Hi David,
Thanks very much for your follow-up.
On Mon, Jul 24, 2017 at 07:28:07PM -0400, David T. Lewis wrote:
Hi Alistair,
I am copying this to vm-dev for follow up on the plugin, see below.
On Mon, Jul 24, 2017 at 08:09:10AM +0000, Alistair Grant wrote:
Hi All,
I'm nearly ready to submit a patch that started with the goal of being able to retrieve the device id and fixing FileReference>>isSymlink and also follows Esteban's suggestion of splitting out file existence and other attributes (which provides a performace gain). See the summary below for a description of the changes.
The patch involves adding a new VM plugin, FileAttriubutesPlugin. To minimise the chance of any problems along the way I'd like to submit the patch in three steps:
- Add the VM plugin (FileAttributesPlugin)
- Add the code the image that allows testing of the code and plugin, but doesn't do any integration with existing functionality. This will allow the plugin to be tested by a few volunteers (hopefully).
- Add the patches that make the switch over to the new implementation.
I think you are handling this in exactly the right way, kudos.
Can someone point me to how to submit a new VM plugin? The code is contained in a subclass of InterpreterPlugin.
Following up on the vm-dev list: If your plugin is available in a Monticello repository, that would be great because VM builders can easily include it and try it out. Any repository would be fine for starters, or if you have an account on squeaksource.com I can add you as developer in the somewhat loosely-related DirectoryPlugin project if that is of any help. Whatever is convenient for you.
smalltalkhub.com is probably easiest for me, thanks.
Your plugin sounds like something that would be stable and require little maintenance over time, so it might make sense to pull it directly into the VMMaker package. We can discuss that vm-dev list.
I'm hoping this will become a core part of Pharo, so ultimately it should live with the other Pharo core plugins like FilePlugin (assuming it is accepted, of course).
Once your plugin code is available, it should be straighforward to start including it in the various VM build configurations.
Great, thanks. build.linux32x86/pharo.cog.spur is probably the best place to start.
I should be able to post the plugin to smalltalkhub.com quite quickly. The smalltalk code will take me a while to repackage as I'll want to test it fairly carefully, and my time is very scattered (this is my hobby, but lots of family demands :-)).
As a side effect, once this has been fully integrated it will be possible to get rid of at least some of the "#if Pharo" type conditionals in the FilePlugin, allowing the code to be tidied up.
I'll post another reply once I've got the plugin code in smalltalkhub.com.
Thanks again, Alistair
Dave
I've been using this as my production environment for about 2 months now on a linux 32 bit VM. Running the full test suite results in the same set of test failing before and after applying the patch.
I've also ran file related tests on linux 64 bit (run the Test Runner, select all packages with "file" as part of the name and run all the available tests) and the full test suite on Windows 32 bit.
The summary is:
- #isSymlink now works properly on Linux (and it should also work on MacOS and BSD).
- The list of file attributes available from FileReference now is: #accessTime (new) #changeTime (new) #creationTime #deviceId (new) #exists #gid (new) #inode (new) #isBlock (new) #isCharacter (new) #isDirectory #isExecutable (new) #isFIFO (new) #isFile #isReadable #isRegular (new) #isSocket (new) #isSymlink (works) #isWritable #modificationTime #numberOfHardLinks (new) #permissions #size #targetFile (new) #uid (new)
- FileReference>>exists is faster than before (well, at least on my linux laptop). This is useful as it is called quite often.
- It is possible to retrieve symbolic link attributes, e.g. all the attributes listed above plus the target path.
Given how similar MacOS and BSD are to linux, I assume that this will all work without problem on those platforms (but it obviously needs to be tested).
As implied above, the changes are all backward compatible (except the broken #isSymlink), although a couple deserve mention:
- Obviously #isSymlink now answers correctly (previously it would only answer correctly for a broken link).
- Requesting any of the attributes listed above (except #isSymlink) will return the value of the target path. If the FileReference is to a broken symbolic link, it will return the attributes of the symbolic link (keeping existing behaviour).
- The attributes of a symbolic link can be retrieved using FileReference>>symlinkAttributes.
Overall, performance is slightly better than before. Code that needs to access multiple attributes and is written to take advantage of the new behaviour will see significant performance improvements.
If you've got this far and forgotten the original question :-)
Can someone point me to how to submit a new VM plugin?
Thanks, Alistair