Currently, when afl-tmin is executed on non-crashing input (instrumentation mode), it can be very slow (especially for targets with a long initialization phase), due to its re-execution of the process for each run. It should be possible to run it like afl-fuzz does - re-run just the target function on each iteration, and keep the process open.
Currently, when afl-tmin is executed on non-crashing input (instrumentation mode), it can be very slow (especially for targets with a long initialization phase), due to its re-execution of the process for each run. It should be possible to run it like afl-fuzz does - re-run just the target function on each iteration, and keep the process open.