Today is an auspicious day for ruby. Exactly 20 years ago today, Matz started work on what was to become ruby. It saw its first public release about three years later, at the end of 1995. Since then many rubies have been cut and polished. So happy 20th birthday to ruby, and a tip of the hat to Matz and all on ruby core who’ve worked hard on it since then.

Hopefully well see another big event today—the release of ruby-2.0. This is a release that’s been years in the making, and it’s looking really good.

I’ve covered the headline features separately, including lots of examples. This post details some of the lesser-known updates in 2.0 that I think will prove to be great changes. There’s a full list in the NEWS file, but these are my favourites:

• The GC is copy-on-write friendly, thanks to Narihiro Nakamura. In the past, forked ruby processes would quickly duplicate their shared memory, because the GC used to modify every object during the mark phase of its mark-and-sweep run. As of 2.0, objects are marked in a separate data structure instead of on the objects themselves, leaving them unchanged and allowing the kernel to share lots of memory. In practice, this means your unicorns will consume less resident memory (although they’ll still appear to have the same resident size because of how RSIZE is reported). Pat Shaughnessy wrote an excellent post detailing how Narihiro’s new GC design works.

• Syck has been removed. The syck/psych yaml gauntlet that we all had to pass through around the 1.9.2 days is completely behind us now, which is great. Ruby now has a hard dependency on libyaml, but it’s bundled for the cases where the library isn’t present locally.

• The vendored rubygems has been upgraded to 2.0.0-preview2, which should be 2.0.0 final by release time. This new rubygems has full support for the new default gems that some vendored libaries have been moved into. Other notable changes include gem search being a remote search by default, and only ri docs being built by default. Good changes all round.

• The default source encoding is UTF-8 now (changed from US-ASCII). This means literal strings are unicode by default in 2.0, and the # coding: utf-8 comment we’ve grown accustomed to adding isn’t required anymore. This is a nice step in the direction of “all unicode all the time”. A bunch of other encoding cleanups were made, too – one example is Time#to_s, which returns a string encoded in US-ASCII instead of BINARY.

• There’s a new #to_h convention to mirror #to_a. There are less hash-like things than array-like things, so it’s not defined in as many places, but one nice addition is Struct#to_h, which returns a hash of the struct’s keys and values: Struct.new(:key).new('value').to_h #=> {:key=>"value"}

• A new %i[] / %I[] array literal was added, which produces an array of symbols, just like %w[] & %W[] produce arrays of strings. I can’t think of many situations where I’d want to use this syntax, but I think it makes sense to have it alongside its string equivalent.

• Zlib supports streaming when deflating and inflating now, which means large files can be processed without using up huge amounts of memory. This is a particularly nice change because MRI doesn’t ever reduce its memory footprint while running – growing the ruby process to allocate large objects is a one-way thing, even after those objects have been garbage collected.

• IO#lines and friends have been deprecated. The list-of-string methods on IO, StringIO, Zlib::GzipReader and so on, like #lines, #chars and #bytes, have been deprecated in favour of #each_line, etc. This is a nice change, pushing the feel of the enumerating API in the direction of the #each / #each_thing convention.