top-10 URIs in the apache log in pure bash

top-10-slow.sh

#!/bin/bash
# This is a very crazy 1-liner that only uses builtins
#set -o noglob;declare -a mapper=() count=();exec 3<access.log; while read -a lines -u 3; do [[ ${lines[5]} != '"GET' ]] && continue; page="${lines[6]}"; ix=0;for i in "${mapper[@]}"; do [[ "$page" = "$i" ]] && break; (( ix++ )); done ; if [[ "$page" = "$i" ]]; then (( count[$ix] ++ )); else mapper+=( $page ); count[$ix]=1; fi; done; (( max=-1, winner=-1, show=10 )); while (( show > 0 )); do ix=0;for i in "${count[@]}"; do (( i > max )) && (( max=i, winner=ix )) ;(( ix++ ));done; builtin echo -e "${count[winner]}\t${mapper[$winner]}";(( count[$winner]=-1, max=0, show-- )); done; 
# Here's a formatted nice version

# We need to make sure that bash doesn't do weird substitution
set -o noglob
declare -a mapper=() count=()
# And to avoid cat we do file descriptors
exec 3<access.log
# We read it in as an array with the default IFS of ' '
while read -a lines -u 3; do 
  # We only look at GETs
  [[ ${lines[5]} != '"GET' ]] && continue
  page="${lines[6]}"
  ix=0
  # Now we do a linear search through things we've seen before
  for i in "${mapper[@]}"; do 
    # If we see it that's great, break out
    [[ "$page" = "$i" ]] && break
    (( ix++ ))
  done
  
  # We see if we broke out because we found it, if so, 
  # increment our counter
  if [[ "$page" = "$i" ]]; then
    (( count[$ix] ++ ))
  else 
    # otherwise, append it to the end of our seen pages
    mapper+=( $page )
    count[$ix]=1
  fi
done

# cool now we sort
# we are doing a linear search for the max value
(( winner=-1, max=-1, show=10 )) 
while (( $show > 0 )); do
  ix=0
  for i in "${count[@]}"; do 
    (( i > max )) && (( max=i, winner=ix ))
    (( ix++ ))
  done
  # When we find it we emit the count and the URI to standard out 
  builtin echo -e "${count[winner]}\t${mapper[$winner]}"
  
  # And we blow away our count of it to -1 so we don't hit it again
  # We decrement our counter, reset our max and then do another pass
  (( count[$winner] = -1, max = -1, show-- ))
done

Backstory: My friend sent me this

one of my standard sysadmin interview Qs is "show me the top 10 URIs in this apache access log". that one is (or should be) easy, though.if they whip out python they're out

I thought "python!? I don't even need sort. Hell, I don't need cat"

I was trying to think of an elegant speedup with minimal code. There's trivial things you can do to avoid the multipass sort (keep track of the top 10 and not do 10 passes) but yawn, unimpressed.

You can do weighted swapping during the insertion phase as well, another big yawn.

They both suck

I'm looking for something really innovative. For instance set shows things in alphabetical order, if you could cleverly encode your values to leverage that then you get sort for free.

Lol i know how. Problem is you'll want grep