Nothing’s the same anymore
When it became clear that Perl 6 was not going to be just the next version of Perl 5 but its own language, the community began to rumble. There were discussions about whether or not to rename the language as early as the mid-2000s, but today Perl 6 has been released and has had its first major post-release update, 6.d. There are those in the community who want to rename it, and Larry Wall—initial creator of Perl—weighed in on the resolution with the name, “Raku” late last year. Reflecting on this, though, it seems a bit late to rename. “Perl 6” as the name of the language is embedded in the source, dozens of libraries, third party tools, markup/down systems, online forums, books, etc. It’s like RMS’s attempt to re-brand Linux systems as “GNU/Linux” which still has echos today and many users end up wondering why Ubuntu is GNU/Linux but Red Hat is Linux… do they differ that fundamentally? Is RedHat lacking essential GNU tools? Does a Perl 6 syntax plugin for my editor work with Raku? Can I import a Perl 6 module into a Rakudo Raku (or is that Raku Rakudo?) program? I know that those are nonsensical questions, but users will have them for years if not decades regardless of how many blog posts try to explain the (now) history!
I’m frankly skeptical that a declarative name change would take, even if it were a good idea…
And that’s the rub. Is it a good idea? It breaks Perl 6 off as a non-Perl and it’s simply not that. Larry made an initial comment long ago that I think actually solved the problem, but what he said didn’t quite register with everyone and no one followed it to its logical conclusion.
He said that Perl 6 and Perl 5 are both members of the Perl Family of languages (I’m going to capitalize “Perl Family” as the name of the whole collective). This has tremendous implications! Is Perl 7 some as-yet unspecified language? If so, can it be my pet language? How Perl-like does it need to be? Does 7 look more like 5 or 6? What would a major revision to the Perl 5 language that doesn’t stem from Perl 6 be numbered? Is there an expectation that a language in the Perl Family will be initiated by those in the development teams of an existing version?
As you can see, this introduces more questions than it answers, but I don’t think that’s a bad thing. Indeed, most Perl innovations have been so. The goal here is to answer a few of those questions and to set the groundwork for how we can arrive at the next ones.
Larry’s use of the word “family” brings to mind a group of people who are related by no choice of their own, and might not necessarily get along or agree to many, if any conventions. Instead, I’d like to agree that we can view our family as a queue. One should have the expectation that it’s possible to get “pushed” onto this queue electively, in which case you receive the largest index.
But we’ll have Perl 7,000 the day after making such a plan available, so I’m going to propose two parts of a solution. The first is the more conservative and least impactful part that addresses what we have today, though it has plenty of impact, and the second is a bit more speculative and the machinery of it may or may not appeal to everyone.
If we are to accept that “Perl 5” is the name of a language, then it becomes impossible for that language to ever change its major version number. That stunts the language in an unreasonable way, and I think we need to re-align expectations around that reality and re-calibrate versioning to remedy the constraint. As such, I’d like to propose the following versioning scheme across all Perl Family languages in the queue:
Version number: 126.96.36.199 Language Name: Perl 5 Language ID: 5 Major version: 30 Minor version: 0 Patch release: 1
This does not mean that there can’t be a
188.8.131.52.1, but it means that
across all languages in the queue, the above expectation should be valid. The
only constraint on this process being that users have a reasonable expectation
to patch releases not breaking their code due to specification changes while
minor releases should have some sort of deprecation cycle or other way of
ensuring that jumps between minor versions are not only possible, but
So the Language ID is a unique, integer key which identifies a language, ordered in time with respect to all other entries by when the identifiers were issued (which may or may not related directly to the age of the language, see below).
But what does this relate to? Within the queue, all we care about is the specification and the existence of one or more implementations. Implementations and specifications might be the same (Perl 5) or they might be closely related but separate (Perl 6). They could even be as broadly detached as the ANSI C spec and GCC! That’s all up to the language. But the queue needs to know where the spec stands.
Perl 5 has an interesting numbering
It reserves the odd numbered “major” number for development releases. It’s my
feeling, but obviously up to them, that this is a good thing to be
maintained, but should be pushed down one level. So
184.108.40.206 would be a
production release while
220.127.116.11 would be a development release. This
means that bumps to the major number could flag incompatible changes such as
a replacement to a major component such as the object system. Again, that’s
all convention that’s entirely up to the language itself, but it would make
some sense and would give everyone plenty of room to breathe.
