Re: Package Pool Proposal

[James Deikun <james@netcsi.com>]

>I was thinking, what if we just special case'd lib and maybe a few others
>and still used prefixes but had a set of prefixes like
>[a-z,lib[a-z],x[a-z],g[a-z]] That gives a much smaller top level dir and
>still manages to create a nice logical distribution with good hash
>properties. We can create new buckets later on if needed.

I tried a couple of schemas myself.  Note that the following distributions
do not contain non-us.

remove 'lib', use second letter:
    400 e
    340 o
    334 a
    316 i
    240 m
    211 t
    211 c
    209 p
    200 l
    198 s

remove 'lib', 'x', 'g', or 'k', use first remaining letter:
    421 p
    382 s
    289 c
    280 t
    277 m
    256 d
    234 n
    228 f
    228 a
    214 l

remove 'lib', 'x', 'g', 'k', or 'p', use first remaining letter:
    390 s
    310 c
    283 m
    282 t
    260 d
    246 a
    235 n
    231 f
    225 l
    225 e

remove 'lib', use second letter, or third if second is a vowel (aeiou):
    363 m
    357 t
    353 n
    343 s
    339 r
    337 c
    283 l
    259 p
    204 d
    176 g

note that all these would lump libs together if the 'lib' prefix isn't
dropped.  (under 'i', 'l', 'l', 'b' resp.)   the total number of packages
is 4335, which means an ideal size distribution among 26 letters would have
166 packages in each bucket.

i think basing the directory on the second letter is better because it
allows better/quicker/more specific tab-completion in each directory, as
well as automatically breaking up single-letter prefixes and separating
perl/python/php/pcmcia into their own directories.  But that's just me.

>I think going by last letter is not a good idea because it is hard to
>predict what a pacakge name will end in, consider zlib[2,2g,g2] I can
>never remember what it ends in!

A very good point, I would have this problem myself, and it makes looking
for the latest version of a shared lib hell, too.

--
James Deikun (with season tickets in the peanut gallery)