Problem Statement

Currently, SQLancer users must manually capture database states and write interestingness scripts when using C-Reduce for test case minimization. This process is error-prone and limits automation capabilities.

Implementation Plan

1. Add CLI option (MainOptions.java):

@Parameter(names = "--use-creduce", description = "EXPERIMENTAL Attempt to reduce queries using c-reduce")
private boolean useCReduce = false; // NOPMD

2. Modify Exception Handling (Main.java):

Inside executeMain(...): extend the existing catch (Throwable reduce) block.

try {
    ...
} catch (Throwable reduce) {
    reduce.printStackTrace();
    executor.getStateToReproduce().exception = reduce.getMessage();
    executor.getLogger().logFileWriter = null;
    executor.getLogger().logException(reduce, executor.getStateToReproduce());
    
    if (options.useCreduce()) {
        Reducer<G> reducer = new CReducer<>(provider);
        creducer.reduce(/* state, reduce, ... */);
    }
    return false;
}

3. CReducer.reduce(...) should:
   • Generate databasex.sql: Dump all statements from state.getStatements(), and append the failing statement (from reduce.detailMessage)
   • Generate databasex.sh (interestingness script): Extract the error substring from reduce.cause.detailMessage (e.g., from [SQLITE_CORRUPT] The database disk image is malformed (database disk image is malformed) grab database disk image is malformed), then can generate a script like: sqlite3 databasex.sql 2>&1 | grep -zq "database disk image is malformed".
   • Invoke C-reduce by ProcessBuilder, like: Process p = new ProcessBuilder("creduce", "databasex.sh", "databasex.sql").start();.

Questions

• Is this design sensible? Any obvious pitfalls?
• How to handle missing or mismatched SQLite versions? The interestingness script may fail.
• This approach only works for SQLite. How can we generalize it to support other DBMSs?

@mrigger @KabilanMA Could you review this design before implementation? Thanks!