[Feature request] Redirect stderr messages to exceptions

Some of the loopy scan realization code calls dim_max/dim_min on ISL sets that may be unbounded, and ISL prints a message to stderr although the exception is caught:

>>> knl = lp.make_kernel(                                                                                                                                     
        "{[i,j,k]: 0<=i<n and 0<=j<=i and 0<=k<10}",
         "out[i] = sum(j, j)")
>>> lp.realize_reduction(knl, force_outer_iname_for_scan="k")
isl/isl_tab_pip.c:499: unbounded optimum

I don't think there is an easy way to avoid hitting the above error message algorithmically.

I'd like to avoid random stderr output in islpy. I think the error message is useful, so it should probably be part of the exception message that is thrown by islpy.

The error handling code is here:

void isl_handle_error(isl_ctx *ctx, enum isl_error error, const char *msg,
        const char *file, int line)
{
        if (!ctx)
                return;

        isl_ctx_set_error(ctx, error);

        switch (ctx->opt->on_error) {
        case ISL_ON_ERROR_WARN:
                fprintf(stderr, "%s:%d: %s\n", file, line, msg);
                return;
        case ISL_ON_ERROR_CONTINUE:
                return;
        case ISL_ON_ERROR_ABORT:
                fprintf(stderr, "%s:%d: %s\n", file, line, msg);
                abort();
                return;
        }
}

Right now, my current solution is loopy@ff43c095. Ideally, we would like to capture the error output as opposed to having it go to stderr.

I can't think of a robust, portable way of capturing the output. My feeling is that this part of the code should be rewritten.

@inducer: Do you agree and do you think it makes sense to send a patch to Sven and then back-port the changes to our version of islpy?