matthew green <mrg@eterna.com.au> wrote on Tue, 27 Nov 2001
at 12:15:54 +1100 in <23592.1006823754@splode.eterna.com.au>:

> it seems to be to generate consistent (== correct) release notes
> one needs to run the toc generation twice in any case

Correct.

> one may need to run it at all.

Parse failure.

> depends could make it so that having the generated files commited
> works but it's always possible for this to get confused as well
> (say, z depends on a and b.  i change a and then cvs update, getting
> changes to b and commited changes to z.  but now my z is newer than
> my a so 'make' will be fooled).

I do agree that if you put in a target to make them automatically,
there's no point in having them in the tree. As it stands, there is
such a target, but not executed automatically.

> what sort of extar time are we talking about here?  i see the
> potential for lossage as being too great.

It's 15 seconds to regen all 4 kinds for the i386 on my architecture.
I'd find it annoying, when debugging changes to the notes,
to have to wait 8 seconds above and beyond the time it takes
to generate one version of the notes just to see if the markup
I'd just added came out looking reasonable.

I suppose we could add a target to generate, e.g., INSTALL.more
without regenning INSTALL.more.toc, but that seems even
kludgier.

--jhawk