August 2000 - Zope-Dev

Re: [Zope-dev] Weird Data.fs corruption... zope-dev@zope.org
by Dieter Maurer 10 Aug '00

10 Aug '00

Martijn Pieters writes: > Ai, NFS! There is a known no-no about using NFS for Data.fs storage, it leads > to data corruptions. Zope and NFS do not mix. Huch, why that? I use Data.fs (occasionally) over NFS and did not yet have had problems. And appending to a file should work over NFS. Locking may be a bit more problematic, but this is not an issue provided I ensure that only on Zope process runs at a time. Dieter

3 2

Re: [Zope] ZPatterns Question
by Steve Alexander 10 Aug '00

10 Aug '00

I've moved your question to zope-dev, as that's where ZPatterns questions probably belong. Roche wrote: > I created a ZClass called "Product" with Dataskin as baseclass. > > I also added a specialist called ProductManager containing a default rack. > Under storage for the rack I selected the newly created Product ZClass. > > My product ZClass needs to query both a RDBMS and the ZODB for > attributevalues. At this point I get lost. > I guess i should add an attributeprovider plug-in??? The easiest way is to use a couple of SheetProviders. Use a PersistentSheetProvider for your ZODB values. You'll have to write your own one for SQL values. You will need to create the appropriate propertysheets in your ZClass class definition in your product. > What should go into the fromexpr and attrsexprs properties of the > attributeprovider? Should one add an attributeprovider for each attribute > of the product? See the IRC chat log: http://www.zope.org/Documentation/Chats/zpatt_chat_log An attribute provider can provide multiple attributes. You can have multiple attribute providers if you like. -- Steve Alexander Software Engineer Cat-Box limited http://www.cat-box.net

5 8

RE: [Zope-dev] RE: [Zope] ZPatterns Question
by RC Compaan 10 Aug '00

10 Aug '00

> At 01:57 PM 8/8/00 +0200, RC Compaan wrote: > >I've added a propertysheet called "properties" to my ZClass and i notice > >there is Persistent Sheetprovider under the default rack already. The > >Sheetprovider has properties Sheet_Names and Sheet_Namespaces. I guess > >Sheet_Names should refer to the sheetname i created for my > ZClass??? but how > >do Sheet_Namespaces come into play. > > Actually, neither relates. Property sheets created on ZClasses > always have > their data stored in attributes of the object itself, so if you want to > control those property sheets you would not use a sheet provider at all. > Sheet providers are only used to provide property sheets whose data is > stored external to the object. > Oh ok, this divide clears it up a bit. I somehow imagined that an object's properties are always stored inside the object and that Providers are the managers of those properties. > The Specialist does not talk to the sheet provider. The Specialist asks > its racks for an object. The Rack which "retrieves" the object tells the > object that it belongs to the rack and should ask the rack for anything it > needs. When the object needs a property sheet or attribute or > whatever, it > asks the rack to give it a list of relevant providers for that > attribute/sheet/whatever. It then walks the list asking those > providers if > they can perform the function it desires. If none of the providers are > successful, it performs a default behavior (such as raising an > AttributeError to indicate the attribute does not exist). This general > pattern is followed for almost anything that can be delegated from a skin > to a provider. I considered the scenario where attributes are stored external to the object according to the filing clerk analogy on the wiki: The Specialist (filing clerk) asks its racks (filing cabinet) for an object (dental records). The rack (filing cabinet) tells the object (the dental records) that it belongs to the rack AND that the object should ask the rack for anything it needs. Why would the object "need" anything. Is it not the specialist with the "need"? If object data is stored external to the dataskin object does the dataskin only implement behaviour? I find it difficult to understand because I associate Dataskin with Storage. If object data is stored externally is it stored in Persistent Property Sheets? How does one define properties for Persistent Property Sheets? Roché

