First impressions of git

So far, I like git a lot. My favorite feature is branches, and the goodies that come with that, such as git fetch, which pulls patches from a remote repository into a branch. I like using git add to flag files to commit more than the darcs "cherry picking", which is tedious and error-prone. I also like it how git status shows me any files which have not been added yet.

The gitk GUI, while not great, is a nice way to get an overview of what's going on with the repository.

Of course the best thing about git is the speed. The Linux Kernel developers use git, and they have a huge codebase with many developers. I'm confident that I've made the right choice by switching to git.

One thing I don't like about git is the complexity -- there are hundreds of commands and it certainly seems like the user interface could be simplified, for the common operations anyway. I'm slightly worried that the increased learning curve might be a hindrance to getting contributors in the future. However, having to wait 10 minutes to darcs push is a turn-off to contributing in itself.

A neat trick for refactoring code

So today I'm going to be cleaning up one of the vocabularies involved in code generation in the optimizing compiler. Each vocabulary has a set of words which are used by other vocabularies -- the "public interface", and implementation words, those only used to implement the interface. I want to rewrite the generator.registers vocabulary but keep the public interface as close to its current state as possible. Trouble is, the public interface is not well-defined here; this vocabulary does not use <PRIVATE and PRIVATE>. However, this is not a big problem, since I can just write a tool:

: used-externally ( vocab -- words )
    dup words [
        usage
        [ word? ] subset
        [ word-vocabulary = not ] curry* contains?
    ] curry* subset natural-sort ;

How does this work? It takes the list of words in a vocabulary, then for each word, checks if the usages of that word include any words outside of this vocabulary. If the answer is affirmative, it adds this word to the sequence that is returned.

Using this tool I found out that generator.registers vocabulary only has a handful of words which are used externally, and so it should be easy to do a radical refactoring which doesn't require many changes to other modules.

It is hard to articulate how powerful this is. Not only is the Factor listener is a great way of testing code, it also provides a way to interact with the system on a meta-level. I wonder how much work it would be to implement something like used-externally in a conventional IDE such as Eclipse; I'm sure all the pieces are there (cross-referencing, etc) but it can't be that easy to just glue things together like that.

How to publish a git repository

Now that I'm using git, all the contributors have to switch, too. Chris Double wrote a blog entry about publishing a git repository. Thanks Chris!

The Factor project switches to git

Why git? Because it is faster than darcs, and has kick-ass branching features.

To check out a repo:

git clone git://factorcode.org/git/factor.git/

You can find documentation at the git website.

A gitweb interface is coming really soon; we have to implement cgi support for the Factor HTTP server first.

The code in the repository runs on the following platforms right now; more will be added very soon:

• Mac OS X/PowerPC
• Mac OS X/x86
• Linux/x86
• Linux/AMD64
• FreeBSD/x86

You will need the latest boot images.

The old darcs repository is still there. It contains the last revision of Factor to include an interpreter, and so represents a historical milestone of sorts.

Factor web site redesign

Thanks to Elie Chaftari, we have a new Factor web site design. Check it out before it goes live! (Which will probably happen with the release of Factor 0.91).

The "Latest News" sidebar is just a placeholder; I will replace it with an RSS aggregator soon.

Also if you look carefully, you'll see mention of a git repository. Yes, I'm switching to git! Today I encountered a darcs bug/flaw/limitation while merging in Eduardo's patches (it was not Ed's fault) and decided that this would be the very last time.

Right now the repository contains the in-progress quotation compiler/compiled continuations code, which only runs on Mac OS X/PowerPC, FreeBSD/x86 and (just barely) Mac OS X/x86. When it is more stable and supports a wider range of platforms, I will make an announcement and formally retire the darcs repository.

Contributors: please continue using darcs for the next few days, but be prepared to migrate to git Real Soon Now.

Building the latest Factor sources on Ubuntu Linux

David Mercer writes about getting Factor from darcs and compiling it on Ubuntu. The hardest part is just knowing what packages to install to get gcc, libc headers, and so on.

Bignum libraries

So does anybody know of a simple, efficient, BSD-licensed bignum library where the radix size is equal to the CPU's word size, and not the CPU's word size minus two bits?