Perl 6 has been using non-numeric versions like 6.d. This is also fine. It
would seem reasonable to require that a base-36 numeric sort of version
components is stable with respect to release date (e.g. in Perl 6,
$major.parse-base(36) increases for each subsequent major version component)
but other than that I don’t think that the queue should care. There’s no
requirement that a version use all 36 digits in a base-36 identifier, of
course, and Perl 5’s mostly base-10 history is not a violation of this
The Registration Process
This is a bit more speculative, and I’m well aware that it might not be the most broadly accepted option, but please bear with me until the end and see how you think the whole thing would operate.
The queue is composed of sequential numeric language identifiers. But what kind of number? Obviously, it has a real part… but is there an imaginary part? I think there should be. In fact, I think that solves the major problem with the idea of a Perl Family language queue: the fact that language ideas come and go ephemerally, but languages that we consider “real” are fewer and further between. So, let’s take the point of view of an aspiring language designer. I first need to compose a basic outline of my language’s name, goals, ancestry within the family and major technical features. For Perl 6 this might be the initial announcement and RFC connected to the State of the Onion 2000. It’s not a high bar, and it might be that a language meeting this bar never gets a line of code, much less a production-ready implementation. So it is subjected to a basic amount of scrutiny by the community and then assigned a complex number of the form:
Where n is a monotonically increasing integer value assigned to each new
language proposal and m is the current next unallocated entry in the queue at
the time (e.g. currently 7). Thus, the recent proposal for a Python-like Perl
6 variant might come in
7+1i.0.0.0. It is now free to keep bumping that version as normal and can
remain with that versioning forever if it wishes. However, once my language
has a quasi-stable implementation (let’s say Perl 6 back in the Pugs days or
Perl 1) then I can apply to actually allocate the next available number
without an imaginary component. By that time, perhaps some other language has
matured faster. Perhaps Perl
7+2i has become stable enough that it has been
moved to Perl 7. So my language moves from
8 and officially
becomes Perl 8. Does that mean that Perl 8 is ready for prime time? No,
but it does mean that it’s demonstrated a level of seriousness about being in
the Perl Family that it is ready to begin allocating more prime real estate in
the versioning landscape.
This requires a registrar, but if we can manage CPAN as a shared resource between Perl 5 and 6, I think it’s possible to manage some version number allocation.
It’s also a bit hokey, and that’s not accidental. I strongly believe that there should be a sense that getting one of these imaginary identifiers should feel less than permanent. It shouldn’t be the case that applying to become a fully realized (pun intended) language in the Perl Family feels like an unnecessary extra step. Every imaginary Perl Family language should want to grow up to be a real language!
Okay, that’s enough puns… maybe.
That’s the proposal. Now on to some of my own thoughts as to what this all means…
All in the Family
So, what constitutes a Perl Family member language? That’s hard, but I think it requires a few things:
Its specification cannot be wholly within another language. That is, a simple slang in Perl 6 is not Perl 7. Perl 7 might get an implementation as a slang in Perl 6, but it should stand on its own as well. This is probably the hardest part to get right. It’s a very fuzzy border, but there are definite black and white areas as well, so I think it works as a high-pass filter.
It should have a practical connection to its ancestor languages. If a language is further down the queue than you are (e.g. has a smaller number) then it should be a reasonable expectation that it is composed of features that either emulate, consciously drop, improve on or seek to correct features in that language. For example, one could see an argument that both PHP and Ruby were within the Perl Family in this sense (I’m retroactively deciding to call them
5+2irespectively and invite them to apply for real version numbers at their leisure…), but Java certainly was not. It doesn’t take Perl into account much at all, and is much more clearly a branch from the C/C++ portion of the language space.
It must not be a trivial modification to an existing language. That is, Perl 5, but with Moose built-in is not a langauge specification. It might be the first step in creating one, but it needs to be more than “you don’t have to ‘use Moose’” to meet this criteria.
Arising from the above, I have some questions.
- Would a JSON-like specification for Perl data be allowed to apply for a
number? Should such DSLs get their own namespace (e.g.
Perl DSL 1.x.y.z) or does it just not matter? I’m inclined to the latter.
- In a case like Ruby, where the new language has no real pretense of being a Perl Family member, what happens if someone applies for them?
- Would it make any more or less sense for imaginary languages to all get their real parts bumped whenever a new language ID was allocated? Is this a reasonable level of disruption given that the language is still in early specification stage?
- If we agree to the above, then do imaginary languages exist in a sparse namespace where new specs can claim now-realized languages’ IDs?
What’s In a Name?
There is no requirement that a language treat its Perl Family queue identifier as its “name”. That is, Perl 7 might be more widely known to the world as Gueuze, or it might be more complicated, as in the Perl 6 world where Perl 6 the spec is also known as Raku and there are many implementations current and obsolete, all with their own names.
Being in the Perl Family merely means that the language can be considered fundamentally “Perlish” but that doesn’t mean that anyone ever calls it by its maiden name…