[arch-dev-public] Fwd: Perl packaging guidelines.
Message below is trimmed, but the summary is: pacpan (CPAN wrapper for pacman packages) has been updated with gusto. There is a request to include using pacpan as part of the official guidelines for packaging perl apps for Arch. I think this is a good idea, but it leaves me wondering what we should do with the tool itself - should we be hosting it in the git repos? What do you guys think? ---------- Forwarded message ---------- From: Xyne <xyne@archlinux.ca> Date: Fri, Feb 5, 2010 at 6:18 AM Subject: Perl packaging guidelines. To: aaron@archlinux.org Hi Aaron, You're probably at least vaguely aware of a tool that I wrote a while ago named pacpan, which can autogenerate PKGBUILDs for CPAN packages. In the past worked well enough in most cases but there was room for improvement and I finally found the time to rewrite it recently. It is now a very useful tool for packaging CPAN distributions. I don't want to bore you with the details but it now does the following: *) correctly maps all modules to their distributions *) maps distributions to unique pacman pkgnames which follow the guidelines *) detects depends, makedepends, optdepends, and provides *) can detect missing provides for installed packages *) can update the local database provides while waiting for the packager to update it The provides are actually really important because many official packages do not list the packages that they provide and/or deviate from the standard naming conventions and this causes problems when trying to install other CPAN packages. Patching the local database works but it's an ugly hack. I've posted a thread on the forum to discuss changing the perl packaging guidelines to officially recommend using pacpan as a starting point for all CPAN packages. In most cases the PKGBUILD it generates should work out of the box. In those when it doesn't, it will still provide a comprehensive provides array and contain the correct name. This will improve the CPAN packaging situation by enabling full dependency resolution and by preventing conflicts from people effectively packaging the same packaging under different names and omitting the provides of those packages. ...snip... If I get some sort of official approval for this proposal, I will update the Perl packaging guidelines page in the wiki and begin to open tickets on the bug tracker to update/fix current CPAN packages which are incorrectly named and/or lack complete provides arrays.
On 06/02/10 09:28, Aaron Griffin wrote:
Message below is trimmed, but the summary is: pacpan (CPAN wrapper for pacman packages) has been updated with gusto. There is a request to include using pacpan as part of the official guidelines for packaging perl apps for Arch. I think this is a good idea, but it leaves me wondering what we should do with the tool itself - should we be hosting it in the git repos?
What do you guys think?
I'm not sure about the idea. From what I understand, Perl comes in CPAN "distributions", which contain some modules. We take a "distribution" and make a package. This script fills out a provides array with all modules that are part of the package and makes the dependencies be these modules. That may be technically more correct, but the sacrifice is a lot of complexity that I am not sure is really needed. How often do modules swap from one "distribution" to another? Robustness to that is the only advantage I see here. I might be missing something... Saying that, in total perl modules on my system are what is needed for SDL_perl (yay frozen bubble!) and that is the limit of the perl packaging I have to deal with. Allan
On 02/06/2010 10:29 AM, Allan McRae wrote:
On 06/02/10 09:28, Aaron Griffin wrote:
Message below is trimmed, but the summary is: pacpan (CPAN wrapper for pacman packages) has been updated with gusto. There is a request to include using pacpan as part of the official guidelines for packaging perl apps for Arch. I think this is a good idea, but it leaves me wondering what we should do with the tool itself - should we be hosting it in the git repos?
What do you guys think?
I'm not sure about the idea.
From what I understand, Perl comes in CPAN "distributions", which contain some modules. We take a "distribution" and make a package. This script fills out a provides array with all modules that are part of the package and makes the dependencies be these modules. That may be technically more correct, but the sacrifice is a lot of complexity that I am not sure is really needed.
How often do modules swap from one "distribution" to another? Robustness to that is the only advantage I see here. I might be missing something...
Saying that, in total perl modules on my system are what is needed for SDL_perl (yay frozen bubble!) and that is the limit of the perl packaging I have to deal with.
Where the rubber hits the road on these issues is packaging software that requires perl modules. Most that I've encountered require particular versions (or greater) of particular perl modules. It's then the job of the perl-module-dependent packager to try to figure out what version of perl incorporated that module. Then as perl incorporates more modules into the main distribution, all the perl-dependent packages must be changed/updated. There are a lot of packages with perl module dependencies. As at least one step, it seems a good idea for perl to fully list the component module provides for what it includes in the distribution. This is would seemingly benefit packagers of perl-module-dependent packages greatly, and it doesn't add complexity over what we currently have (a partial such set of provides). As for the rest of the proposal, it's less clear to me the right path. Perhaps others will have experience to share here. - P
---------- Forwarded message ---------- From: Xyne <xyne@archlinux.ca> Date: Sat, Feb 6, 2010 at 2:32 PM Subject: [arch-dev-public] Fwd: Perl packaging guidelines. To: aaron@archlinux.org Hi, Sorry that I'm posting outside of the thread. I just subscribed to the list so I didn't have the previous messages to reply to.
I'm not sure about the idea.
From what I understand, Perl comes in CPAN "distributions", which contain some modules. We take a "distribution" and make a package. This script fills out a provides array with all modules that are part of the package and makes the dependencies be these modules. That may be technically more correct, but the sacrifice is a lot of complexity that I am not sure is really needed.
How often do modules swap from one "distribution" to another? Robustness to that is the only advantage I see here. I might be missing something...
