Re: Re: cvs: pear(MAINT_1_1) /Net_NNTP package.xml

From: Date: Fri, 02 Dec 2005 20:47:14 +0000
Subject: Re: Re: cvs: pear(MAINT_1_1) /Net_NNTP package.xml
References: 1 2 3 4  Groups: php.pear.qa 
Request: Send a blank email to pear-qa+get-3301@lists.php.net to get a copy of this message
Hi Greg, and Sebastian, First of all I’d like make it clear that I’ve been more or less out of office when it comes to Net_NNTP for at least a year, and that I am currently preparing the package for a possible future status as unmentioned! For quite some time I’ve been unhappy with the progress within the PEAR channel – the installer etc. not included – and I’m not certain if I want to participate any longer. At the moment I have some spare time, and I thought I might finish my current work here so that I can leave at will any time. From my point of view the upcoming and for some reason apparently feared Zend PHP Framework is a result of the failures in PEAR. PEAR is now just one among many code directories of sometimes doubtful quality… Greg Beaver wrote: > > Sebastian Nohn wrote: > >>Greg Beaver wrote: >>>> >>>This HAS to stop Heino, you can't maintain 3 copies of the same damn >>>program in PEAR. As a matter of fact only one copy of NET_NNTP exists, and only two stable v1+ releases have been published! >>>For the uninitiated, version 1.0.x is stable, version 1.1.x is beta, and >>>version 1.2.x is alpha. Even for the initiated this is exactly how the versioning works – some uses just choose to use beta and even alpha quality code knowing full well that any release with a higher version number might break something! >>>Stop being such a male and commit to an API. Maintaining duplicate >>>packages in PEAR is expressly forbidden, and certainly maintaining 3 of >>>the same package using branches and preferred_state is just loopy. >> >>Then this rule should be adjusted in my eyes. It makes full sense to me >>likt it makes sense having PHPUnit and PHPUnit2. >> >>Why not have a stable 1.0.x and a stable 1.1.x at the same time? Makes >>absolutely sense in terms of configuration management. > > It makes absolutely *no* sense because we do not allow BC breakage in a > minor release, which apparently does not apply to PHP itself. Version > 1.1 of a package should do everything version 1.0 does, unless there is > a critical bug such as a security vulnerability that can only be fixed > by breaking backwards compatibility with the API. Maintaining multiple stable releases makes a lot of sense, but only in very rare cases. If I’m not mistaken, you did in fact yourself once propose that the PEAR package v1.5 could be made PHP5+ only, since package requirements could prevent PHP4 systems from upgrading to any version above v.1.4.*. I do believe this would in fact work, but I honestly think such a decision should result in a rewrite of the current versioning scheme, since the only right solution would be to make an increase of the major version number (without duplicating the package) of the PEAR package so that future maintenance of both branches would be possible. This would of cause make releases of the same package mutual exquisite. I would however still have preferred to have PEAR reduced to the bare minimal amount of packages needed to run the installer – the official installer and all sorts of unmentioned code should definitely not be published via the same channel! >>What I want to say: If someone is willing to do the work of maintaining >>different stable trees of the same software, why would you want to stop >>him from that? Heino should be encouraged to continue his work and >>hopefully other OSS developers adopt that. > > The problem is not that Heino is maintaining different branches, it is > that he is maintaining a "stable" branch, a "beta" branch, and an > "alpha" branch. This is simply not the case! Every goddamn serious software project have a branch marked stable I some way (after the first stable release that is, of cause). As do Net_NNTP, and currently that CVS branch is MAINT_1_0. First of all one can’t tag a branch as being ultimately neither stable, beta, alpha nor devel. The mere fact that I’ve not yet made any stable release above v1.0 is not a result of me stupidly comparing code repository branch stability and release stability. How could I ever make a release candidate? Putting a ‘beta’ (RC is still marked beta, right ?) release in the ‘stable’ branch would break such awkward branching practice, wouldn’t it? (Speaking about release candidate; how about making gamma available as a release state, so that release candidates could be differentiated from actual beta releases) > The labels are supposed to apply to code stability and > API firmness. In other words, "alpha" means the API may change at any > moment, "beta" means the code is almost judged to be stable and "stable" > means the code should work, and is fully documented, and can be depended > on in the most rigorous production environment. Quite frankly I haven’t felt that any of my milestone requirements had been meet, and so there was absolutely no reason to take the package to the next level. And since the documentation (including the phpdoc) wasn’t up to date, I wasn’t even allowed to make such an upgrade. Actually I’ve been extremely conservative regarding the release stability compared to many other PEAR developers. This stands in contrast to so many other packages. Just think about those young packages which have gone from non-existent to stable in only a handful of weeks! Although stable doesn’t have to be ‘Debian’-stable, it would be nice if ‘PEAR’-stable meant well-written _and_ thought through, right? >>The PEAR installer needs to be able to distinguish between bugfix-, >>feature- and incompatibility-releases to make that usable but that >>should be possible. >> > This is already possible, and the recommended way is quite clear. > PHPUnit2 is an incompatibility release of PHPUnit. If Heino wishes to > have 3 versions of Net_NNTP, they should be Net_NNTP, Net_NNTP2 and > Net_NNTP3. And this is, in my opinion, exactly what should never happen unless the package has been completely rewritten for the upcoming NNTP version 2. Wouldn’t it be awkward having both NET_NNTP2 and NET_NNTPv2 in PEAR simply because of package maintainer stupidity? If I’m not mistaken, the IETF working group has put two and a half yours of work and debate into the current NNTP version 2 draft. If those involved could spend so much time updating the standard, I’m most willing to be even more conservative regarding package stability! > This is straightforward, and not confusing. Using stability to do > branches is completely unacceptable as it perverts the meaning of the > system and will only confuse users. Not only does it confuse users, but in the long run users willing to the necessary alpha and beta testing would end up with faulty applications, since they would end up with unmentioned and/or instable releases with higher version numbers than the actual stable code. Greg, honestly I do in fact understand how and why you came to the conclusion that I had gone mad. I’ve updated the info on Net_NNTP’s package-homepage – the choice of words was properly the most misleading… Kind regards, Heino H. Gehlsen

« previous php.pear.qa (#3301) next »