While reproducing it, I found that in the bap/ocaml directory's input.ml, there is a mutable list that is being updated by functors in speclist when parse_argv or parse is called; it retains the old list between calls to my function. So I need to reset it.
(line 6 at https://github.com/argp/bap/blob/master/ocaml/input.ml)

But now I get a strange compiler error! I don't know how ocaml could be such a hard language to use...

Input.inputs:=ref [];

Error: Unbound value Input.inputs

But you can know that I have included the ocaml directory and linked it correct, since using Input.get_program already worked...

On Fri, Dec 5, 2014 at 9:38 AM, Kenneth Adam Miller <kennethadammiller@gmail.com> wrote:
Yes, I'll try and recreate it for you.

No, the backtrace in gdb is useless. All it says is:
#0  0x0000000000843033  in caml_c_call ()
#1  0x0000000000000000  in ?? ()

On Fri, Dec 5, 2014 at 4:14 AM, Anders Fugmann <anders@fugmann.net> wrote:
On 12/04/2014 10:48 PM, Kenneth Adam Miller wrote:
Well I am just no thorough and you are correct.

The sending of data over a zmq socket and the conversion of that data
from string to protobuf encoded string all occurred in one line. One I
added a print statement and then segregated them more cleanly, I can see
that it is most certainly the line that converts to protobuf.

The exact function that fails (on my end, could be deeper within this)
is to_pb from here:

https://github.com/argp/bap/blob/master/ocaml/piqi/ast_piqi.ml#L186

In any case, I did a test, and in my first function when to_pb gets
called the first time and succeeds, I added an additional call to it...
which also succeeded. But then in a subsequent unit test, the one that
has been failing, still segfaults.

If I turn off the tests prior to the segfaulting test, to_pb works in
this particular run. But if the tests run before hand, something goes
awry between the tests. Is it possible that to_pb is using some shared
state between calls?

 I would not expect so.

If you create a failing unittest that I could try?

Also, does the segfault contain a usable back trace (using gdb)? That might give some insights into which code is failing.

/Anders