This repository contains the Curry package manager (CPM).
To build the Curry package manager, you need to run make inside this directory. The Makefile assumes that the curry executable and git are on your path. If the build was successful, a cypm binary will be placed in the directory ~/.cpm/bin (which is also the directory where CPM installs binaries of tools distributed in packages). Therefore, you should add this directory to your path. Afterwards, run cypm update to clone a copy of the central package index repository. More information can be found in the manual, see the docs directory.
Use make manual to generate a PDF version of the manual. A working LaTeX installation is required. make doc generates the CurryDoc documentation for the CPM source code in the cdoc directory.
Please run the tests using make test before publishing your changes. You should also run the performance tests when you make changes to the API or behavior comparison modules or the resolution algorithm. To run the performance tests, build the performance test program using make buildperf. You can then use bin/perftest to execute the different performance test.
To test the API comparison algorithm, use bin/perftest api -n NUMBER, where NUMBER is the number of added, changed and removed functions and types each that you want to compare. Note that when you specify 1000, the API comparison will compare 6000 elements: 1000 added functions, 1000 removed functions, 1000 changed functions, 1000 added types, 1000 removed types, and 1000 changed types.
The behavior comparison algorithm can be tested using bin/perftest behavior -t T -f F, where F is the number of functions to compare and T is the depth to which the type of each function's argument is nested. For example, if T is set to 2, each generated function will take a type Nested1, which is defined as follows:
data Nested1 = Nested1 Nested2
data Nested2 = Nested2 StringTo test the resolution algorithm, you need a set of test data, which you can find in the cpm-perf-test-data repository. Make sure that packages.term from thate repository is available in the current directory and then run bin/perftest resolution --packages=P, where P is a comma-separated list of package identifiers. A complete list of package identifiers available in the test data set can be found in the packages.txt file alongside packages.term. A good set of packages to start with is the following:
express-4.14.0 has 1,759 dependencies available in 23,295 different versions. Resolution succeeds relatively quickly, since a solution can be found early in the candidate tree.express-3.9.0 has 1,794 dependencies available in 23,286 different versions. Resolution fails in about a quarter second on KiCS2 since a package is missing in the sample data set.chalk-1.1.3 only has 8 dependencies available in 65 different versions. Resolution succeeds very quickly.request-2.74.0 has 1,789 dependencies available in 23,229 different versions. Resolution still succeeds quickly on KiCS2, but takes over a second on PAKCS.mocha-1.21.5 has 1,789 dependencies available in 23,229 different versions. Resolution fails with a dependency conflict in about 4.5 seconds on KiCS2, but fails to finish in a reasonable timeframe on PAKCS.karma-1.2.0 has 1,850 dependencies available in 24,264 different versions. Currently, the resolution algorithm is too slow and does not arrive at a solution in a reasonable timeframe.