Re: pkg/56125 (Many haskell related errors when running pkgin upgrade)

[Jonathan Perkin <jperkin%joyent.com@localhost>]

The following reply was made to PR pkg/56125; it has been noted by GNATS.

From: Jonathan Perkin <jperkin%joyent.com@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: pkg-manager%netbsd.org@localhost, pho%netbsd.org@localhost, pkgsrc-bugs%netbsd.org@localhost,
	gnats-admin%netbsd.org@localhost, nia%pkgsrc.org@localhost
Subject: Re: pkg/56125 (Many haskell related errors when running pkgin
 upgrade)
Date: Tue, 27 Apr 2021 15:44:22 +0100

 * On 2021-04-27 at 15:03 BST, pho%NetBSD.org@localhost wrote:
 
 >I just updated Makefiles so that they now set 
 >CHECK_SHLIBS_SUPPORTED=no. The resulting binary packages should no 
 >longer contain invalid REQUIRES entries.
 
 I just want to register my opposition to this approach, it's completely 
 wrong, and just hides bugs.  Historically in pkgsrc we have tried to be 
 better than this, and it's very sad to see that we're now happy to just 
 churn out broken packages by turning off warnings rather than actually 
 fixing things.
 
 I appear to be in the minority however, and I have no authority to ask 
 that this be reverted or anything, but I do want it on record that I 
 believe this is a terrible approach to software engineering, and would 
 encourage it to be fixed properly.
 
 It should also be pointed out that this was only a problem on NetBSD, so 
 we're now producing packages with no safety checks on platforms that 
 were previously fine, e.g. SmartOS.
 
 CHECK_SHLIBS_SKIP and CHECK_SHLIBS_SUPPORTED are only supposed to be 
 used in situations where the files in question are rebundling of 
 existing binary files, or are very special files.  The list of matches 
 across all of pkgsrc is very small and with the exception of the ghc 
 packages fit this description.
 
 We do not support $ORIGIN and never will, it is the wrong approach when 
 building software that is not designed to be relocatable.  pkgsrc 
 packages have one and only one prefix, and packages should always be 
 built with specific rpaths to only those locations.  The end result is 
 vastly more reliable software, and it's somewhat ironic that it is 
 Haskell, a language purportedly for "correctness", that is the only one 
 in pkgsrc that is abusing these rules.
 
 Regards,
 
 -- 
 Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com