Todd Vierling writes:

> : Which was the problem, and it was because #define's had changed,
> : because they didn't exist at all when the libc I had installed was
> : compiled. :)
> : 
> : So this problem only arises if you happen to install newer include
> : files, don't update your library,
> 
> If you install newer include files, you better be intending to install a new
> kernel, and do a "make && make install" in src/lib and src/gnu/lib (part of
> "make build"), or you're likely to be hosed everywhere you compile.

:)

It was all part of a tricky operation where I was playing with new
kernels but needed to be able to roll back.  I didn't want to
recompile the libraries until I was happy.  Two different kernels on
different machines share the libraries you see, but this is usually ok
because the libraries track the older kernel.

> Mind if I close this PR, then?

Even though I jumped through some weirdly shaped hoops to produce the
problem, I do think the check is worth having.  It's nice to catch the
problem if it's there, and it's a trivial addition to the code.

And it's a valid return value from sysconf(). :)  I do think missing
out the check is an ommission, albeit a usually safe one.

But I'm not going to push too hard for it, it's unlikely to personally
affect me again.