Not logged in
Log in now
Create an account
Subscribe to LWN
LWN.net Weekly Edition for May 23, 2013
An "enum" for Python 3
An unexpected perf feature
LWN.net Weekly Edition for May 16, 2013
A look at the PyPy 2.0 release
A conversation with Linus at LinuxCon Japan
Posted Jun 2, 2011 15:28 UTC (Thu) by viiru (subscriber, #53129)
Posted Jun 2, 2011 16:50 UTC (Thu) by Yorick (subscriber, #19241)
Posted Jun 2, 2011 19:54 UTC (Thu) by jimparis (subscriber, #38647)
Posted Jun 3, 2011 0:01 UTC (Fri) by aegl (subscriber, #37581)
That way you have a clear reason why you want to start the 4.x series - either because there is not much 3.0 left, or because the additions dwarf the original.
Here's an out-of-date graph that shows how total code size grew from 2.6.12 to 2.6.37 ... while code from each release was gradually nibbled away by later releases:
Posted Jun 3, 2011 11:41 UTC (Fri) by jengelh (subscriber, #33263)
Posted Jun 3, 2011 22:28 UTC (Fri) by giraffedata (subscriber, #1954)
But I wish Linus would have given some insight as to what the number 20 is too big for. My own mind has no trouble grasping 21 (or 321) releases of something.
To avoid that, Linus is just inventing a new numbering system rather than use the decimal system that people know so well. (I.e. he plans 3.0, 3.1, 3.2, ... 3.20, 4.0, 4.1, etc. whereas the decimal 3, 4, 5, ... 23, 24, 25 makes more sense to me).
Posted Jun 3, 2011 23:05 UTC (Fri) by elanthis (guest, #6227)
20 isn't "too big" for anything, it's just an arbitrary number in an arbitrary range chosen as a reason to increase another arbitrary number.
I'd have preferred the calendar-based numbering (like Ubuntu's) simply because then the numbers mean something logically to lay users and folks who want version numbers to imply _something_ without actually implying anything that Linus doesn't want the version number to do (that is, the 2012.04 release doesn't indicate some major new feature or change from the 2011.10 release, it's just the next version and it happens to be released in 2012 rather than 2011).
Posted Jun 4, 2011 2:26 UTC (Sat) by giraffedata (subscriber, #1954)
Version numbers aren't decimal numbers. A lot of Americans especially
tend to forget that, due to our use of the dot
Sorry, I didn't notice that opportunity for confusion. I didn't mean
3.21 is a decimal number because it has a decimal point in it. I meant 21 is a decimal number. Linus plans to use an x.y number scheme, with the y part resetting and the x part incrementing at arbitrary points instead of using decimal, where the cipher is a string of digits, each turning over at 10.
20 isn't "too big" for anything, it's just an arbitrary number in an
arbitrary range chosen as a reason to increase another arbitrary number.
But those are Linus' word. The reason to increase the first number from 3 to 4 is that (approximately) 20 is "too big."
Posted Jun 7, 2011 10:21 UTC (Tue) by marcH (subscriber, #57642)
Interesting... Is this why dots became popular in software version numbers, I mean because they make numbers more or less look like fractions?
> (In many/most other parts of the world, "1,000.25" is written "1.000,25".)
"1 000,25" is also quite common:
periods in release numbers
Posted Jun 8, 2011 1:58 UTC (Wed) by giraffedata (subscriber, #1954)
I don't see how it can be anything else. Early software products had plain
natural numbers: 1, 2, 3, 4. OS/360 got up to the 30s or 40s that way, with
several releases a year. Later, the bureaucracy surrounding putting out a new
release got so heavy that people wanted a way to improve Release 31 without
actually putting out Release 32, so the natural thing was to make it
fractional. It was probably a while before they had more than 9 of these
subreleases, and then there was probably some dilemma about calling it 31.10,
but I guess we got over that. And eventually, as people put more and more
stuff into these cheap subreleases, the bureacracy grew up around those too,
and so we added another fraction: 31.10.1 etc.
If they weren't thinking of fractions, it would have been more normal to call
it "Release 31, Subrelease 10, Modification 1" and abbreviate it "31-10-1".
Posted Jun 7, 2011 15:44 UTC (Tue) by marcH (subscriber, #57642)
It is too big for the human brain
Posted Jun 8, 2011 14:46 UTC (Wed) by nye (guest, #51576)
> It is too big for the human brain
That's... not relevant in any way. The number 20 is still one number, with two digits. If the discussion were over version 17456295702 or something else with a lot of digits to remember, *then* it would be relevant.
(PS. The idea in question is, of course, utter bollocks. For supporting evidence, see the 3 billion people who happily memorise lots of 11-digit telephone numbers, for example)
magic number 7
Posted Jun 8, 2011 15:44 UTC (Wed) by giraffedata (subscriber, #1954)
I think you missed the point. Few of those people could repeat a random 11 digit number you say slowly back to you. The 7 limit is working memory -- where things stay in your brain before they are memorized. Also, many people could repeat back an 11 digit telephone number because it contains multi-digit chunks which are a single piece of information. For example, a North American area code is 3 digits, but in most cases, that would count 1 toward the limit.
Incidentally, as the Wikipedia article admits, current thinking is that the real number is less than the 7 +/- 2 from the 1950s research.
If there were some reason that people needed to think about some feature of all minor releases of a certain major release of the kernel at once, then 20 would be way too many. But I think marcH was just making a light-hearted comment about a number being intrinsically too big. It's pretty clear that the only thing 20 is too big for is some aesthetic feeling Linus has about normal looking release numbers.
Posted Jun 8, 2011 23:04 UTC (Wed) by neilbrown (subscriber, #359)
The difference between 38 and 39 is simply a lot smaller than the difference between 8 and 9 (is some cognitive sense which I believe psychologist can measure) and I have trouble remembering which of those two is 'next'.
It felt really good when 2.6.39 was done because then the "next" version would be 2.6.40, and the difference between 40 and 39 is MUCH bigger than the difference between 39 and 38. And then 2.6.42 wasn't far away and the difference between 42 and any other number is clearly very big too. So we were coming to a time when I would not be confused about version numbers for at least a year or so.
So it seemed like a strange time to change numberings, just when the up coming numbers would be so much easier to work with. But I definitely understand how numbers can be "too big" - it is when their differentiation becomes too small.
Posted Jun 9, 2011 11:15 UTC (Thu) by marcH (subscriber, #57642)
> 3 billion people who happily memorise lots of 11-digit telephone numbers,
> utter bollocks
Posted Jun 2, 2011 18:09 UTC (Thu) by iabervon (subscriber, #722)
Posted Jun 3, 2011 0:09 UTC (Fri) by nicooo (guest, #69134)
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds