From: Tom Hirschowitz <tom.hirschowitz@ens-lyon.fr>

> We are trying to modify the ocaml compiler in order to have it verify
> the well foundedness of recursive module definitions, and have a
> question for the long standing ocaml source hackers.

Interesting endeavour.
Of course I hope that you can handle my pet case:
http://www.math.nagoya-u.ac.jp/~garrigue/papers/mixmod5.ml.txt

> >From early experiments, it seems that the dependencies in the various
> .depend files are insufficient. For example, try to modify the version
> string, say "3.08.3", in stdlib/sys.cmo, and make the world. Launch
> the new toplevel; nothing has changed. How do people deal with this
> issue? In the cases we have, cleaning up the executables before making
> the world again is enough. Is it in general? 

There are indeed a few things that are not redone automatically.
But of course make clean; make world should always solve the problem;
except when you have changed some internal data and you need to
bootstrap.

The kinds of problem I see frequently are:
* after modifying the standard library implementation, need to relink
  everything. Dirty workaround: touch utils/config.ml; this will force
  to relink the compilers, but not all tools.
* any change in the interface of a standard library module used
  somewhere in the compiler requires recompilation. Generally you're
  better off recompiling everything than trying to track the
  dependencies by hand.
* after changes in the compiler, ocamlbrowser compilation breaks
  sometimes. (cd otherlibs/labltk/browser; make clean) solves it.
  
The dependencies are not all that clever because:
* some cases are really cyclic (eg ./ocamlc and the standard library
  clearly depend on each other)
* having too many dependencies automatically enforced can be a pain.
  (For instance, on slow machines I sometimes disable the rule forcing
  ocamlc.opt and ocamlopt.opt to be recompiled from scratch everytime
  ocamlopt has changed.)

Jacques Garrigue