On Sun, 2020-02-09 at 22:30 -0800, Zac Medico wrote:
> Hi all (especially package owners in CC),
> 
> In various packages we have inconsistent use of || preferences for
> www-client/elinks, links, lynx, w3m, and w3mmee. This means that the
> default preference depends on the combination of packages that one has
> installed and the order that one has installed them, leading to
> unpredictable results.
> 
> Here is a list of relevant packages and their dependencies:
> 
> app-text/docbook-sgml-utils: || ( www-client/lynx www-client/links
> www-client/elinks virtual/w3m )
> app-text/sgmltools-lite: || ( www-client/w3m www-client/lynx )
> app-text/xmlto: || ( virtual/w3m www-client/lynx www-client/elinks )
> dev-lang/mono: || ( www-client/links www-client/lynx )
> mail-client/mutt: || ( www-client/lynx www-client/w3m www-client/elinks )
> mail-client/neomutt: || ( www-client/lynx www-client/w3m www-client/elinks )
> net-irc/irssi: || ( www-client/lynx www-client/elinks )
> sys-fs/gt5: || ( www-client/links www-client/elinks www-client/lynx )
> x11-base/xorg-server: || ( www-client/links www-client/lynx www-client/w3m )
> 
> How about if we create some more virtuals to cover all of the relevant
> cases?

I don't think that's a valid case for a virtual since those tools do not
provide a consistent API for other packages.  It just happens that some
packages explicitly support multiple choices, and this is exactly what
|| indicates.

The virtuals would really be arbitrary here.  Developers would
repeatedly fail to use them because they wouldn't naturally expect
the virtual to exist.

> 
> We've already got virtual/w3m to cover this particular case:
> 
>     || ( www-client/w3m www-client/w3mmee )
> 
> We might want to use nested virtuals, in the same way that virtual/jdk
> nests under virtual/jre.

-- 
Best regards,
Michał Górny