LegoHypothesis

Overview

The LegoHypothesis is that software can be put together like Lego, out of lots of small interchangeable components.

We are attempting to conclusively disprove this hypothesis, by showing that component sizes follows a PowerLaw, and thus that there is no natural scale to software componentry. Bigger systems will have larger components than smaller ones.

This TR demonstrates this for OO Programs

• http://www.mcs.vuw.ac.nz/comp/Publications/CS-TR-02-30.abs.html

Now actually published: yay!

• http://portal.acm.org/citation.cfm?id=1060716

We now wish to generalise it (particular this component size data) to lots of other systems in [paper writing] by our [pool cleaner].

See ShapeOfJavaSoftware for the latest developments with UoA.

Lego Research Notes

• Existing Publications
  • LegoPowerLawPapers
  • LegoLandClaims
• [fast cash]
• Patterns to Look For
  • LegoPossibleDistributions
  • PowerLaw
  • LogNormal
• Interim Results
  • LegoInterimResults
  • LegoJavaCorpus
• Results Summary
  • LegoResultsSummary
  • EntropyMaximisation

[karaoke equipment]

• TheNetworkHypothesis

The main work will probably consist of

• finding softare systems (source or binary)
• measuring the sizes of components used in that system and/or the incoming references to those components
  • I think I'm generally more interested just in size than in incoming refs for this project, partly because component size measures will be easier across a wide range of systems - JamesKjx
• determining what distribution they follow - see DetectingPowerLaws.
• for more discussion, see the LegoHypothesisFanClub

we should also

• find (heavily cite) publications claiming software is like lego.
  • e.g. that cover of CACM
  • what else?
  • RoBt sez: Somewhere here, or in my office, is a bunch of stuff I found when considering this earlier in the year. As I recall, the main people I found making what seemed to be the claim were Brad Cox (in the many "Softare IC" pitches), and intro OO textbooks. I'll start looking for the refs.
  • Let's record stuff we find in LegoLand: oh no - we already have a page called that!
  • So let's use LegoLandClaims instead.
• LegoPowerLawPapers - papers about the lego hypothesis and power laws

questions to answer

• what is a component -- dunno?
• source or binary or running -- dunno?
• (Ok, so what I mean here is: what can we find? it may be different for different systems?)

• Things that could be measured: LegoPossibleDistributions

software to analyse

• TeX
• emacs
• linux
• unix

Here's a page where we can summarise results in one place: LegoResultsSummary

A program to read in a list of values and produce various outputs: PowerStats (Obsolete, use Marcus's script)

TheNetworkHypothesis: A working hypothesis that says that the distributions that are from network-related things are power laws; other stuff tends to be LogNormal.

A note about how log-normals could look like power laws: WikiOfJerome:PowerLawLogNormal

EntropyMaximisation

---

New stuff Nick is working on:

Component size dependence on system size: ComponentSystemSizeCorrelation

InterComponentGraph

ClusteringCoefficient: a metric describing the amount of clustering among vertices in a graph.

---

So, I've been reading many papers recently about WebServices?. I wonder what the failure of the LegoHypothesis would say about their success? --EwanTempero

---

I was the architect of a commercial software tool called AppComposer that was successful at building large applications out of software components, but only if you slightly change the definition of what a component is. I agree that you cannot build software out of reusable objects, and unfortunately when most people say "component" they are really saying "object". I plan to give a talk about this, probably sometime in February. --WmLeler