Some ideas of possible solutions for this bug:

1) As Roger Upole suggests, throw an readable error when GetStdHandle fails. This would not really change much, besides of subprocess being a little less confusing. 

2) Automatically create PIPEs for those handles that fails. The PIPE could either be left open or closed. A WriteFile in the child would get ERROR_BROKEN_PIPE, if the parent has closed it. Not as good as ERROR_INVALID_HANDLE, but pretty close. (Or should I say pretty closed? :-)

3) Try to attach the handles to a NUL device, as 1238747 suggests. 

4) Hope for the best and actually pass invalid handles in startupinfo.hStdInput, startupinfo.hStdOutput, or
startupinfo.hStdError. It would be nice if this was possible: If GetStdHandle fails in the current process, it makes sense that GetStdHandle will fail in the child as well. But, as far as I understand, it's not possible or safe to pass invalid handles in the startupinfo structure. 

Currently, I'm leaning towards solution 2), with closing the parents PIPE ends.