I figured out the problem. My set up was
a.ml
b.ml (* uses module C *)
c.ml (* uses module A *)
mylib.mlpack (* packs modules A and B *)

If you ignore mylib, then we have the topological order B --> C --> A.
However, actually there is no module A and B, only Mylib.A and Mylib.B.
Thus, C actually requires Mylib.A, which means Mylib must be first compiled.
However, Mylib requires B, which requires C, and there's the recursion.

What's odd is that the compilation behavior is different for byte code and
native code. I guess the byte code compiler does packing later than the
native code compiler. The correct behavior should be that compilation of
c.ml fails with an "unidentified module A" error because in c.ml, I am
referring to A directly, not Mylib.A. Is this a bug?

On Sat, Nov 8, 2008 at 3:16 AM, ygrek <ygrekheretix <at> gmail.com> wrote:

> On Fri, 7 Nov 2008 14:15:20 -0500
> "Ashish Agarwal" <agarwal1975 <at> gmail.com <agarwal1975%40gmail.com>> wrote:
>
> > I'm getting a circular dependency error when compiling a native library,
> but
> > the byte code library compiles fine. How is this possible? I'm using
> > ocamlbuild, and as far as I can tell, there is no circular dependency. I
> > would appreciate any suggestions on how to investigate this.
>
> Maybe
> http://brion.inria.fr/gallium/index.php/New_kinds_of_build_errors#Circular_build_instead_of_unbound_symbol
>
> --
> ygrek
>
>
>

[Non-text portions of this message have been removed]