noSideEffect in os module

[GULPF]

The noSideEffect pragma behaves incorrectly with some procs in os. Using symlinkExists, fileExists or dirExists in a noSideEffect proc compiles without error.

[GULPF]

After looking up the reason for this I feel a bit confused. The reason that most procs in the os module are marked with sideEffect is because they call osLastError(). The funny thing is that osLastError() accesses the global errno on POSIX, but on Windows it calls some WINAPI function. This means that osLastError() (and by extension most procs in the os module) is only marked with sideEffect on POSIX, not on Windows. To me this just seems like an implementation detail - surely calling a WINAPI function to get the last error is a sideEffect just like getting it from a global.

Is the problem that all importc procs needs to be explicitly marked with sideEffect when applicable? If so, that seems easy to fix but will likely break a bunch of code that relies on the current situation.

[dom96]

Is the problem that all importc procs needs to be explicitly marked with sideEffect when applicable? If so, that seems easy to fix but will likely break a bunch of code that relies on the current situation.

Yep. Breakage is fine as whatever code that breaks will be wrong anyway.