upset, genini: different version ordering

[Jon Turney]

On 20/07/2015 19:10, Jon TURNEY wrote:
> On 20/07/2015 19:03, Achim Gratz wrote:
>> I've just found that upset and genini will order versions differently.
>> For perl-Carp, genini produces:
[...]
> If I recall correctly, genini is just broken, doing some kind of lexical
> sort which e.g. sorts 1.10 before 1.9.
Something which seems to be missing from this discussion is how Cygwin 
version numbers are supposed to be sorted.

This needs to be simple, documented (and ideally, similar to other 
package managers), so that package maintainers can easily anticipate how 
their package versions are going to be ordered, but [1] seems silent on 
this subject, apart from describing it as 'intuitive'

Here's a list of existing package versions ordered by upset, which I 
think sort differently in setup:

x86:

makeself 2.1.5+20120813+gitdcbe778-1 < 2.1.5-3
tcl-itcl 3.4b1-1                     < 3.4.1-1
tidy     20090325-1                  < 041206-1
wput     0.6.2+git20130413-2         < 0.6.2-1

x86_64:

bzr      2.6.0+bzr6602-1 < 2.6.0-2 < 2.6b2-1
cgdb     0.6.7+20150214+git3a710f9-1 < 0.6.7-1

The result for tidy is particularly unexpected, and looking at the date 
of the package files, just wrong. (This appears to be due to upset 
truncating numeric parts to 4 digits!)

I don't really want to spend effort on unravelling the complexities of 
the sorting that upset does, since I don't think it's worth keeping, and 
we should switch to a scheme which can be described in a paragraph,
e.g. the scheme used by setup and RPM:

Compare contiguous chunks of digits or non-digits.
Non-digit chunks sort before digit chunks.
Digit chunks sort numerically, non-digit chunks sort lexicographically.
The version with a suffix remaining is the greater

[1] https://cygwin.com/setup.html