On Tue, Mar 2, 2010 at 2:58 PM, Sven Neumann wrote:

On Tue, 2010-03-02 at 08:01 -0500, lloyd konneker wrote:

I wonder if importing a plug-in from another plug-in is really

something

that we want to support. If the goal is to share code, then perhaps

the

code that is worth sharing should be factored out into a Python

module

that the plug-ins can import.

Every Python program is also able to be a python module that plug-ins can import. We should preserve this feature of the language.

(For example, one can implement an app with a comand line interface, and then just add a GUI in another file that uses the functions defined on the stand-alone first file).

David Gowers wrote:

I've considered this problem a fair bit, and my opinion is that if you
want this functionality, you should simply guard your register()s. We
cannot safely 'co-opt' python plugins that are not written with this
functionality in mind, as they are designed to be run always in an
independent process (hence they may do initialization which confuses
the calling program, or vice versa); there is no modification to GIMP
which could permit that, it is a logistical problem not a technical
one.

Allowing python plugins to make separate modules available on installation, similar to Sven's suggestion, seems to me the most practical suggestion. This means we would add two items to sys.path --
one the site modules* directory, and the other the modules* directory
belonging to the specific user, which the installation of the plugin
package could put modules into.
We could further postulate that the normal python modules directory
should be the destination of modules that do not require GIMP running
in order to function, and only GIMP-requiring modules would be installed in it's modules* directory. I make this distinction because there are various good reasons not to
install gimp-dependent modules in the global namespace (for example,
pydoc and the general help() facility get confused because the imports
of gimp modules fail.. so you can look up a specific module, but not
search.)

* I realize 'modules' is a term already used in the gimp directory
structure. This is meant as a placeholder for something else...python-modules?

I agree with Joao S. O. Bueno and disagree with David Bowers.

Its better to make plugins meet the normal expectations of Python programmers (you can import any Python file to use pieces of it) than to add new conventions and directories for shared Python plugin code. Simpler is better?

Often, authors don't plan their code will be useful to others. It's just serendipity. If authors don't plan to share code and put it in these new directories, it thwarts serendipitous reuse and experimentation.

It is inconsistent for a duplicate call to register() to be harmless with a warning while a duplicate call to main() is fatal.

Also, new conventions and directories does not solve my wish to call pydoc on plugins (which is not very important. I hope to download a prototype Inspect plugin to gimp registry soon.) But it illustrates that you can't always anticipate what people will want to do.

Thanks

On Sat, 2010-03-06 at 08:41 +1030, David Gowers wrote:

On Tue, 2010-03-02 at 08:01 -0500, lloyd konneker wrote:

I agree with Joao S. O. Bueno and disagree with David Bowers.

Well, I'm glad you don't disagree with me, but who is that?

Sorry I mangled your name. BTW David Bowers is a famous environmentalist, Sierra Club etc. Mr. Gowers was a pharmacist in the movie "Its a Wonderful Life." He boxed Jimmy Stewart up the side of the head, like you did me. Just kidding.

Its better to make plugins meet the normal expectations of Python programmers (you can import any Python file to use pieces of it) than to add new conventions and directories for shared Python plugin code. Simpler is better?

The normal expectations of Python programmers are that the main loop will be guarded.

Granted.

This is exactly the thing that allows you to import otherwise executable modules;
without it, no guarantees can be made that the code you are importing will not interfere with global state in a way which causes your plugin to crash. You can depend on a specific version of a plugin which has known behaviour; In that case you prevent the user from upgrading that plugin, and you rely on knowledge of the internals of that plugin, which is very likely to get you into trouble if you ever need to upgrade it.

The same principle applies to python scripts that are unguarded: importing them will, at the least, have significant side effects. Often they are entirely incapable of being cooperative.

Hence, it's not practical to import a plugin module unless it specifically indicates that it is fit to be imported (by guarding the main loop)

But Gimp shouldn't protect programmers from themselves. If they import a plugin that is not fit, or create dependencies on versions, let them.

Often, authors don't plan their code will be useful to others. It's just serendipity. If authors don't plan to share code and put it in these new directories, it thwarts serendipitous reuse and experimentation.

It is inconsistent for a duplicate call to register() to be harmless with a warning while a duplicate call to main() is fatal.

Currently, the latter is serving the function of preventing the previously mentioned possibility of side effects of importing a plugin in this way.

If there is another way of preventing that problem, then we could look at removing that aspect of main() calls. Otherwise, we absolutely must not.

Also, new conventions and directories does not solve my wish to call pydoc on plugins (which is not very important. I hope to download a prototype Inspect plugin to gimp registry soon.) But it illustrates that you can't always anticipate what people will want to do.

I wanted to do that too.
Just because you want to do something, doesn't mean it's wise to do so.

Calling pydoc on unguarded python modules is just as unpredictable as importing unguarded python modules. Why? Because the way it gets the documentation is by first importing the module in question. Furthermore, all GIMP plugins' dependency on the 'gimp' (and/or gimpfu, etc..) modules mean that even if they are in the module search path, pydoc's attempt at importing them will fail (because the gimp and gimpfu,etc.. modules are only available to Python scripts that are being run by GIMP).Changing the mechanics of main() would not effect that.

If you don't believe that, it's easy to check; set PYTHON_PATH to your plugins directory (full, absolute path like /home/llkk/.gimp-2.7/plugins), start Python, and try to import one of those plugins as a module.

David

My Inspect plugin IS run by Gimp. Thus importing pydoc and invoking it on plugins DOES work, as far as I can tell by my limited testing, with the patch to Pygimp that I have proposed.

Also, you CAN invoke pydoc on the pygimp modules from an Inspect plugin run by Gimp. It turns out not to be really useful, since the existing documentation for pygimp is rather good.

IF you can't invoke pydoc on the command line on a plugin, it's not because it can't find the pygimp modules (because you can set the path so it will) but for the reason you described: they need Gimp running. I really haven't tested it.

But this does seem to work: start Gimp, open the Python console, import pydoc, and call pydoc.render_doc('gimp'). I believe this works even without the patch under discussion. But pydoc.render_doc('colorxhtml') crashes pyconsole lacking the patch under discussion.

My brain hurts.

Lloyd