On Sat, 20 Feb 1999, Mason Loring Bliss wrote:
>  * pkg/REQ:
>    Require-script that is invoked before installation and de-installation
>    to ensure things like certain accounts being available, user/sysadmin
>    agreeing with usage policy, etc.

You need to provide this script in your package. For the calling, see
pkg_add(1):

     6.   If the package contains a require script (see pkg_create(1)), it is
          executed with the following arguments:

          pkg-name      The name of the package being installed

          INSTALL       Keyword denoting to the script that it is to run an
                        installation requirements check.  (The keyword is use-
                        ful only to scripts which serve multiple functions).

          If the require script exits with a non-zero status code, the instal-
          lation is terminated.

See also: pkg_delete(1).


> 1) Am I correct in assuming that the proper thing to do is to have a package
> require the existance of a particular account, and refuse to build without
> it, rather than blindly creating the account?

No. the package system's goal is to take as much work from the (possibly
non-technically knowledged) user as possible. Thus, do what you can if it
comes to account creation. (There seems to be some consensus on not to
touch /etc/* ... you'd have to tell the user to do this manually in
pkg/MESSAGE).


> 2) Should this check be done in .../pkg/REQ? If not, what's the appropriate
> place?

seems best for me... i think postgresql does it in the PLIST, though.


> 3) And is there an accepted way to allow a user to change such an account
> name, if they wish? I would assume that setting the desired default in the
> Makefile would work... Is it safe to assume that variables defined in the
> Makefile exist for, e.g., .../pkg/REQ? I can find this out from the source,
> but I'd be just as happy with commentary.

Um, no, , see above. You'd have to edit this (automatically) into the REQ
script. Several other packages do such editing to PLIST and MESSAGES -
please look around for some examples.


> PS: I've grown accustomed to a feature in fvwm2 2.1.7... Is there any
> opposition to having our package source use this version? If not, I'll do
> up a package, once I've gotten the basics down. That version, anyway,
> compiles out of the box under NetBSD, I think. (I don't remember having
> to do anything at all funky to make it work.)

I also have 2.1.7 running here at home, and I also have a package for it
(in case you want it, let me know). There's also 2.1.10 out, but I didn't
find time to update my package.

The reason this isn't in pkgsrc is that all these versions are tagged
"beta". (Although I have to say I have _never_ seen 2.1.7 or anything else
crashing... maybe I should just commit it? What do people think?)


 - Hubert

-- 
Hubert Feyrer <hubert.feyrer@rz.uni-regensburg.de>