Eventually, I think using cabal during development may be convenient. The only drawback is that you have to specify each dependency and -- above all -- every module each time you add one.
When writing bindings-posix, bindings-glib etc., which have lots of modules, I used a shell script to take all modules under ./src to .cabal. What it did was: 1) for each subdirectory, create a .hs file that reimport all modules under that subdirectory; 2) list all .hs (and, in my case, also .hsc) under ./src and insert then into .cabal. I've found that scripts of this kind have been very usefull. They require you to follow some rules (like, say, all modules with name mapping to directories are always a reimport of submodules).
Then, to compile you executable with ghc (because Cabal is definitely not convient when you have a lib and an executable in the same package): ghc --make --package-conf pkg.conf.d main.hs
Again, should you have better/simpler ways to achieve this, I would be glad to know them.
I usually find useful, at first, to build only the executable and leave the library. When modules get stable enough, I separate both. Other scripts have been also usefull: one to check uncommited changes in all packages I'm working in, other to sync all my local packages with their repos. You could have a single 'b' script you could run like: b l # build and install your library in local database b t # build and run your test package executable b m # update module list in .cabal file for all your packages etc. Best, Maurício