Date: Sun, 08 May 2005 20:54:58 -0600 From: "James E. Flemer" <jflemer@uvm.edu> To: freebsd-ports@freebsd.org Cc: Michael Nottebrock <lofi@freebsd.org> Subject: Re: Why are implicit package dependencies recorded? Message-ID: <427ED102.6040406@uvm.edu> In-Reply-To: <200505082049.31145.lofi@freebsd.org> References: <427E4687.40901@uvm.edu> <200505082044.52084.lofi@freebsd.org> <200505082049.31145.lofi@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
This is a multi-part message in MIME format.
--------------050605030609090400050402
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Michael Nottebrock wrote:
> On Sunday, 8. May 2005 20:44, Michael Nottebrock wrote:
>
>
>>>Both of these cases can be avoided by building P with the same flags
>>>that D was built with. In the simple case, that does not seem like a
>>>bad solution, but in practice there can be a huge number of deps and
>>>sub-deps (mplayer has around 45).
>
>
> Ah, I've been reading too fast again. Yes, all packages in a dependency-chain
> will have to be built to match. It would be possible to make the pkg_* tools
> smarter in order to avoid that, but at the moment quite everything (from
> pkg_* to the package cluster scripts) relies on prerecorded (or
> pre-generated, i.e. INDEX) sub-deps.
>
Actually, pkg_add does indeed do the right thing when only first-order
dependencies are recorded in the package. I was pretty sure that was
the case, so I cooked up a trivial case and the (implicit) sub-dep was
installed correctly.
The interesting thing is that of the many *-depends and *-depends lists
targets in bsd.port.mk, only one needs changing to change the packaging
behavior. The PACKAGE-DEPENDS-LIST macro is used for the
package-depends-list and package-depends targets, and the latter is what
determines the pkgdep lines in the package. Neither of these two
targets are used for any other purpose in the Mk/bsd.*.mk files.
A simple three line patch[1] is all that is needed to make the change to
record only the first-order dependencies. The all-depends-list target
and macro are left recursive, preserving all other recursive aspects of
the port infrastructure.
I cannot say what external consumers there are of the
PACKAGE-DEPENDS-LIST macro, or of the package-depends and
package-depends-list targets. However, there are no consumers other
than do-package within the port makefile infrastructure.
[1] Hopefully attached and at:
http://www.cs.rpi.edu/~flemej/freebsd/pkg-first-order-deps/
-James
--------------050605030609090400050402
Content-Type: text/plain;
name="bsd.port.mk.diff"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
filename="bsd.port.mk.diff"
--- bsd.port.mk.orig Sun May 8 20:18:18 2005
+++ bsd.port.mk Sun May 8 20:42:20 2005
@@ -608,7 +608,7 @@
# for this port.
# package-depends-list
# - Show all directories which are
-# package-dependencies for this port.
+# first-order dependencies for this port.
# run-depends-list
# - Show all directories which are run-dependencies
# for this port.
@@ -4522,7 +4522,7 @@
fi; \
done | ${SORT} -u
-# Package (recursive runtime) dependency list. Print out both directory names
+# Package first-order dependency list. Print out both directory names
# and package names.
package-depends-list:
@@ -4543,7 +4543,7 @@
for pkgname in $$installed; do \
${ECHO_CMD} "$$pkgname ${.CURDIR} ${PKGORIGIN}"; \
done; \
- fi; \
+ else \
checked="${PARENT_CHECKED}"; \
for dir in $$(${ECHO_CMD} "${LIB_DEPENDS} ${RUN_DEPENDS}" | ${SED} -e 'y/ /\n/' | ${CUT} -f 2 -d ':') $$(${ECHO_CMD} ${DEPENDS} | ${SED} -e 'y/ /\n/' | ${CUT} -f 1 -d ':'); do \
dir=$$(${REALPATH} $$dir); \
@@ -4562,7 +4562,8 @@
else \
${ECHO_MSG} "${PKGNAME}: \"$$dir\" non-existent -- dependency list incomplete" >&2; \
fi; \
- done
+ done; \
+ fi
# Print out package names.
--------------050605030609090400050402--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?427ED102.6040406>