Re: [Kimchi-devel] [PATCH V2 6/7] add an optional to toggle the sample plugin

on 2014/05/21 18:28, Yu Xin Huo wrote: > On 5/21/2014 5:19 PM, Zhou Zheng Sheng wrote: >> on 2014/05/21 14:31, Yu Xin Huo wrote: >>> On 5/21/2014 11:27 AM, Zhou Zheng Sheng wrote: >>>> on 2014/05/21 09:22, Sheldon wrote: >>>>> On 05/20/2014 02:53 PM, Yu Xin Huo wrote: >>>>>> Sample plugin has no difference from other plugins, it is wrong to >>>>>> specially >>>>>> design a command for that sample plugin. >>>>>> A plugin should have a way to disable itself, I prefer the original >>>>>> way to >>>>>> disable a plugin like below. >>>>>> >>>>>> In plugin descriptor xml file, comment out all tabs, if no tabs is >>>>>> defined, >>>>>> then the plugin will not be loaded. >>>>>> By this way, no special command is needed, no additional overhead in >>>>>> coding is >>>>>> needed. >>>> @Yu Xin, Your idea of building plugins inside of Kimchi works for the >>>> plugins that comes with Kimchi originally. However usually third-party >>>> plugins should be able to build themselves separately and outside from >>>> kimchi code repository. What Kimchi needs to do is just to discover the >>>> build result of the pluginX and load it, regardless whether it contains >>>> tabs or not. >>>> >>>> For example, to compile a Linux kernel module, all we need is some >>>> kernel headers that describe the data structures used by the module >>>> interface. We do not need the kernel source itself. I have a example >>>> kimchi plugin that build outside of kimchi in this way. (We are working >>>> on making it open-source.) After I build the plugin separately, I can >>>> just copy the build result files to kimchi's plugins dir and it loads >>>> automatically. >>>> >>>> We also do not enforce all plugins to contain tabs, because some >>>> plugins >>>> can just extend the kimchi API. And the ".conf" file in the specific >>>> plugin dir already contains a "enable = True/False" switch. There is no >>>> need to "comment out all tabs". >>> Obviously, you mean that there are various types of plugins for kimchi. >>> If there is a sample for each type of the plugins, no wonder there will >>> need a command to enable/disable the sample for that type of plugin? >>> >> No, we don't need a command to enable the plugins in build time. The >> third-party plugins are built without Kimchi code, and outside of Kimchi >> source code directory. You can build Kimchi without any plugins then >> publish "kimchi.rpm" package. Another developer builds a plugin >> separately on a different machine in a different time and publish >> "kimchi-pluginX.rpm". The user download all the RPMs and install. >> >> As I said, Kimchi does not provide any plugins, so it does not need to >> build any plugins, and we don't need any command to enable the plugins >> in build time. All Kimchi knows is that it scans the plugins directory >> in runtime and loads all plugins in the directory. >> >> In case a third-party developer want to test his plugin without >> generating and installing a RPM file, he can build the plugin >> separately, then get Kimchi source code and build Kimchi, copy the >> plugin build result to Kimchi source code plugins directory, then at >> last run Kimchi from the source code directory. >> >>> If I understand you correctly, you mentioned something like ".conf" >>> which is configuration file for the plugin to enable/disable itself. >>> I already stated in my previous mail, I prefer a way to get plugin to >>> enable/disable itself. >> You are mixing runtime and build time configurations. The "pluginX.conf" >> file is for the runtime. After the a plugin is installed, it is enabled >> by default. If the admin want to disable the plugin temporally, he can >> edit the "pluginX.conf" file. The build time of the plugins is not >> related to the build time of Kimchi. > All the discussion has *nothing* to do with "build time" at all. It is > all about run-time about how kimchi handle plugin. > Please *stop* to get anything about "build time" involved to make things > confusing. This patch is all about build time configuration. "--enable-plugin-X" is a switch given to the build scripts. The switch sets the default availability of a plugin.

As regard to runtime configuration, manually changing "enabled = False/True" is enough. For now the only valid use case to enable and disable a plugin is that a plugin developer experimenting some new things on the "sample", and this case is rare. I think there's little value to add new command for a rare case. Notice that even the third-party developers do not need to toggle the sample plugin, and only when we want to extend the plugin framework itself, we have to enable the sample and test it. So this use case is really rare.

In all, "--enable-sample" is an acceptable solution for the developer to toggle the sample plugin occasionally in the developing environment. It's also acceptable that we don't have any switches or options at all, and the "enabled = False" is the default in "sample.conf". Whenever the developer wants to experiment something, just manually edit it and restart Kimchi back-end server.

> Anything that is not target for production(for example, the sample) > should *not* be built into the final binary delivery package. True, the sample plugin didn't, doesn't and won't be in the binary package. The switch in this patch is just to toggle its availability when developer runs Kimchi from source code.

> For run-time, 2 scenarios, under development or staged into production > for operations. > > As you said, the plugin can be installed separately as a binary package > on top of kimchi installation. > If the admin want to get rid of any installed plugin, he need a way to > disable it. If the user does not want a plugin, he can just uninstalls it.

> So there is a need to disable plugin. Either 1 or 2 below is ok. > 1. Design a way for plugin to disable itself. > 2. Kimchi provide a command to disable/enable any plugin. > >>>> @Shaohe, @Yu Xin, as a conclusion, we do not need to add any mechanisms >>>> to define which plugin to build/load. In Kimchi upstream, all the new >>>> features should be added to Kimchi itself, not to plugins. The plugins >>>> are for third-party developers and they should build their plugins >>>> separately. Only the sample plugin that serves as a example on how to >>>> write plugins should be treated specially. So --enable-sample is just >>>> enough, we don't draw the feet of a snake. >>> Have a command "--enable-sample" specially for a shipped sample in >>> source distribution is definitely wrong. >>> There is no justification to have additional overhead in coding for >>> non-product stuff. >>>>> Does that mean the user needs to find this file and uncomment these >>>>> codes when >>>>> he want to try the plugin. >>>>> >>>>> Usually, a user wants to add a new plugin, he can reference the >>>>> sample plugin codes. >>>>> He can try it first to how it works and then read the code to how he >>>>> can make >>>>> himself plugin. >>>>> >>>>> we can change "--enbale-plugins" to "--enable-sample". >>>>> our switch command("--enable-sample") is also an example, ti tells >>>>> user how he >>>>> build a plugin with kimchi >>>>> together if he just add one plugin. >>>>> >>>>> Also he can add "with" command for his plugins if he wants to add >>>>> more than one >>>>> plugin. >>>>> such as: >>>>> --with-plugins=plugin1,plugin2,plugin3 >>>>> The "with" command is similar to switch command, a plugin developer >>>>> can read >>>>> the autotool document to >>>>> learn more about it.