2 1

zpatterns: how to invalidate the attribute cache?
by Jephte CLAIN 10 Aug '00

10 Aug '00

Hello, I feel a need to invalidate a dataskin attribute cache after it has been edited, because other objects might want to use the just modified dataskin, but with the new values. Of course, these values come from some attribute providers that should be re-queried. Any thoughts? It seems that the only way is to redefine _v_attrCache. Please comment this patch: --- DataSkins.py.orig Thu Aug 10 17:20:53 2000 +++ DataSkins.py Thu Aug 10 17:21:47 2000 @@ -97,6 +97,11 @@ """Return attribute cache""" return self._v_attrCache + def _invalidateCache(self): + """invalidate the current attribute cache""" + v = self._v_attrCache = {} + return v + def _v_readableSlot(self,_v_dm_=_v_dm_): return self.__dict__[_v_dm_]._readableSlotFor(self) regards, jephte.clain(a)univ-reunion.fr

2 1

Permission
by Tom Deprez 10 Aug '00

10 Aug '00

Hi, I have this q'n already a long time in my head and I still don't know the answer to it. It has to do with the security-view. I hope that someone can shed a light on my brains. Why is their an Acquire permission (referred to as acq_perm in following text) checkbox column? In order to change a permission for a certain role, we have to check or uncheck this permission at the roles permission checkbox. However, for this to work we have to uncheck the acq_perm column (because if this one is checked, the permission is taken from parent-folder-objects anyway, the value of the role checkbox doesn't counts at that moment). As a drawback, this means that all roles have to be checked/unchecked, because they don't acquire the permission anymore from one of its parents-folder-objects. Now, why couldn't this be implemented like the following? Isn't this easier to grasp? The Acq_perm column doesn't exists. Assume that acquisition of a permission is always Active. The checkboxes of the role show their actual value (according to the acquisition). Thus if permissionA is checked in the parent-object, permissionA is also checked. If you want, for a certain role that this permission is not given, you uncheck permissionA. The subobject will show an unchecked permissionA box (because it acquires from its parent). In order to give the subobject again the permission, we only have to check permissionA. A) In the first approach we've to uncheck the acq_perm checkbox and have to set the checkbox of every role according to how we want the permission to be handled by that role. B) In the second approach we've to uncheck the permission checkbox of the role(s). Other roles are not effected. Am I missing something here? Is my explenation somewhere wrong? Why does Zope uses the first approach? Which, sounds for me, a little bit more complicated. Thanks for any explanation! Tom Deprez.

2 4

OFS.objectManager checking object Ids
by Jay, Dylan 10 Aug '00

10 Aug '00

This seems to be a new thing in Zope 2.2, however it is breaking functionality that I had before. I was able to add objects with ids that had a "?" in them. This gets quoted in links and seems to be handled fine by zope (both 2.2 and older versions). Why can't I have question marks in my objects then? Should I submit the inclusion of a the "?" as a valid char in and id as a bug? ---- Dylan Jay Tel: +61 2 9886-8961 Avaya Communication FAX: +61 2 9352 9224 Level 5, 15 Talavera Road Mobile:+61 409 606 171 Nth Ryde, NSW, 2113 ICQ: 520341 AUSTRALIA

1 0

Membership Security Update
by Bill Anderson 10 Aug '00

10 Aug '00

A day or so ago, I made an announcement that there may be a problem with Zope storing passwords for members in the clear in the SystemProperties propertysheet. After som einvestigation, it does exactly what it is supposed to do. That is, store them encrypted with crypt by default. The catch is, if crypt is not importable for some reason, it stores them in plain text. Systems tested Linux, RedHat 6.2: Crypt works fine HP-UX 11.0: Crypt does not import, thus it stores them as plaintext. If you install the Membership product, add a user, and then look at the property sheets for the user (see the wiki), you can verify whether your platform encrypts correctly or not. If your system is not on the list above, please let me know the results. Odds are all the current Linux distributions work fine. Otherwise, it works fine. :^)= Bill -- Do not meddle in the affairs of sysadmins, for they are easy to annoy, and have the root password.

