On Wed, Mar 19, 2014 at 09:45:14AM +0100, su_v wrote:
On 2014-03-19 06:50 +0100, Bryce Harrington wrote:
On Tue, Mar 18, 2014 at 01:13:52PM +0100, su_v wrote:
On 2014-03-18 12:32 +0100, Daniel Macks wrote:
On Tue, 18 Mar 2014 00:35:48 -0700, Bryce Harrington <bryce@...961...> wrote:
Identify any critical OSX/Win32 packaging issuesconfigure.ac still only accepts lcms (not lcms2) on OS X, with embedded comment "lcms 2.2 & 2.3 have problems on OSX". Upstream now has lcms 2.4; is the problem still present? I don't see a link to a bug report or discussion to be able to know what the problem is.
Discussed in the comments of these reports:
Request to migrate OS X port to lcms2 tracked in:
AFAICT building with lcms1 or with lcms2 on OS X seems ok (personally tested on OS X 10.7.5, with trunk only), as long as there are not both libraries loaded at runtime (pulled in by another shared library, or due to overlinking (e.g. in earlier MacPorts versions)).
Thanks Dan and su_v, this sounds like exactly the sort of packaging problem we want to get resolved up front. Can you two compare notes in more detail and determine if this is indeed a real problem, and let me know the plan of action for addressing it?
In my understanding 'lcms1 vs lcms2' on OS X is not a packaging issue, it's a porting issue.
Actually, porting issues are probably relevant to track with the packaging issues too, if they're severe enough to risk being a "show stopper" class problem that blocks the release.
A possible solution would be a minor change to simply remove the restricting test in 'configure.ac' (the patch attached in bug #1024344 still applies). I'm waiting/hoping for a reply or comment by Jon Cruz (I asked him the same question on irc the night before this issue was raised here on the mailing list).
Okay, thanks. Give Jon one more ping and if you don't get an answer within a couple days, go ahead and commit it. This looks like it would be trivial to back out if there were problems, and the more time we have it in there for folks to test the more confidence we'll gain in it.
As commented in the referenced bug reports, personally I did not want to change configure.ac to enable lcms2 detection (and usage) on OS X without prior feedback about (Jon's) builds on SL: detailed information
SL?
about what type of crashes had been the reason for the restriction, and whether they still occur with clean builds of current trunk using latest lcms2 2.4 or 2.5 (current version in MacPorts), is not available (as dan already mentioned in his mail).
Once lcms2 support is allowed for OS X builds and confirmed to work on OS X, support for lcms1 (including all related ifdefs in the code) could be removed completely, as requested in bug #1133014 - that's not really a porting or packaging issue though.
Right, and unless having it there causes an actual problem, I would suggest leaving it there. That way post-release if people find problems with cms2, they'd need only a minimal patch to revert back to legacy to work around the problem. Then once 0.91 is out the door, the cruft code can go!
Bryce