El sáb, 20-03-2010 a las 14:12 -0700, Dale Henrichs escribió: > Miguel, > > I should emphasize that conditional load/compile in Metacello is a pragmatic > feature. I don't disagree that the combinatorials can quickly spiral out of > control, but for the short term there is no alternative. I won't argue that > conditionals _should_ be included in the package management layer, either. > > As with all other problems there are several ways to "skin the cat". I > _could_ have extended Monticello to have platform-specific code components. I > _could_ have added conditional compilation to the Smalltalk compiler. In many > ways those solutions are uglier than putting conditional loading into > Metacello:) > > In the end I added the minimum set of features to Metacello that solved the > problems that I was facing on a daily basis and conditional load/compile was > high on my list. > > With that said, I encourage and support any effort aimed at avoided the use > of conditionals:)
Dale, I agree with your arguments and I am not criticizing or commenting about the features of Metacello, I am a happy user of Metacello as you know, and it is very easy to use it. I know that the conditionals are a fundamentally pragmatic solution (an this is good, as that is the way that most real world problems are solved, by being pragmatic and solving the problem, not caring if it is pure design or not. :) But I think that with enough time (or maybe never, miracles happen, they say), Pharo and Squeak will be different enough that will be difficult that a universes scheme can work for several forks or Smalltalk versions. I think that they will need each one its own package set. For example, squeak5.0 universe will have package A, B and D pharo3.0 universe will have package B, C, H and K with package B being a ConfigurationOfB that has been copied to both repositories and that maybe have parted ways and both of them have different versions methods, for example: squeak5.0 has ConfigurationOfB version1 ConfigurationOfB version2 ConfigurationOfB version3 and pharo3.0 has ConfigurationOfB version1 ConfigurationOfB version2 ConfigurationOfB version3 ConfigurationOfB version4 ConfigurationOfB version5 And maybe even version3 is different in squeak5.0 and pharo3.0 universes. But we are theorizing, so far Metacello is working very good, we'll attack the problem when is presented. So, keep the good job you are doing Dale. Cheers > > Dale > ----- "Dale Henrichs" <[email protected]> wrote: > > | Miguel, > | > | Keep in mind that the conditional load code in Metacello is the moral > | equivalent of #ifdef in C. The debian package universe has the > | advantage of pushing the conditional compilaton of _source code_ down > | a level. > | > | With Metacello, I purposely use the version attribute for conditional > | loading of code (as close to a conditional compile as you can get). In > | Metacello we do the load and the compile... > | > | Where common code can be written it should be, but as versions and > | platforms diverge in their API, something has to give ... If you can > | afford to seal off the source code for an older version then that's > | fine, but in a project that is under active development and is > | _expected_ to run on multiple versions of multiple platforms one must > | resort to "conditional loading" - conditional loading is what makes > | Grease work, BTW. > | > | When you look at the universe of #ifdef in C, you see that some of the > | ifdefs are adhoc and some of the ifdefs are based upon platform > | constants ... and that is what the metacello attributes give you ... > | Right now, you can create project-specific attributes and there are a > | handful of pre-defined attributes ... As I said, I am purposefully > | avoiding adding new pre-defined attributes willy nilly, but anticipate > | that some will be needed (GemStone3.0 and GemStone2.4 and even > | GemStone2.3 need to have difference code compiled/loaded for projects > | that are in active developement and expect to be run on all three > | versions of GemStone the low-level apis are different and different > | code has to be written for each different version). > | > | Having the conditional compile/load component in Metacello isn't > | necessarily an elegant feature, but given that no other part of the > | Smalltalk system currently provides a conditional compile/load, I > | added it to Metacello. Since I am maintaining common software on three > | different versions of gemstone, I consider it a requirement. > | > | If and when an elegant conditional compile/load solution becomes > | available, then this particular feature of Metacello can be allowed to > | die in peace. > | > | I will agree that the MYSQL maintainers don't worry whether their > | executables load on Ubuntu or Fedora, but I can almost guarantee you > | that they have conditional code that makes sure that there their code > | will compile/build on the different OSes. > | > | The ifdef world is as ugly as it comes, but it does exists for a > | pretty good reason. > | > | > | Dale > | ----- "Miguel Enrique Cobá Martinez" <[email protected]> wrote: > | > | | El sáb, 20-03-2010 a las 21:54 +0300, George Herolyants escribió: > | | > Thanks for the explanation, Dale! > | | > > | | > May be it's obvious and I'm asking stupid questions but then I > | | don't > | | > understand how can I specify in my configuration the differences > | | > between the versions of the target platform? > | | > > | | > 2010/3/20 Dale Henrichs <[email protected]>: > | | > > George, > | | > > > | | > > The mcz files containing the Metacello configurations _are_ > | | copied, but the configuration itself is not modified in the scheme. > | | > > > | | > > The idea of using different repositories is to indicate which of > | | the many different projects are _expected_ or _known_ to work in > | | Pharo1.1 or Pharo1.0. > | | > > > | | > > For GemStone, I have createed a GemSource MetacelloRepository > | and > | | I have copied the metacello configurations that I know have been > | | ported to GemStone. I've done this so that folks can know what is > | | supported. I make sure that all of the tests run green and that all > | of > | | the configs in the GemSource MetacelloRepository can be loaded. > | | > > > | | > > It does not stop someone from loading a configuration into > | | GemStone that isn't in the GemSource MetacelloRepository, but if you > | | do and run into trouble, my answer will probably be that it hasn't > | | been ported yet. > | | > > > | | > > The same thing can apply to Squeak, but someone would have to > | step > | | up to testing/approving metacello configurations that run with the > | | various versions. > | | > > > | | > > Without this extra step of approval/testing, it's difficult for > | | someone that doesn't have their finger on the pulse of the latest > | and > | | greatest happenings to know what to expect ... > | | > > > | | > > Dale > | | > | | The problem is, you are thinking in upstream development, and the > | | idea > | | of universes is that each package has maintainer for each platform > | | involved. This maintainer can be the same person that the upstream > | | developer, but it can also be someone else that make sure that a > | | given > | | package load correctly in *a give* fork. > | | > | | Making an analogy, what you are saying is that the maintainers of > | | MySQL > | | deb package in Debian should be worried about the package not > | loading > | | in > | | Ubuntu or Fedora. That is not the case. The upstream code can run in > | | all > | | of three distros, but the packages specific to each one (.deb in > | | debian > | | and ubuntu and .rpm in fedora) *must* be build and tested for their > | | own > | | maintainers. The upstream developers can rightfully don't give a > | dime > | | about where is package is being used and maybe they only release > | | a .tar.gz and know nothing (but also they don't stop efforts) about > | | external package for specific distros. > | | > | | So, returning to your question, maybe sometime in the future Squeak > | | (if > | | interested) will have to build its own repositories with its own > | | universes and put there the configurations that are proved to work > | in > | | Squeak. > | | > | | Right now as Pharo and Squeak are very similar the configurations > | | (and > | | the #common you mention) works, but this isn't going to scale for > | | different Smalltalks in the long term. > | | > | | Cheers > | | > > | | > _______________________________________________ > | | > Pharo-project mailing list > | | > [email protected] > | | > > | http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > | | > | | -- > | | Miguel Cobá > | | http://miguel.leugim.com.mx > | > | _______________________________________________ > | Pharo-project mailing list > | [email protected] > | http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project -- Miguel Cobá http://miguel.leugim.com.mx _______________________________________________ Pharo-project mailing list [email protected] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
