Date: Sat, 10 Mar 2007 06:42:31 -0800 From: Ade Lovett <ade@FreeBSD.org> To: Jean-Yves Lefort <jylefort@FreeBSD.org> Cc: freebsd ports <freebsd-ports@freebsd.org>, Ade Lovett <ade@FreeBSD.org> Subject: Re: Ports 104877 causing big problems Message-ID: <D450EB9D-77BB-47AA-A107-B5EC9567F5FD@FreeBSD.org> In-Reply-To: <20070310144528.1224e8f1.jylefort@FreeBSD.org> References: <45F1DDE2.5030404@FreeBSD.org> <BE66AB56-E0B4-420A-910D-9C10DB9AF24D@FreeBSD.org> <45F1EA6A.6070904@FreeBSD.org> <FB399CF7-11E2-4CC9-8C91-7D6850B7B2D8@FreeBSD.org> <20070310023034.c5939c48.jylefort@FreeBSD.org> <7CF1749C-3254-46AC-ABDD-BAB0D84ED7A1@FreeBSD.org> <20070310033000.c9d2a66f.jylefort@FreeBSD.org> <8E182699-3175-447C-92EF-B6F0E84B4244@FreeBSD.org> <20070310144528.1224e8f1.jylefort@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Mar 10, 2007, at 05:45 , Jean-Yves Lefort wrote: > Repeatedly read the PR until you understand it. Pay attention to the > provided objdump diff, and to the fact that the patch modifies the > link_all_deplibs variable. Understand what that variable > does. Understand why it should be set to "no" rather than to > "unknown". There is no empirical evidence in the PR. Statement #1: "This slows down the linking considerably" This one should be easy enough to prove. Providing timings. Pay particular attention to the relative timing of the final link-loader step compared with the time taken to register all dependencies as part of the pkg_install process. Statement #2: "... creates a direct dependency between a library down the tree and a leaf program" Incorrect. Take three libraries; A, B, and C. A depends on B. B depends on C. In order to ensure deterministic behavior in the event of 'C' being upgraded, *all* ports depending on it should also be rebuilt. Thus, if something relatively low level like libpng is updated, then to guarantee at least deterministic failure modes, everything that depends on libpng should also be rebuilt (either manually, or via portupgrade/portmaster, both of which have flags to make this an easy process). Statement #3: "The attached patch fixes the problem." Unknown. There are no comparative timings for the link-loader stage (statement 1), and merely obfuscates the dependency issue (statement 2), particularly given that there are *two* dependency chains being discussed, the one inherent to .la files, and the one provided by the pkg_* tools. > You managed to introduce this regression in one go with your .la > resurrection. Again, empirical evidence, please. Cold, hard, facts, preferably without the negative emotions. ports/94826 may be educational with regards to a possible approach. - -aDe -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (Darwin) iD8DBQFF8sPYpXS8U0IvffwRAnRaAJ4oJUZ+PCoy/7cBqwxmGszNEgFrRwCdGugN 5iEjVNjrxg7Iwng32aGPyE4= =LSgp -----END PGP SIGNATURE-----
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?D450EB9D-77BB-47AA-A107-B5EC9567F5FD>