[lsb-discuss] roadmap synchronization and backward compatibility

[Wichmann, Mats D]

>I think you'd be astonished at how long the likes of Solaris, 
>HP-UX and AIX maintain deprecated interfaces. Their policy more
>closely resembles "we take things out when it becomes technically 
>impractical to continue to support them" than "we give a few years 
>warning and then remove them". 

So I guess I might be a little better off if I gave an
example of a place where I want to move more quickly.

matherr() is in the LSB, but should never have been. this
was an old construct to allow a kind of exception handling
in the pre-IEEE-floating-point days (there are new mechanisms
for that for IEEEfp).  the last reference specification that
listed this interface was the old SVID 3rd edition, which
listed it as deprecated as of 1989! LSB-conforming programs
that use the math library are to use IEEE floating point.
So including this in LSB and requiring distros to provide
support is just plain silly and I'd like to get rid of it.

So this is in the category of "mistakes", and I believe
there ought to be a little flexibility to deal with
mistakes like this more expeditiously than we're talking
about in the more general case. I don't believe there are
a lot of these, it's even possible we've done our work
well enough that this is the only one left (sure, dream on).

And yeah, I bet Solaris is still carrying matherr, and
I don't really care :-)