@bp1222 I think the patch should return \SIG_DFL if the previous handler is null. This would eliminate some userland logic when restoring the previous handler. Sorry, I thought \SIG_DFL was a non-zero integer... turns out it is 0. Returning true is better for BC. Either that or \SIG_DFL should be redefined to something else (e.g. 2).

@krakjoe @dshafik Could we consider merging this for 7.1? It can't go into an prior version as it does have a BC break against === true return value checks.

It is a weird sitch. Since SIG_DFL, in signal.h is a functionptr, and pcntl is just casting it to a zend_long. I wouldn't want to have PHP's SIG_DFL be some arbitrary number to handle this case.

However, we could just document that the return value will be false on error, true on first set, and callable when overriding. Does that involve some aspect of user management? Maybe. Not sure if ugly or not, but if $handler === true, we could assume SIG_DFL. But that's also a bit clunky. So I'd opt to leave it kindof as-is where if you set a callable, it'd return the callable, else true.

I will re-write this to patch master .. as I assumed, it probably wouldn't want to be applied to the 5.6.x branches from the BC problem