Why Properties?

Just as Subversion uses properties to store extra information about the files, directories, and revisions that it contains, you might also find properties to be of similar use. You might find it useful to have a place close to your versioned data to hang custom metadata about that data.

Say you wish to design a web site that houses many digital photos and displays them with captions and a datestamp. Now, your set of photos is constantly changing, so you'd like to have as much of this site automated as possible. These photos can be quite large, so as is common with sites of this nature, you want to provide smaller thumbnail images to your site visitors.

Now, you can get this functionality using traditional files. That is, you can have your image123.jpg and an image123-thumbnail.jpg side by side in a directory. Or if you want to keep the filenames the same, you might have your thumbnails in a different directory, such as thumbnails/image123.jpg. You can also store your captions and datestamps in a similar fashion, again separated from the original image file. But the problem here is that your collection of files multiplies with each new photo added to the site.

Now consider the same web site deployed in a way that makes use of Subversion's file properties. Imagine having a single image file, image123.jpg, with properties set on that file that are named caption, datestamp, and even thumbnail. Now your working copy directory looks much more manageable—in fact, it looks to the casual browser like there are nothing but image files in it. But your automation scripts know better. They know that they can use svn (or better yet, they can use the Subversion language bindings—see the section called “Using the APIs”) to dig out the extra information that your site needs to display without having to read an index file or play path manipulation games.

Custom revision properties are also frequently used. One common such use is a property whose value contains an issue tracker ID with which the revision is associated, perhaps because the change made in that revision fixes a bug filed in the tracker issue with that ID. Other uses include hanging more friendly names on the revision—it might be hard to remember that revision 1935 was a fully tested revision. But if there's, say, a test-results property on that revision with the value all passing, that's meaningful information to have.

Searchability (or, Why Not Properties)

For all their utility, Subversion properties—or, more accurately, the available interfaces to them—have a major shortcoming: while it is a simple matter to set a custom property, finding that property later is whole different ball of wax.

Trying to locate a custom revision property generally involves performing a linear walk across all the revisions of the repository, asking of each revision, "Do you have the property I'm looking for?" Trying to find a custom versioned property is painful, too, and often involves a recursive svn propget across an entire working copy. In your situation, that might not be as bad as a linear walk across all revisions. But it certainly leaves much to be desired in terms of both performance and likelihood of success, especially if the scope of your search would require a working copy from the root of your repository.

For this reason, you might choose—especially in the revision property use case—to simply add your metadata to the revision's log message using some policy-driven (and perhaps programmatically enforced) formatting that is designed to be quickly parsed from the output of svn log. It is quite common to see the following in Subversion log messages:

Issue(s): IZ2376, IZ1919
Reviewed by:  sally

This fixes a nasty segfault in the wort frabbing process
…

But here again lies some misfortune. Subversion doesn't yet provide a log message templating mechanism, which would go a long way toward helping users be consistent with the formatting of their log-embedded revision metadata.

Prev Up Next
Properties Home Manipulating Properties