- xml_file_reader.mod
| | |-- xmllib_parser.mod
| | |-- xmllib_pathquery.mod
| | `-- xmllib_types.mod
| `-- lib
| |-- libxmllib.a
| `-- pkgconfig
| `-- xmllib.pc
`-- lib # this can be completely removed if we start using -rpath
|-- libxmllib-4.0-gnu-4.8.5.a -> ../4.0-gnu-4.8.5/lib/libxmllib.a # this can be completely removed if we start using -rpath
`-- libxmllib-4.0-intel-17.0.4.a -> ../4.0-intel-17.0.4/lib/libxmllib.a # this can be completely removed if we start using -rpath |
RPATH - compiling codes with fixed location of libraries
One of the drawbacks of the module based layout is the problem of pointing to proper library (if we don't have symbolic links with fully qualified names).
...
| Pros | Cons |
|---|---|
| different codes can use different versions of libraries | with each release of new version of the library user's code must be recompiled |
| each library is released as a separate installation (it's easy to maintain each version separately) | it's not possible to move libraries from their original location |
| we are not forced to build log LD_LIBRARY_PATH variables | |
| LD_LIBRARY_PATH has lower precedence than rpath | |
| It is still possible to use LD_PRELOAD to enforce given library to be used (e.g. debugging purposes) |
RPATH - slightly different layout of directories
In case of rpath based compilation we don't need centrally located lib directory at all. All files (paths) will be provided by pc files for each and every module.
...