10 Problems Large Companies Have Managing C++ Dependencies and How to Solve Them Augustin Popa, Microsoft C++ TeamIntroduction Augustin Popa Senior Product Manager, Microsoft C++ Team @augustin_popa

over time? – Create a fresh environment. As environments grow, they become harder and harder to solve. Working with small, dedicated environments can be much faster. Read more about how we made conda do. Reducing the index: • Reduces unnecessary input into generating solver clauses. • Reduces solve complexity. • Prefers newer packages that apply constraints. Read more on Understanding and Improving conda install python=3.7 codecov This will download and install numerous additional packages to solve for dependencies. This will introduce packages that may not be compatible across platforms. If you

over time? – Create a fresh environment. As environments grow, they become harder and harder to solve. Working with small, dedicated environments can be much faster. Read more about how we made conda do. Reducing the index: • Reduces unnecessary input into generating solver clauses. • Reduces solve complexity. • Prefers newer packages that apply constraints. Read more on Understanding and Improving conda install python=3.7 codecov This will download and install numerous additional packages to solve for dependencies. This will introduce packages that may not be compatible across platforms. If you

errors are more likelyExample: Domain decomposition • Solve big linear system of equations Ax=b – Decompose into many small systems (“subdomains”) – Solve small systems independently and combine results – non-parallel code try { preSolveWork(); for(auto&& subdomain: subdoms) { solve(subdomain); } postSolveWork(); } catch(/* what solve throws */) { useSlowerSolver = true; } Parallelization attempt FixedMemoryPool nseq, begin(subdoms), end(subdoms), [&] (auto&& subdomain) { solve(subdomain, pool); }); postSolveWork(); } catch(/* what solve throws */) { useSlowerSolver = true; }Parallel domain decomposition

rotation • Matrix-vector: matrix-vector multiply, triangular solve, outer product update • Matrix-matrix: matrix multiply, triangular solve with multiple vectors, symmetric matrix update • Layer 1: rotation • Matrix-vector: matrix-vector multiply, triangular solve, outer product update • Matrix-matrix: matrix multiply, triangular solve with multiple vectors, symmetric matrix update • Layer 1: rotation • Matrix-vector: matrix-vector multiply, triangular solve, outer product update • Matrix-matrix: matrix multiply, triangular solve with multiple vectors, symmetric matrix update • Layer 1:

Commands conda Documentation, Release 4.6.0 -y, --yes Do not ask for confirmation. --download-only Solve an environment and ensure package caches are populated, but exit prior to unlinking and linking packages change already-installed dependencies. --update-deps Update dependencies. -S, --satisfied-skip-solve Exit early and do not run the solver if the requested specs are satisfied. Also skips aggressive updates twice for DEBUG, three times for TRACE. -y, --yes Do not ask for confirmation. --download-only Solve an environment and ensure package caches are populated, but exit prior to unlinking and linking packages

Commands conda Documentation, Release 4.6.1 -y, --yes Do not ask for confirmation. --download-only Solve an environment and ensure package caches are populated, but exit prior to unlinking and linking packages change already-installed dependencies. --update-deps Update dependencies. -S, --satisfied-skip-solve Exit early and do not run the solver if the requested specs are satisfied. Also skips aggressive updates twice for DEBUG, three times for TRACE. -y, --yes Do not ask for confirmation. --download-only Solve an environment and ensure package caches are populated, but exit prior to unlinking and linking packages

conda install python=3.7 codecov This will download and install numerous additional packages to solve for dependencies. This will introduce packages that may not be compatible across platforms. If you over time? – Create a fresh environment. As environments grow, they become harder and harder to solve. Working with small, dedicated environments can be much faster. Read more about how we made conda do. Reducing the index: • Reduces unnecessary input into generating solver clauses. • Reduces solve complexity. • Prefers newer packages that apply constraints. Read more on Understanding and Improving

conda install python=3.7 codecov This will download and install numerous additional packages to solve for dependencies. This will introduce packages that may not be compatible across platforms. If you over time? – Create a fresh environment. As environments grow, they become harder and harder to solve. Working with small, dedicated environments can be much faster. Read more about how we made conda do. Reducing the index: • Reduces unnecessary input into generating solver clauses. • Reduces solve complexity. • Prefers newer packages that apply constraints. Read more on Understanding and Improving

conda install python=3.7 codecov This will download and install numerous additional packages to solve for dependencies. This will introduce packages that may not be compatible across platforms. If you over time? – Create a fresh environment. As environments grow, they become harder and harder to solve. Working with small, dedicated environments can be much faster. Read more about how we made conda do. Reducing the index: • Reduces unnecessary input into generating solver clauses. • Reduces solve complexity. • Prefers newer packages that apply constraints. Read more on Understanding and Improving