[alephnil]

It is not completely arbitrary though. Usually 1.0 for proprietary software means "the first version shipped to customers", while for open source, the very first commit is 0.0, and then it goes from there. Since it is public from the start there is no 1.0 moment. Sometimes, especially for programming languages, the 1.0 version is used to signal that from no on there will be no arbitrary change to syntax or semantics after that, so that it can be relied on.

Traditionally, version numbering has been used to signal the significance of the release. for version x.y.z, you could expect that

  * x is incremented: Major new features, possibly incompatibilities
  * y incremented but not x: Minor new features and bug fixes.
  * z changed only: Bug fixes only.

This was generally observed in both proprietary and open source software alike, and is still used in many projects. Recently many projects has abandoned this pattern, including Chrome and Firefox, the Linux kernel and others.

Of cause there has always been a pressure from the marketing departments to have a new major release, while the engineers has been holding back, so you have always seen major releases that isn't that major, and sometimes incompatible changes sneaked into minor releases. The latter has generally been considered bad form.