UUID::NCName: Turn UUIDs into NCNames (and back)

require 'uuid-ncname'
require 'uuidtools'

uu = UUIDTools::UUID.random_create
# => #<UUID:0x3fff0e597ef8 UUID:df521e0a-9d57-4f04-9a95-fc2888decc5a>

nc64 = UUID::NCName.to_ncname uu
# => "E31IeCp1X8EqV_CiI3sxaJ"

nc32 = UUID::NCName.to_ncname_32 uu
# => "E35jb4cu5k7yevfp4fcen5tc2j"

orig = UUID::NCName.from_ncname nc64
# => "df521e0a-9d57-4f04-9a95-fc2888decc5a"

orig == UUID::NCName.from_ncname nc32 # => true
orig == uu.to_s                       # => true

# then you can turn it back into an object or whatever
uu == UUIDTools::UUID.parse(orig)     # => true

The purpose of this module is to devise an alternative representation of the UUID which conforms to the constraints of various other identifiers such as NCName, and create an isomorphic mapping between them.

The UUID is a generic identifier which is large enough to be globally unique. This makes it useful as a canonical name for data objects in distributed systems, especially those that cross administrative jurisdictions, such as the World-Wide Web. The representation, however, of the UUID, precludes it from being used in many places where it would be useful to do so.

In particular, there are grammars for many types of identifiers which must not begin with a digit. Others are case-insensitive, or prohibited from containing hyphens (present in both the standard notation and Base64URL), or indeed anything outside of ^[A-Za-z_][0-9A-Za-z_]*$.

The hexadecimal notation of the UUID has a 5/8 chance of beginning with a digit, Base64 has a 5/32 chance, and Base32 has a 3/16 chance. As such, the identifier must be modified in such a way as to guarantee beginning with an alphabetic letter (or underscore _, but some grammars even prohibit that, so we omit it as well).

While it is conceivable to simply add a padding character, there are a few considerations which make it more appealing to derive the initial character from the content of the UUID itself:

• UUIDs are large (128-bit) identifiers as it is, and it is undesirable to add meaningless syntax to them if we can avoid doing so.

• 128 bits is an inconvenient number for aligning to both Base32 (130) and Base64 (132), though 120 divides cleanly into 5, 6 and 8.

• The 13th quartet, or higher four bits of the time_hi_and_version_field of the UUID is constant, as it indicates the UUID's version. If we encode this value using the scheme common to both Base64 and Base32, we get values between A and P, with the valid subset between B and F.

Therefore: extract the UUID's version quartet, shift all subsequent data 4 bits to the left, zero-pad to the octet, encode with either base64url or base32, truncate, and finally prepend the encoded version character. Voilà, one token-safe UUID.

The ID production appears to have been constricted, inadvertently or otherwise, from Name in both the XML 1.0 and 1.1 specifications, to NCName by XML Schema Part 2. This removes the colon character : from the grammar. The net effect is that

<foo id="urn:uuid:b07caf81-baae-449d-8a2e-48c0f5fa5538"/>

while being a well-formed ID and valid under DTD validation, is not valid per XML Schema Part 2 or anything that uses it (e.g. Relax NG).

Blank node identifiers in RDF are intended for serialization, to act as a handle so that multiple RDF statements can refer to the same blank node. The RDF abstract syntax specifies that the validity constraints of blank node identifiers be delegated to the concrete syntax specifications. The RDF/XML syntax specification lists the blank node identifier as NCName. However, according to the Turtle spec, this is a valid blank node identifier:

_:42df00ec-30a2-431f-be9e-e3a612b325db

despite an older version listing a production equivalent to the more conservative NCName. NTriples syntax is even more constrained, given as ^[A-Za-z][0-9A-Za-z]*$.

There are only two hard things in computer science: cache invalidation and naming things [and off-by-one errors].

-- Phil Karlton [extension of unknown origin]

Suppose you wanted to create a literate programming system (I do). One of your (my) stipulations is that the symbols get defined in the prose, rather than the code. However, you (I) still want to be able to validate the code's syntax, and potentially even run the code, without having to commit to naming anything. You are (I am) also interested in creating a global map of classes, datatypes and code fragments, which can be operated on and tested in isolation, ported to other languages, or transplanted into the more conventional packages of programs, libraries and frameworks. The Base32 UUID NCName representation should be adequate for placeholder symbols in just about any programming language, save for those which do not permit identifiers as long as 26 characters (which are extremely scarce).

After careful consideration, I have decided to change the UUID-NCName format in a minor yet incompatible way. In particular, I have moved the nybble containing the variant to the very end of the identifier, whereas it previously was mixed into the middle somewhere.

This can be considered an application of Postel's Law, based on the assumption that these identifiers will be generated through other methods, and potentially naïvely. Like the version field, the variant field has a limited acceptable range of values. If, for example, one were to attempt to generate a conforming identifier by simply generating a random Base32 or Base64 string, it will be difficult to ensure that the variant field will indeed conform when the identifier is converted to a standard UUID. By moving the variant field out to the end of the identifier, everything between the version and variant bookends can be generated randomly without any further consideration, like so:

B64_ALPHA = ('A'..'Z').to_a + ('a'..'z').to_a + ('0'..'9').to_a + %w(- _)

def make_cheapo_b64_uuid_ncname
  vals = (1..20).map { rand 64 }               # generate the content
  vals.push(rand(4) + 8)                       # last digit is special
  'E' + vals.map { |v| B64_ALPHA[v] }.join('') # 'E' for UUID v4
end

# voilà:

cheap = make_cheapo_b64_uuid_ncname
# => "EXSVv8ezPbSKWoKOkBNWKL"

# now try changing it to a standard UUID:

UUID::NCName.from_ncname cheap, version: 1
# => "5d256ff1-eccf-46d2-b296-a0a3a404d58a"

Furthermore, since the default behaviour is to align the bits of the last byte to the size of the encoding symbol, and since the variant bits are masked, a compliant RFC4122 UUID will always end with I, J, K, or L, in both Base32 (case-insensitive) and Base64 variants.

Since I have already released this gem prior to this format change, I have added a :version parameter to both to_ncname and from_ncname. This parameter, which controls the compact UUID spec behaviour, defaults to 1, as of the module version 0.2.4.

Unless you have to support identifiers generated from version 0.1.3 or older, you should be running these methods with version: 1.

Generated and deposited in the usual place.

You know how to do this:

$ gem install uuid-ncname

Or, download it off rubygems.org.

Bug reports and pull requests are welcome at the GitHub repository.

©2018 Dorian Taylor

This software is provided under the Apache License, 2.0.