Date: Tue, 13 Jul 2004 09:31:46 +0200 From: Oliver Eikemeier <eikemeier@fillmore-labs.com> To: Herve Quiroz <herve.quiroz@esil.univ-mrs.fr> Cc: Rui Lopes <rui@ruilopes.com> Subject: Re: ports/68769: [PATCH] devel/maven: update to 1.0.r4 and MASTER_SITE_APACHE Message-ID: <B1E729D3-D49E-11D8-A6C6-00039312D914@fillmore-labs.com> In-Reply-To: <200407121150.i6CBoE05096031@freefall.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Herve Quiroz wrote:
> On Mon, Jul 12, 2004 at 01:19:49PM +0200, Volker Stolz wrote:
>> Am 09. Jul 2004 um 13:17 CEST schrieb Herve Quiroz:
>>> - Dynamic pkg-plist (hence I removed the maintainer target)
>>
>> Hm, I'm strongly against dynamic plist -- partly because they're
>> always right when something goes wrong and partly I like to look
>> at them occassionally. Is this *really* necessary?
>
> Not necessary at all. It's just that some other commiter (eik) just
> asked me for a dynamic plist for one of my ports.
I asked you to `sort the pkg-plist or generate a dynamic one'. The
dynamic generation came from the fact that you already did this in a
`x-generate-plist:' target, plus that everything in the packing list
were data files in one directory that was completely owned by the port.
IMHO the decision whether to use a (partly) dynamic packing list should
be made based on:
- is the installation process dynamic too, so that I'm sure exactly
those files are added to the plist that are installed (and no other ones
accidentally in the target directory)
- does it ease upgrades for potentially fast changing packing lists, or
are there pitfalls in the process
- is the dynamic part of the plist only for auxiliary data and does not
hide files I potentially need to depend upon or conflict with (e.g. the
main executable)
some motives for using a dynamic list here could be the sheer size of
the packing list and entries that have '-20030211.142932' in the name.
Reasons against it might be that it is not clearly defined how
PLIST_FILES and a (dynamic) packing list play together, and that it is
not clear that the plist generation (with two find commands) does
exactly the same like the install process (with cpio).
In PORTDOCS I reasoned the many ports install changing documentation
which might be generated during build time (especially the java ones)
that is hard to maintain (we had a lot of packing list errors),
installed to directories where it is unlikely that another port (or a
user) will add his own files. Besides, I'm making sure that this list is
correct by generating it from the *installed* files.
I'm currently working on PR 66032, which should make the process of
packing list generation more clearly defined, and should assure
correctness. In the meantime there is no `good practice', or as the perl
people say: TIMTOWTDI. The considerations above may help to make your
own decision.
-Oliver
P.S.: Some style nits:
The saved `>>${PLIST}' is not worth the subshell spawned by `( ... )'
and generally the stuff is less clear. Try how the code reads with two
FINDs (easier to understand IMHO). You don't need to pipe through SORT,
find -s and find -d should suffice (especially since find -s -d is what
you want, not sort -r). Also, cpio might copy stuff not matched by the
plist generation. Also the chmod/chown stuff looks tricky to me.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?B1E729D3-D49E-11D8-A6C6-00039312D914>