Project

General

Profile

Bug #5058

Access wizard mishandles certain combinations of allow/deny elements

Added by Jim Regetz about 9 years ago. Updated about 9 years ago.

Status:
Resolved
Priority:
Immediate
Category:
morpho - general
Target version:
Start date:
06/24/2010
Due date:
% Done:

0%

Estimated time:
Bugzilla-Id:
5058

Description

Morpho substrings XPaths improperly in certain cases when extracting data for the Access wizard, ultimately causing it to unnecessarily punt to the tree editor.

Here are some reproducible cases:

  • If there is an 'allow' element for principal 'public', along with 10 or more 'allow' elements for (non-public) principals, Morpho incorrectly parses the 10th one.
  • If there is an 'allow' element for principal 'public', Morpho incorrectly parses any 'deny' elements for (non-public) principals.
  • If there is a 'deny' element for principal 'public', along with at least one 'allow' element and at least one 'deny' element for (non-public) principals, Morpho incorrectly parses those 'deny' elements.

For a live example of the first case, try to open the Documentation->Access Information interface for this data package:
http://knb.ecoinformatics.org/knb/metacat/krobinson.12.6/nceas

I believe the problem lies at least partly in lines 1304-1305 of AccessPage.java (as of rev 4925). For the cases above, this.xPathRoot.length() isn't actually the right value to pass to the substring statement. Depending on the case, the trimmed XPath ends up either having an extra character at the start, or having its first character truncated.

From stderr.log, here's an example of the first case above (note the extra ']' as the first character of the trimmed nextXPaths):

...
Access: nextXPath = /access/allow10/principal1
nextVal = user10
Access: TRIMMED nextXPath = ]/principal1
Access: nextXPath = /access/allow10/permission1
nextVal = read
Access: TRIMMED nextXPath = ]/permission1
...

And here's an example of the second case (note the missing first letter in the trimmed nextXPaths -- actually, maybe they're missing a leading '/' too?):

...
Access: nextXPath = /access/deny1/principal1
nextVal = user09
Access: TRIMMED nextXPath = rincipal1
Access: nextXPath = /access/deny1/permission1
nextVal = read
Access: TRIMMED nextXPath = ermission1
...

Access.java.patch (1.9 KB) Access.java.patch Jim Regetz, 06/25/2010 11:58 AM

History

#1 Updated by Jim Regetz about 9 years ago

I think there are a few separate issues, all in the Access setPageData method. Proposed patch attached, with explanations below.

1. The paths in accessAllowList and accessDenyList are each constructed with their own indices, like this:

/access/allow1
/access/deny1
/access/deny2

...but the while loops that iterate over these two lists use a common accessPredicate counter, as though they were like this:

/access/allow1
/access/deny2
/access/deny3

This is why 'deny' elements are not processed correctly if any 'allow' elements exist in the EML. I split it into two separate counters.

2. When the while loop encounters a public map, it still needs to increment the counter before continuing to the next loop iteration. This is changed in the patch. To be conservative, I left it so it still doesn't iterate when encountering an empty map, but I haven't totally thought it through.

3. This one doesn't seem to make a difference, but the path constructed and passed to nextStep.setPageData does not have a trailing slash in the allow case, but does in the deny case. For consistency, I changed the latter to be like the former, but perhaps the reverse is preferable?

#2 Updated by ben leinfelder about 9 years ago

Jim - these changes are very good.
I tracked down where the empty maps were being added to the access rule lists (your point #2) and made sure that's not happening, at least for normal/valid eml.
You are correct with comment #3, it should not have a trailing slash as the AccessPage appends this when using the passed in xPath.

#3 Updated by ben leinfelder about 9 years ago

now in the SVN trunk

#4 Updated by Redmine Admin over 6 years ago

Original Bugzilla ID was 5058

Also available in: Atom PDF