Simon Marlow <[EMAIL PROTECTED]> wrote,
> > The scheme is good - although, I basically have the same
> > comments as Michael:
> >
> > * It is simpler to parse the version numbers if stable
> > versions at patchlevel 0 also include the patchlevel.
>
> You shouldn't *have* to parse anything after the x.yy, because none of it
> affects the behaviour of the compiler. A compiler named x.yy(.z), where yy
> is even, is a compiler with a well-defined set of library interfaces.
> Conditional compilation should check for these values of yy only.
That's the theory - the praxis is that a good portion of the
autoconf macros actually test for bugs in used
systems/libraries and conditionally enable workarounds. So,
generally, patchlevels are important, because depending on a
patchlevel, a workaround might have to be enabled or not.
> An odd-numbered compiler is viewed as approaching the limit of the next
> even-numbered compiler, so the interfaces presented by x.yy.z (y odd, z is a
> date) are the same as x.(yy+1). This of course won't be true for certain
> values of z because the interfaces evolve over time, but the fix is to
> "upgrade to the latest value of z".
That's a problematic fix, because it means that I have to
tell a user of my code to update their compiler to get my
code running. The same user might use the same compiler
with other code, which has not yet been updated to run with
the latest compiler. So, that's again a situation, where I
might want to special case depending on "z". But I agree
that it is less of an issue for unstable versions, because
it is always possible to blame all problems on the fact that
the person is running unstable software.
> > * There is not much point in having the substring "pre" in
> > the patchlevel of unstable versions - especially, as the
> > "odd versions are unstable" scheme is becoming more and
> > more widespread (ie, the need for "pre" as a comment is
> > not really clear and as the scheme for the third digit is
> > completely different from stable versions, unstable
> > version numbers stand out anyway).
>
> Agreed. Let's remove the "pre", leaving the unstable compilers named
> x.yy.YYYYMMDD.
Good :-)
Cheers,
Manuel
