rdmd is insufficient to build D applications: (A) mainly because of this 6 year old bug https://issues.dlang.org/show_bug.cgi?id=7016 local import does not create -deps dependency (B) no support for partial recompilation (blocked by lack of -oq, which ldc has but not dmd: https://github.com/dlang/dmd/pull/1871 has been pending for 4 years; see also https://issues.dlang.org/show_bug.cgi?id=3541: Add -oq to dmd (use fully qualified module name as object filename) ; https://issues.dlang.org/show_bug.cgi?id=12116: dmd -op -od broken) I built a build system for my use cases that addresses (A) and (B) (suboptimally though) based on forking off rdmd but it's very hard to keep in sync with rdmd because of rdmd's monolithic design: * it contains a main function (so can't be used as a library) * it's one big file, the private functions can't be reused elsewhere * in some sense it does too much (--eval=code, --loop should belong in a separate module) Recommendations: * split rdmd into several reusable modules (eg: dtools.rdmd.dependency, etc) * make the main subject to a switch
Wouldn't it be a better case for DMD as a library ?
Going to close this as there is [a pull request for it](https://github.com/dlang/tools/pull/344) that has attracted a lot of controversy, especially by Andrei. There are a few good point which highlight actual missing features / borderline bugs, which would be more suited to their own issues. If you are inclined to pursue this, please open issues at https://github.com/dlang/tools/.