I didn't understand what you meant by "makes the dependencies be these modules". You're right that pacpan fills out the provides array with the modules contained in a distribution and that modules do not jump between distros very often. I considered that as well, but that is not a rule and distributions are occasionally split and merged. This would prevent breakage in such cases.* The real advantage of this is that the metadata provided by CPAN distributions in META.yml files often specify specific modules, not distributions. The upstream packagers and CPAN itself obviously think that this is the right way to do it, probably for the aforementioned possibility of splits and mergers. Including the provides, depends, makedepends and optdepends (with occasional tweaks) produced by pacpan is hardly difficult for the packager and that is exactly what these arrays are for. It thus guarantees that pacman can handle dependencies regardless of what happens upstream and it follows CPAN's packaging standards, which is thus in line with Arch's goal of being as vanilla as possible. Coincidentally, the Arch Haskell group recently decided to make it official policy to explicit list provides and depends in all Haskell PKGBUILDs because they've found that this is a better solution for long-term dependency handling. Considering how this simply uses existing pacman functionality, I don't really understand your argument about complexity either. All the packager has to do "pacpan --get-pkgbuilds <foo>" and check the output, then, after installation, optionally run "pacpan --check-local" to double-check. This presents no added complexity to pacman itself. This is far simple than not specify explicit dependencies and dealing with breakage when it occurs. The latter approach is lazy and short-sighted in my opinion, especially in light of how trivial it would be to prevent. Aside from the comprehensive arrays, pacpan will also recommend standard names for modules which do not have them, and this makes finding packages easier if applied to all packages in the official repos as well as the AUR and elsewhere. It prevents conflicting and/or redundant packages which package the same distribution under different names, it turns up the correct package when searching for modules, etc. Having a tool which provides a standardize starting point for all CPAN modules should be a good thing and I think it could do for Perl packages what cabal2arch has done for Haskell packages. * Just to show that this actually does happen, this is from my pacman.log: The following official packages can be removed since the modules are now included in the standard perl library: perl-archive-tar perl-compress-raw-zlib perl-compress-zlib perl-extutils-cbuilder perl-io-compress-base perl-io-compress-zlib perl-io-zlib perl-module-pluggable perl-pod-escapes perl-pod-simple perl-module-build perl-version These community packages are also included in the standard perl library: perl-archive-extract perl-cpanplus perl-digest-sha perl-file-fetch perl-extutils-parsexs perl-ipc-cmd perl-locale-maketext-simple perl-log-message perl-log-message-simple perl-module-corelist perl-module-load perl-module-load-conditional perl-module-loaded perl-module-pluggable perl-object-accessor perl-params-check perl-term-ui perl-time-piece As you can see, things do change and having explicit dependencies will only prevent breakage. Regards, Xyne
* Just to show that this actually does happen, this is from my pacman.log:
The following official packages can be removed since the modules are now included in the standard perl library: perl-archive-tar perl-compress-raw-zlib perl-compress-zlib perl-extutils-cbuilder perl-io-compress-base perl-io-compress-zlib perl-io-zlib perl-module-pluggable perl-pod-escapes perl-pod-simple perl-module-build perl-version
These community packages are also included in the standard perl library: perl-archive-extract perl-cpanplus perl-digest-sha perl-file-fetch perl-extutils-parsexs perl-ipc-cmd perl-locale-maketext-simple perl-log-message perl-log-message-simple perl-module-corelist perl-module-load perl-module-load-conditional perl-module-loaded perl-module-pluggable perl-object-accessor perl-params-check perl-term-ui perl-time-piece
As you can see, things do change and having explicit dependencies will only prevent breakage.
I'm +1, but don't feel very strongly about this. I see that it's more complex than what we have now, but I think the world of perl dependencies is pretty complex, and the existing system isn't great at handling either. Modeling it in this more complete way and mirroring the upstream model seems like a good long-term choice in terms of reducing work for packagers. - P
On 06/02/2010 00:28, Aaron Griffin wrote:
Message below is trimmed, but the summary is: pacpan (CPAN wrapper for pacman packages) has been updated with gusto. There is a request to include using pacpan as part of the official guidelines for packaging perl apps for Arch. I think this is a good idea, but it leaves me wondering what we should do with the tool itself - should we be hosting it in the git repos?
What do you guys think?
This is an area where I have a pretty solid experience, as as perl dev, arch user, and maintainer of several perl packages in extra and community. I tend to agree with Allan here. We should only package and take into consideration what on CPAN is called a distribution, i.e. a tarball containing a bunch or related perl modules. The mapping between modules and distributions is available in a plain text database on each CPAN mirror, and can be figured out by the common tools used to install cpan stuff from the command-line (cpan and cpanp). I think it is quite a BAD idea to put all module names in the provides array, as this can easily yield hundreds of elements without obvious advantage I can think of. Our poor lil Pacman has better things to do than take that overload into account. That said, I do acknowledge Xyne's effort! I have written a very similar tool years ago, which is still available in AUR (perl-cpanplus-pacman), but for which I have alas not dedicated as much effort and commitment as Xyne did with pacpan. My own approach has however inspired another project, called perl-cpanplus-dist-arch (also in AUR), which IMHO is superior to both my own cpan4pacman and Xyne's pacpan. (That said I still use cpan4pacman (together with a few helper shell scripts and the devtools) to maintain my own local repository of 550 CPAN packages, all of which I keep uptodate with relative ease). So -1 from me: the added complexity this proposal would bring clearly outweighs its advantages! But OTOH I am fine with the idea of hosting pacpan on git, it is a good tool. I just don't want the policy it adheres to become our official one. F
participants (4)
-
Aaron Griffin
-
Allan McRae
-
Firmicus
-
Paul Mattal