Happens to us at my day job too. The codebase is primarily C++. My most recent horror story is that I was stepping through some code in an object that was statically initialized and a variable that was initialized as `static const double foo = nan;` had a value of 0 in it. This was very surprising to me.
I look at how we defined nan and it turns out that nan is a singleton that was initialized in a DLL somewhere. My variable was being initialized before the singleton nan was initialized. I asked around, and someone with access to the previous version control system (we migrated to git in 2016) discovered that this was part of the original commit to that VCS back sometime in 2003-2006 or something. We think that was probably from before our C++ compiler was updated to support C++98 and `numeric_limits` was added.
So of course I moved this over so that accessing our special nan singleton is just a static constexpr call to `std::numeric_limits<double>::quiet_NaN()`. But our special nan singleton is used all over the place in our codebase. So of course I have to check to see nobody's doing something weird with it.
Of course they are.
There are about a hundred lines of code that boil down to `if (foo == special_nan_singleton) { /* ...handle nan / }` which of course...isn't how nan works. This is a no-op and the compiler just straight up compiles it out of binary. This happens a lot*. Including fundamental infrastructure, like the special JSON serializer somebody reinvented.