1 0

Broken product instances when using SiteAccess
by Helge Tesdal 09 Aug '00

09 Aug '00

We have set up a site using SiteAccess. The site itself works just fine, but when moving Data.fs to other machines the instances of our products appear broken. There is no problem adding new instances. Currently we're running Zope 2.1.6, and will probably continue to do so unless this is fixed... We have copied the whole zope directory from our main site, including all installed Products, Extensions and so on. Could it be that Zope is looking for the product in the wrong places? We had SiteAccess running when the instances were added, and but have to turn it off when used on other machines. The rewriterule is included below. Any suggestions on how to make this work again? Is it possible to get some more debug info from Zope on where the instance wants the product to reside? Our rewriterule: ========================================== <dtml-call expr="REQUEST.setURL(base='http://toplevel.domain')"> <dtml-let path="REQUEST.path"> <dtml-if path> <dtml-comment> There is a path </dtml-comment> <dtml-if expr="path[-1]=='Z'"> <dtml-comment> '/Z/...' explicitly forbids redirection to 'artists' </dtml-comment> <dtml-call expr="path.pop()"> <dtml-call expr="REQUEST.setURL(path='Z')"> <dtml-elif expr="path[-1] in objectIds()"> <dtml-comment> Don't redirect existing objects </dtml-comment> <dtml-else> <dtml-comment> Everything else goes to 'artists' </dtml-comment> <dtml-call expr="path.append('artists')"> <dtml-call expr="REQUEST.set('SiteRootPATH', '/')"> </dtml-if> </dtml-if> </dtml-let> ============================================== Thanks in advance -- Helge Tesdal tesdal(a)v160b.studby.ntnu.no

1 0

Re: [Zope] Permission
by Dieter Maurer 09 Aug '00

09 Aug '00

Tom Deprez writes: > The Acq_perm column doesn't exists. Assume that acquisition of a permission > is always Active. The checkboxes of the role show their actual value > (according to the acquisition). Thus if permissionA is checked in the > parent-object, permissionA is also checked. If you want, for a certain role > that this permission is not given, you uncheck permissionA. The subobject > will show an unchecked permissionA box (because it acquires from its > parent). In order to give the subobject again the permission, we only have > to check permissionA. I do not like the proposal. I want to have some information, when permissions change automatically if something changes further up in the tree (that is what the "Acq_perm" tells). I would, however, like it, if the permissions had 3 values: * inherit setting further up determines whether the permission is granted or denied * explicitly grant permission granted independent of setting further up * explicitly denied permission denied independent of setting further up Dieter

1 0

Weird Data.fs corruption...
by Carl Robitaille 09 Aug '00

09 Aug '00

Hi everybody, I looked around the Zope site and mailing lists, but I wasn't able to find an answer to my problem. I'm using the CVS version of Zope2. I did an update yesterday. Since then, there's been some commits. I just started my web site project using Zope running with ZServer, when I began to have problems doing Cut-Paste operations. I thought at first that I messed the permissions or something because of my lack of experience with Zope. Little did I know at that moment that my Data.fs was corrupted very badly. So I tried to restart Zope, but it refused giving me this backtrace: [zope@petechie ~/Zope2]$ ./start Traceback (innermost last): File "/home/zope/Zope2/z2.py", line 554, in ? exec "import "+MODULE in {} File "<string>", line 1, in ? File "/home/zope/Zope2/lib/python/Zope/__init__.py", line 110, in ? DB=ZODB.FileStorage.FileStorage(Globals.BobobaseName) File "/home/zope/Zope2/lib/python/ZODB/FileStorage.py", line 303, in __init__ self._pos, self._oid, tid = read_index( File "/home/zope/Zope2/lib/python/ZODB/FileStorage.py", line 1515, in read_index if read(4) != packed_version: raise FileStorageFormatError, name ZODB.FileStorage.FileStorageFormatError: /home/zope/Zope2/var/Data.fs I then logged to Zope web site to get information on how to get my Data.fs fixed. I read the http://www.zope.org/Members/itamar/CorruptedZODB document and downloaded the tranalyzer.py python program. I then understood what the backtrace already told me. The first 4 bytes of any ZODB files should be something like "FS21". Instead, I have garbage in my Data.fs: [zope@petechie var]$ python tranalyzer.py Data.fs | more Traceback (innermost last): error: Not a .fs file (at least, not a kind I know about) magic = 6TÆ Just for the record, Data.fs.in is read easily by tranalyzer.py. How can this happen? Any idea? My only "guess" is that it has to do with the Data.fs file being used by ZServer over an NFS partition on Linux. This is only a guess, but I really can't see any other thing I could have done wrong. This doesn't really bug me for the moment as I only had a couple of hours put in my site, but I would like to prevent it from happening again.... yes I do backups, but only once a day ;-) Thanks for your help/comments. Carl Carl Robitaille robitail(a)giref.ulaval.ca

3 4