unnecessary OS redundancy in druntime

[Sean Kelly via Digitalmars-d]

On Friday, 12 December 2014 at 15:47:09 UTC, Joakim wrote:
>
> It seems like pointless repetition and there are many more 
> examples of this, as I keep running into it when adding 
> bionic/Android support.  Martin suggested that it would be 
> useful to have a default case that asserts for an unsupported 
> OS, but many of these blocks don't have that either.
>
> Why not just declare them once for Posix and then specialize 
> for a particular OS later on when necessary

Because therein lies madness.


> as seems to have been done in these instances?

I could be convinced to allow required function prototypes to be
in version(Posix) blocks, but never data definitions.  And even
putting function prototypes in common version blocks risks link
errors.  I would *love* it if platforms that contained a header
actually implemented all of the functions required by a specific
tag (required, xopen, etc) when choosing any one from that tag,
but even this can't be relied upon.

In my C/C++ code I have compatibility modules that conditionally
implement missing functions based on the platform and libc
version, but this is so not fun.  Particularly when you're
targeting as many platform as D is trying to.  So really, you
should never see a version(Posix) block in core.sys.posix,
because it means that for some platform, compilation/linking will
fail.


> As can be seen from these links, druntime is not consistent in 
> how it's separating declarations for different OSs, sometimes 
> using the latter, more compact declarations, other times highly 
> redundant for no apparent reason.  If we can hash out how it 
> should be done, I'll submit a pull to fix it up.

It should all be redundant.  It makes for large files, but is
far easier to maintain and debug against than mixing everything
together.