Hi. I'm working on a shared lib which I would like to optimize with profile guided optimization. I would prefer to have multiple small programs (one for typical use case), each examining a single module of the lib rather then one big program covering complete lib's API, and to generate the profile by running all those use cases. When built with -fprofile-generate, and the resulted binary is executed multiple times, are the .gcda files (re)written from scratch every time, or are they updated so they contain a combined profile data from multiple runs? Any other gotchas this approach might bring? gcc manual does not provide any details about this... Thanks in advance, Mity
