And I was tired of maintaining bibliographic databases for my research group, where each database duplicates a fair fraction of the references with slight variations. The same conference would sometimes have an "Eighth" in front of it, sometimes it would say "Proceedings of," sometimes "Proc. of," sometimes nothing at all. Then our paper deadline would arrive, we would be a page over the limit, and we would have to edit all the entries down to cryptic shorthand, only to expand them later for the camera-ready. Even with heavy use of @string macros, this was a tedious procedure. Changing style files required learning an arcane language. Databases were not modular.

It was 2006, and we knew better, and I thought it would be worth it to spend a few days writing a modern replacement for BibTex.

If you are using a BibTex style file, you are probably using some goobledygook that someone kludged together without understanding, and chances are that no mortal will be able to translate it into anything reasonable. But you can probably write something from scratch in half an hour that'll do what you want in CrossTeX. CrossTeX's internal programming language is Python, a nice and clean platform that is easy to learn.

• Python 2.0 or better, which runs on almost every platform.
• Ply 2.1 or better, which also runs on almost every platform.

The long answer is: you may be a genius and you may even have OCD, but it's too easy to make mistakes when using bibtex. Bibtex provides no support for avoiding common errors such as inconsistent capitalization in titles, spelling errors in author names, and errors in conference locations. Its input language actually encourages errors of this kind. People usually get by simply because they do not cite that many papers and because they have accepted that they'll need to waste their time checking computer output. Things can be drastically better.

• The entire data model is object oriented, as opposed to a special magic hack that applies only to a single field per object. One can inherit any number of attributes from any number of objects. All object treatment is uniform and consistent.
• The use of the crossref field in BibTeX requires that a @proceeding object be declared for each instance of a conference. For long-runing conferences, you would need dozens of @proceeding objects, with extensively repeated information. Conditional fields in CrossTeX enable a single object definition to cleanly subsume all of that information.
• The implementation of the crossref field in CrossTeX is simply as an optional field whose value is never used in formatting. Because of the simple inheritance model in CrossTeX, any object value assigned to that (or any) field will provide values for inheritance. Thus, you may still use the crossref field to inherit information in CrossTeX without actually using another object's value.
• CrossTeX generates professionally-formatted web pages. Default bibtex has no support for the web, and other tools that convert to XML and back to HTML produce strange-looking citations.
• CrossTeX provides fine-grain control over the formatting of the resulting references, without forcing the user to write any code.

• You can cite any paper in the CrossTeX database using its DBLP key. Often, this is the key you see when googling the paper. These keys are almost impossible to remember for multiauthor papers though, so there are two other ways to identify the paper you had in mind.
• You can introduce your own aliases for papers in the database, or even change their fields if you like. So instead of citing DBLP:NSDI:2004:WalshSirer, you can introduce Credence as an alias in a central file and simply cite Credence.
• You can use constrained citations to cite papers by specifying some constraints about them. For instance, if you remember that the paper appeared at NSDI'04 and had Sirer as a co-author, you can cite !Sirer:NSDI:2004.