Steve Spicklemire wrote:
I don't know if anyone out there is actually trying to *use* my ZCVSMixin/ZCVS Folder stuff, but if you are you certainly want to get the latest release, ZCVSMixin-0.1.1.tgz from the zope.org site:
Hi, Steve. This work looks very interesting. I decided to try it out and noticed it was not possible to manage a product folder with ZCVSMixin, so I started hacking up your product. (Images that resemble a horror movie come to mind.) :-) I've attached my changes in a private e-mail so as to not flood the list. Basically I made it possible to override the behavior of ZCVSMixin so that it can manage a folder other than itself. Then I created a new class derived from ZCVSMixin which is a "CVS" Folder. That way, it's possible to manage any objects in the ZODB, regardless of whether CVS had been considered in the structure.
I put some screenshots up there so you can see what it looks like at:
It looks good, but the interface could be a lot cleaner. I envision a table with three distinct column groups: the ZODB version, the FS version, and the repository (with details). When the ZODB and FS are in sync, a small black equal sign will appear between them (preferrably a graphic to make it more visible.) When the FS and repository are in sync, there will be a small equal sign between them also. When something is not in sync, a green arrow pointing in the direction the file needs to go will replace the equal sign. The buttons below would be placed in such a way to make their function obvious. In fact, the submit buttons could be images containing the arrow icon. Thus the interface would be simple and obvious.
I've also found it useful for managing Python products and other software through the web! I have had a couple of minor 'breakthrough's that *seem* cool, but I might just be fooling myself. If an object is not folderish I guess that it's 'simple' so I export it as XML with the following code:
(snip)
This looks good, Steve.
If objects are 'folderish' or 'complex' (and the user is permitted to dictate this when objects are added to CVS), they are exported as 'zexp' files and cvs adds them as 'binary'. I was having trouble with certain kinds of object's xml representation. I decided that this was really an xml implementation problem, since I couldn't reliably export these as xml through the normal Zope export mechanism either. Anyway.. let me know what you think if you are using this stuff!
I think we need to look into performing complete CVS checkouts which create entire folder hierarchies.
Finally I'm aware that there are also security issues with the way this works now. I need to look at all the things a user might need to do and eliminate some of the 'type in' arguments that are in the product now. These arguments wind up on 'os.system' calls, and could be abused by a ill-meaning user (with access to the CVS Admin page!) Any thoughts there?
Imagine this possibility: Zope would start what I'll tentatively call an "execution proxy" server at startup. When a user want to access CVS through Zope, they would be required to start a background "execution client" which makes a connection to Zope's execution proxy server. When the user accesses a CVS function in Zope, Zope would look at the list of connected execution clients and ask the appropriate execution client to access the file system and execute cvs commands based on that user's system privileges. Advantages: - Unless a given user is running the execution client, the possibility of intruders gaining access to that user's files is not increased. - Nothing has to run with root privileges. - The execution proxy could run on a privileged port, making it impossible for intruders to trick users into using a counterfeit execution proxy (unless, of course, the intruder already has root privileges.) - No need for administrators to fiddle with filesystem permissions. Disadvantages: - If Zope is compromised while execution clients are connected, the intruder may be able to execute commands through the execution clients. To reduce the impact, we can reduce the commands allowed to be executed. Note, however, that this problem is very likely inherent in CVS access through the web. Even the current implementation suffers from this problem. - Users will have to manually start their execution clients, which will stop automatically when they log out. This is actually designed to be an advantage in terms of security. On the other hand, it would be possible to write an execution client that runs as root and automatically spawns child clients for each user. That could be dangerous, though. It should be relatively easy to write all this IMHO. Shane