Re: sys/ types policy

[Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>]

    Ick.  no.  We are not going to start creating _blah.h
    header files.  If you want to phase out BSD types
    then we need to clean up any remaining uses of them, then
    we can conditionalize them within types.h, or
    something like that.


					-Matt
					Matthew Dillon 
					<dillon@xxxxxxxxxxxxx>

:
:Hi all,
:I had some discussion with Hiten yesterday about the use of types in
:system headers and want to discuss our consensus.
:
:At the moment most header files depend on the inclusion of sys/types.h
:and the file included there. The types defined there fall in one of following
:categories:
:- POSIX complaient types used both by system-dependent and portable code
:- BSD specific types (e.g. u_intN_t), not needed by portable code
:- machine dependent types of interest only for kernel apps (e.g. vm_size_t)
:
:My proposal is to split sys/types.h in two part:
:- sys/_types.h which defined only POSIX types and includes only those
:  files necessary for general userland applications or files which have
:  proper _KERNEL/_KERNEL_STRUCTURES guard
:- sys/types.h which includes sys/_types.h and defines the rest
:
:Afterwards all system headers must either depend on _KERNEL/_KERNEL_STRUCTURES
:to be defined or only use types defined in other sytem headers or _types.h.
:Esp. the use of historic BSD types like u_short, u_intN_t and similiar has
:to be exchanged by either protected names like __u_short or POSIX types.
:
:The rational is that most header file will break anyway if your programs
:don't follow POSIX, so obfuscating the system headers for POSIX types is
:unnecessary. The situation is somewhat different for types not defined there.
:
:Joerg
: