Mixing recipes (zc.recipe.cmmi reuse)
Hi there, Not sure this is the right list, but let's give it a try. I would like to use the 'patches' functionality from zc.recipe.cmmi together with other recipes. I believe this is useful functionality and is interesting to all sorts of recipes, not only to cmmi-based ones. So the question is, does the zc.buildout architecture support reusing options from a recipe on other recipes? Or would it require the recipe writer to explicitly 'subclass' (?) cmmi to get patch functionality? Or even, would a 'post-fetch'/'pre-build' generalization be desired, 'patch' being one such application? -- Sidnei da Silva Enfold Systems http://enfoldsystems.com Fax +1 832 201 8856 Office +1 713 942 2377 Ext 214
Sidnei da Silva wrote:
Hi there,
Not sure this is the right list, but let's give it a try.
I would like to use the 'patches' functionality from zc.recipe.cmmi together with other recipes. I believe this is useful functionality and is interesting to all sorts of recipes, not only to cmmi-based ones.
So the question is, does the zc.buildout architecture support reusing options from a recipe on other recipes? Or would it require the recipe writer to explicitly 'subclass' (?) cmmi to get patch functionality? Or even, would a 'post-fetch'/'pre-build' generalization be desired, 'patch' being one such application?
I've seen subclassing, or explicit instantiation, e.g. instantiate the dependent recipe class as an instance variable and just call its methods. I don't think you really need any magic from buildout here. Recipes are pretty simple Python classes. Just make sure you pass the options etc from __init__.py and you should be fine. I think plone.recipe.plone uses plone.recipe.distro and plone.recipe.egg in various ways. You may want to look at that. Martin -- Acquisition is a jealous mistress
On Sep 18, 2007, at 6:25 PM, Sidnei da Silva wrote:
Hi there,
Not sure this is the right list, but let's give it a try.
I would like to use the 'patches' functionality from zc.recipe.cmmi together with other recipes. I believe this is useful functionality and is interesting to all sorts of recipes, not only to cmmi-based ones.
So the question is, does the zc.buildout architecture support reusing options from a recipe on other recipes?
It is just Python. Buildout doesn't provide any support fro this because it isn't really needed.
Or would it require the recipe writer to explicitly 'subclass' (?) cmmi to get patch functionality?
I prefer composition. A number of recipes reuse the egg recipe through composition. Generally, an egg will need to explicitly provide a Python API for other eggs. See, for example, http://pypi.python.org/pypi/ zc.recipe.egg/1.0.0b6#egg-recipe-api-for-other-recipes
Or even, would a 'post-fetch'/'pre-build' generalization be desired, 'patch' being one such application?
I have no idea what that is. :) Jim -- Jim Fulton Zope Corporation
On 9/24/07, Jim Fulton <jim@zope.com> wrote:
Or would it require the recipe writer to explicitly 'subclass' (?) cmmi to get patch functionality?
I prefer composition. A number of recipes reuse the egg recipe through composition.
Generally, an egg will need to explicitly provide a Python API for other eggs. See, for example, http://pypi.python.org/pypi/ zc.recipe.egg/1.0.0b6#egg-recipe-api-for-other-recipes
Gotcha, that makes sense.
Or even, would a 'post-fetch'/'pre-build' generalization be desired, 'patch' being one such application?
I have no idea what that is. :)
My intent is: "Given any recipe, no matter where data is coming from (be it an egg in PyPI, a Subversion checkout or a tarball), I would like to be able to perform an operation in the 'local copy' of the data, without depending on the person that wrote the recipe to have allowed me to do so" An example being, after a recipe that does a Subversion checkout runs, my custom operation kicks in and applies a patch to the local copy before the buildout processing continues. I believe that this might be doable today by writing custom configuration, but maybe it's such a common case that could be simplified. Maybe something along the lines of (note: this is all pseudo-config): """ [Step1] recipe = some.recipe.checkout url = svn://url-to-repo/package [Step2] recipe = zc.recipe.cmmi source = {Step1:location} patch = /path/to/my/patch [zope2] products = {Step2:location} """ I believe that something like this might already work today, if not it might be easy to make it work that way. But what I'm after is to avoid Step2 above by listing the patch to be applied in Step1 even if 'some.recipe.checkout' does not support the 'patch' option directly. All in all, if you tell me that the hypotetical Step2 above really can't be avoided and that the best option is to make 'some.recipe.checkout' and any other recipes out support the 'patch' option directly, I would be fine with that too. Does it make more sense now? -- Sidnei da Silva Enfold Systems http://enfoldsystems.com Fax +1 832 201 8856 Office +1 713 942 2377 Ext 214
Sidnei da Silva wrote:
My intent is:
"Given any recipe, no matter where data is coming from (be it an egg in PyPI, a Subversion checkout or a tarball), I would like to be able to perform an operation in the 'local copy' of the data, without depending on the person that wrote the recipe to have allowed me to do so"
An example being, after a recipe that does a Subversion checkout runs, my custom operation kicks in and applies a patch to the local copy before the buildout processing continues.
I believe that this might be doable today by writing custom configuration, but maybe it's such a common case that could be simplified.
Maybe something along the lines of (note: this is all pseudo-config):
""" [Step1] recipe = some.recipe.checkout url = svn://url-to-repo/package
[Step2] recipe = zc.recipe.cmmi source = {Step1:location}
Do ${Step1:location}
patch = /path/to/my/patch
[zope2] products = {Step2:location} """
I believe that something like this might already work today,
It does. Again, this is "just Python". You have a dict-like variable 'buildout' passed to your recipe's __init__(). This has keys for each section, which contains a dict with keys for each option. So above, you could address buildout['Step1']['url'], say. You can also put things into this dict-of-dicts. Once a recipe has been run, subsequent recipes will be able to see things put into the dict.
if not it might be easy to make it work that way. But what I'm after is to avoid Step2 above by listing the patch to be applied in Step1 even if 'some.recipe.checkout' does not support the 'patch' option directly.
How would you avoid naming conflicts? How would you declare where that option comes from? I really don't think the overhead of having to specify a new recipe (as in your Step2 above) is very much, and it's a lot more explicit.
All in all, if you tell me that the hypotetical Step2 above really can't be avoided and that the best option is to make 'some.recipe.checkout' and any other recipes out support the 'patch' option directly, I would be fine with that too.
I'd suggest that's the best way. Only a limited number of recipes would really need 'path' - those could re-use your patch recipe via composition if necessary, but you have a fallback of using the recipe on its own. Martin -- Acquisition is a jealous mistress
participants (3)
-
Jim Fulton -
Martin Aspeli -
Sidnei da Silva