Re: Merging bignum to master

[Paul Eggert]

Richard Stallman wrote:
>    > That being said, the distinction between --with-FOO and --enable-FOO has 
> long
>    > been a confusing area of the GNU coding standards -- as witnessed by the 
> other
>    > packages you're thinking about.
>
> Is that because some cases have been handled erroneously?  Or is it
> because some cases don't really fit into either --with or --enable?

In some cases it's erroneous coding. But more often, I think, it's because the 
two names are both plausible and developers and users are easily confused about 
which names to use.

For example, coreutils ./configure has a --with-gmp option, which causes two 
things to happen: first, 'expr' and 'factor' link to the GMP library (which is 
appropriate for --with-FOO) and second, these two programs support the feature 
of arbitrary-length integers (which is appropriate for --enable-FOO).

Strictly speaking, I supose the GNU Coding Standards require Coreutils to have 
both a --with-gmp option (to control whether the GMP library is used) and an 
--enable-bignum option (to control whether bignums are supported). But that 
would be more confusing than what Coreutils has now, which is a single option 
that controls both things, and which is more natural since the only reason you'd 
want to link to GMP is to have bignums, and if you want bignums with Coreutils 
the only way to get them is to link to GMP.

> If the latter, I think we need to take the bull by the horns
> and work out a good way to handle those cases, rather than forcing them
> into one of two slots which they do not fit.

I hope that we needn't make the coding standards even more complicated than they 
already are in this area. When complexity causes confusion, adding more 
complexity is likely to cause more confusion.

I'd be more inclined to go in the direction Tom Tromey suggested, which is to no 
longer require a sharp distinction between --with-FOO and --enable-FOO since the 
distinction's confusion is more trouble than it's worth.