Charles-François doesn't like packed structure (frame_t) because he fears crash on architectures not supporting non-aligned memory access or bad performances. Antoine and me want them to reduce the memory footprint of the tracemalloc module (tracemalloc.get_tracemalloc_memory()).

I think that the memory footprint has an higher price than the performances for tracemalloc: I can wait longer for a result, whereas I may not be able to use tracemalloc if it uses too much memory.

I propose to pack frame_t structure, and only disable it explicitly on architectures where tracemalloc does crash. What do you think?