Thanks for the details, Arthur!It sounds like your process is working well. If you are going to test GCC in this fashion, I may focus our CPU cycles in a LLVM for the time being. I'm just not finding a lot wrong with GCC these days, for the first time in several years (I wonder if others besides you are running Csmith?).
Your reduction process sounds good-- perhaps we can use these tricks in an automated reducer.
Do you know about this delta debugger? It tends to work very well for crash bugs (less so for wrong-code bugs):
http://delta.tigris.org/The problems that are left when this one finishes, like removing unused function arguments, is what I'm now working on automating. Ugh.
I find that most bugs aren't affected by search-and-replacing all type names with "int". (Actually, I'm pretty sure I'd prefer it if Csmith didn't use typedefs for the integer types in the first place.)
We did this based on the (naive and wrong) idea that it would make Csmith output produce the same checksum across different embedded platforms.
Can you try compiling Csmith output with the CSMITH_MINIMAL CPP symbol defined? This includes an alternate set of files where (among other things) types like int32_t is a macro instead of a typedef. The result is easier to reduce.
John