Tim Zingelman <tez%netbsd.org@localhost> writes:
> I would ask instead that the binary package managers please learn how
> to show this information to users, because (at least in my opinion)
> where I have added them, they provide information that will avoid the
> potential for serious disappointment by the users. Such as
> security/mit-krb5/MESSAGE and security/openssl/MESSAGE.SunOS
Agreed that if MESSAGE stays at all binary package managers should show
them.
The first part of the mit-krb5 one is already in DESCR, which I think is
sufficient. I can see the point of the second hunk, but at the same
time that's normal documentation.
I didn't find security/openssl/Message.SunOS (in 2014Q4)??
I think the heart of the issue is that upstream documentation doesn't
have a universal form, and also that there's a desire to warn people who
don't read documentation of things. While I can see where that desire
comes from, I think we should optimize for people who do.
> If 'make install' leaves the software in an unsafe (default password)
> or unusable state, I think a MESSAGE is warranted.
That is more or less what I meant by "very limited circumstances where
there is an articulated reason why it really makes sense". If MESSAGE
were only used for those situations, I wouldn't have commented. In wip,
there are 258 MESSAGE files and 27 MESSAGE.* files. I read a few of the
unsuffixed ones and skimmed the MESSAGE.* and didn't see anything that
needed to be shouted at the user vs just documentation. In net/, there
are 90 MESSAGE files in 794 packages. I skimmed those, and they seem
almost entirely normal documentation. There are currently 1064 MESSAGE
files in my tree (mostly 2014Q4, some HEAD, including wip). Without a
real basis, that seems like far more than are warranted.
Also, I'd say that if make install leaves the system in a bad state
(rather than with just extra software not being run), then the package
is buggy and should be fixed.
> The latter of my examples reminds me... what about all the platform
> specific MESSAGE.platform files? Would you deprecate them too?
I would think so; those would be
${PREFIX}/share/doc/pkgname/README.platform.
Really I posted this becuase I perceived a trend that a lot of new
packages have MESSAGE, leading me to think people think it's a good
thing in general, rather than a place to put a very unusual caution.
Greg
Attachment:
pgp559ZgpevoN.pgp
Description: PGP signature