Few days ago there was a lot links on the Internet to gitshelve, which implements persistent versioned storage of objects in the git. I’ve read it description and realized that there are few serious design flaws:

• Can store only strings
• Uses subprocess.PIPE for interconnection with git
• Uses bunch of C+Perl+Shell code in Python library instead of using another Python library ;-)

History

We’ve discussed it with Pythy, and I’ve convinced him to write hgshelve, which should be easy and pain-free to implement. First thing which we came to is storage of objects through usage of simplejson, which generates (opposing to pickle) easy-readable string representation of objects. Plus objects, dumped by simplejson, can be edited by hands without fear of occasional corruption caused by newline or space.

While I was theorizing, he just wrote code and got hgshelve working. ;-) And from starts all mentioned flaws of gitshelve was fixed. ;-)

Usage

It’s really simple to use, just as usual shelve:

>>> import hgshelve
>>> data = hgshelve.open('path/to/repo') # repository is opened/created here
>>> data['key'] = {1: "Hello, world!", 'b': 'just test'}
>>> data.sync()
>>> data
{'key': {u'1': u'Hello, world!', u'b': u'just test'}}

Additionally you can use data.commit, which optionally accepts two arguments: commit message and key to commit (if you want to commit single key).

There are no support of advanced mercurial features, like branches¹ and named branches, though it can arise lately. get_data method can return data from another revision through getting optional argument called rev, which can be one of: revision number, partial (or full) revision hash, mercurial tag.

Dependecies

There are only two of them:

• Mercurial
• simplejson

Where to get

Just clone repository:

hg clone http://hg.piranha.org.ua/hgshelve/