Looking at the parse_argv documentation (man Arg), I see two possible reasons:
- Arg expects the passed string array to start with the name of the
program at index 0; does your custom_arguments array follow this
convention?
- parse_argv takes an (optional) integer reference ~current that tells
it where to start parsing the argument array; by default, it points to
a global mutable reference Arg.current (so that several calls in a row
start on the next element); if your library has used Arg.parse itself
before, Arg.current may be set too far. You should try passing
~current:(ref 0) to make sure you only use local state.

On Mon, Nov 24, 2014 at 6:42 PM, Kenneth Adam Miller
<kennethadammiller@gmail.com> wrote:
> So, I'm using a library that calls Arg.parse to build up it's inputs in
> another data structure. I don't want to rewrite any code, and I'm consuming
> the library in a different way than how the binary consumes it, which is
> just to feed it input from the command line.
>
> Not wanting to replicate code in two locations, I chose to use
> Arg.parse_argv, and supplied it with an array that would normally be on the
> command line. For some reason, supplying the exact same speclist as what was
> used in the original statement that works as a command line tool doesn't
> result in the parameters being parsed.
>
> command line tool:
> Arg.parse speclist anon usage;
> get_program () (* consumes the mutable list that speclist modified *)
>
> (* speclist modifies a mutable list used to hold arguments *)
>
> my code that consumes speclist as a library:
>
> Arg.parse_argv custom_arguments speclist anon usage
> get_program () (* exact same function as above, exact same variables to
> Arg.parse* *)
>
> For some reason, Arg.parse_argv seems to complete, but when get_program
> proceeds, it sees the mutable argument list as being empty. Why?