Jeff King <peff@xxxxxxxx> writes: > On Mon, Aug 28, 2017 at 10:52:51AM -0700, Stefan Beller wrote: > >> >> I'm curious, too. I don't think the valgrind setup in our test suite is >> >> great for finding leaks right now. >> >> This discussion comes up every once in a while, >> the last time I was involved in this discussion I proposed >> to have an "optional_free(void *)", which only frees memory >> in e.g. the developer build/debug build. >> >> That way we can have a strict "no leaks in developer build" >> policy (as it is easy to detect!), but it doesn't slow down the >> widely distributed version of Git. > > Personally I am not that worried about slowing down program-exit with > some free() calls (though I would reserve judgement to see how slow it > actually is). > > It is that I do not want to deal with the complexity and maintenance > headache of dropping values which are program-lifetime caches. If we > introduce a free-all-object-structs function, now everybody needs to > remember to call it (even if they didn't realize they were allocating > object structs in the first place, as it may have happened under the > hood of a sub-function). And we have to wonder what might happen when > somebody calls that function _not_ at program exit. In addition, the code earlier may have a variable point at a compiled in literal string or an allocated string depending on the situation and it would have been perfectly fine to rely on exit() to clean it up. "We free all ourselves before exit()" would mean these codepaths now need to be updated to keep track of what needs to be and what must not be freed, or just duplicate everything to make the life of that "free everything" part easier, which somehow feels like a tail wagging a dog. > If we can declare "reachable things are not leaks" (and certainly > valgrind is aware of that distinction), then all of that goes away. And > you're just left with local variables in main() (or our cmd_* > equivalents) that appear as leaks. And then we can solve that either by > freeing them, or just calling them program-lifetime and telling the > compiler that it's so by declaring them "static". Yup.
