用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/GEOS-ESM/MAPL --skill mapl-build命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | mapl-build |
| description | Build MAPL with NAG, gfortran, or Intel compilers on macOS and Linux |
| compatibility | opencode |
Provide step-by-step instructions for building MAPL with different compilers on different platforms, including:
Use this skill when:
IMPORTANT: First question to answer - Which branch should we start with?
develop - For legacy MAPL (MAPL 2.x) workrelease/MAPL-v3 - For MAPL v3 developmentYour actual work will be on a feature/bugfix/hotfix branch, but these are the base branches.
Golden Rule: Use separate build directories for each compiler. Never mix compilers!
./nag # For NAG Fortran builds
./gfortran # For gfortran builds
./intel # For Intel compiler builds (on bucy)
Alternative pattern (if you prefer):
./build-nag
./build-gfortran
./build-intel
Configure your preference in ~/.opencode/mapl-config.yaml:
paths:
build_dir_prefix: "" # For ./nag pattern
# OR
build_dir_prefix: "build-" # For ./build-nag pattern
NAG is the fastest compiler for MAPL (~81 seconds from scratch) and provides strictest checking.
Check ~/.opencode/mapl-config.yaml for your module names, or use common defaults:
module load nag mpi baselibs
Common variations you might need:
module load compiler/nag-fortran openmpi MAPL-baselibs
# Or
module load nag-fortran/7.1 mpi/openmpi baselibs
module list
which nagfor
cmake -B nag -DCMAKE_BUILD_TYPE=Debug
cmake --build nag
For parallel builds (faster):
cmake --build nag -j 8 # Use 8 cores
120 seconds (2 minutes) - provides 25% padding over measured time
gfortran is slower (~6m 43s from scratch) but more portable and widely available.
module load gfortran mpi baselibs
Common variations:
module load gcc/13 openmpi MAPL-baselibs
# Or
module load gnu-fortran mpi/openmpi baselibs
module list
which gfortran
gfortran --version # Should be 10.x or newer
cmake -B gfortran -DCMAKE_BUILD_TYPE=Debug
cmake --build gfortran
Parallel build (recommended):
cmake --build gfortran -j 8
540 seconds (9 minutes) - provides 25% padding
For Intel compiler testing, use the remote bucy server. This requires PIV card authentication.
See the remote-build skill for complete bucy workflow, including:
Quick reference for bucy build:
ssh bucy "cd ~/swdev/VS/MAPL && source /etc/profile.d/modules.sh && module load ifx-stack && cmake -B intel -DCMAKE_BUILD_TYPE=Debug"
ssh bucy "cd ~/swdev/VS/MAPL && source /etc/profile.d/modules.sh && module load ifx-stack && cmake --build intel -j 6"
IMPORTANT: Commands must load modules in the same shell session (see remote-build skill for details).
cmake -B <compiler> -DCMAKE_BUILD_TYPE=Debug
cmake -B <compiler> -DCMAKE_BUILD_TYPE=Release
cmake -B <compiler> -DCMAKE_BUILD_TYPE=RelWithDebInfo
Use -j flag to speed up builds:
# Auto-detect available cores
cmake --build nag -j
# Specify number of cores
cmake --build nag -j 8
# Conservative (leave cores for other work)
cmake --build nag -j 4
NOTE: NAG builds are so fast (~81s) that parallel builds give minimal improvement. Gfortran benefits much more from parallelization.
Problem: CMake cannot find the Fortran compiler Solution: Forgot to load compiler module
# Check loaded modules
module list
# Load compiler module
module load nag mpi baselibs # Or gfortran
# Verify compiler accessible
which nagfor # or which gfortran
Problem: Reused build directory after switching compilers Solution: Use separate build directories, never reuse
# WRONG - reusing directory
cmake -B build -DCMAKE_BUILD_TYPE=Debug # with NAG
module load gfortran
cmake -B build -DCMAKE_BUILD_TYPE=Debug # ERROR!
# CORRECT - separate directories
cmake -B nag -DCMAKE_BUILD_TYPE=Debug
cmake -B gfortran -DCMAKE_BUILD_TYPE=Debug
If you made this mistake:
rm -rf ./build # Delete confused directory
cmake -B gfortran -DCMAKE_BUILD_TYPE=Debug # Start fresh
Problem: Cannot find MPI, NetCDF, or other dependencies Solution: Load baselibs module
module load baselibs # Provides ESMF, NetCDF, HDF5, etc.
Problem: Build uses cached old values Solution: Clean and reconfigure
rm -rf ./nag
cmake -B nag -DCMAKE_BUILD_TYPE=Debug
cmake --build nag
Problem: Loaded wrong module combination Solution: Purge and reload
module purge # Clear all modules
module load nag mpi baselibs
module list # Verify correct modules loaded
After making code changes, just rebuild (no need to re-run cmake):
cmake --build nag
CMake automatically detects changed files and rebuilds only what's needed.
Typical incremental build times: < 3 minutes
If working specifically on generic3g code, see the mapl-testing skill for a faster incremental build workflow that rebuilds only generic3g tests.
See the compiler-switching skill for detailed instructions on safely switching between compilers.
Key points:
module listSometimes you need to start completely fresh:
# Remove all build directories
rm -rf ./nag ./gfortran ./intel
# Or remove just one
rm -rf ./nag
# Reconfigure
cmake -B nag -DCMAKE_BUILD_TYPE=Debug
cmake --build nag
After successful build, run tests with:
cd nag # or gfortran, or intel
ctest --output-on-failure
See the mapl-testing skill for detailed testing workflows, including the fast generic3g-only testing method.
Standard build process applies. Most features are stable.
MAPL v3 pre-release branch. Build process is the same, but may have:
Check branch-specific README or INSTALL.md for any special requirements.
For continuous integration:
module load nag mpi baselibs
cmake -B nag -DCMAKE_BUILD_TYPE=Debug
cmake --build nag -j 8
cd nag && ctest --output-on-failure
module load gfortran mpi baselibs
cmake -B gfortran -DCMAKE_BUILD_TYPE=Debug
cmake --build gfortran -j 8
cd gfortran && ctest --output-on-failure
See remote-build skill for complete workflow.
which nagfor or which gfortran)