2015/06: 40x speedup in part 2, 2x in part 1.

Subshell calls are slow. Collect all the changes for each round and do
a single call to $(tr ...), instead of doing one per line.

From +5 minutes to under 10 seconds for both parts, so it's now faster than awk.

Author: einarjon

2015/06.sh

@@ -11,9 +11,10 @@ while read -r action x y _ X Y; do
             AA[i]=${AA[i]:0:x}${one:x:X-x+1}${AA[i]:X+1}; done;;
         off) for i in "${idx[@]:y:Y-y+1}"; do
             AA[i]=${AA[i]:0:x}${zero:x:X-x+1}${AA[i]:X+1}; done;;
-        toggle) for i in "${idx[@]:y:Y-y+1}"; do # swap using "t" as tmp
-            T=${AA[i]:x:X-x+1};T=${T//0/t}; T=${T//1/0}; T=${T//t/1};
-            AA[i]=${AA[i]:0:x}${T}${AA[i]:X+1}; done;;
+        toggle) PART="";
+            for i in "${idx[@]:y:Y-y+1}"; do PART+="${AA[i]:x:X-x+1} "; done
+            TMP=($(tr '01' '10' <<< "$PART"))
+            for i in "${idx[@]:y:Y-y+1}"; do AA[i]=${AA[i]:0:x}${TMP[i-y]}${AA[i]:X+1}; done;;
     esac
 done < <(sed "s/turn.//" "$input")
 IFS='' ANS=${AA[*]//0}
@@ -23,14 +24,14 @@ if [[ -n ${2:-$PUREBASH} ]]; then
     IFS=$' ,\n'
     AA=($(printf "%.s$zero\n" "${idx[@]}"))
     while read -r action x y _ X Y; do
+        PART=""
+        for i in "${idx[@]:y:$Y-y+1}"; do PART+="${AA[i]:x:X-x+1} "; done
         case $action in
-            on) for i in "${idx[@]:$y:$Y-y+1}"; do  # shift up by 1. '?' means overflow
-                AA[i]=${AA[i]:0:x}$(tr '0-9a-zA-Z' '1-9a-zA-Z?' <<< "${AA[i]:x:X-x+1}")${AA[i]:X+1}; done;;
-            off) for i in "${idx[@]:y:Y-y+1}"; do  # shift down by 1
-                AA[i]=${AA[i]:0:x}$(tr '1-9a-zA-Z' '0-9a-zA-Y' <<< "${AA[i]:x:X-x+1}")${AA[i]:X+1}; done;;
-            toggle)  for i in "${idx[@]:y:Y-y+1}"; do  # shift up by 2
-                AA[i]=${AA[i]:0:x}$(tr '0-9a-zA-Z' '2-9a-zA-Z??' <<< "${AA[i]:x:X-x+1}")${AA[i]:X+1}; done;;
+            on)     TMP=($(tr '0-9a-zA-Z' '1-9a-zA-Z?' <<< "$PART"));;  # shift up by 1. '?' means overflow
+            off)    TMP=($(tr '1-9a-zA-Z' '0-9a-zA-Y' <<< "$PART"));;   # shift down by 1
+            toggle) TMP=($(tr '0-9a-zA-Z' '2-9a-zA-Z??' <<< "$PART"));; # shift up by 2
         esac
+        for i in "${idx[@]:y:Y-y+1}"; do AA[i]=${AA[i]:0:x}${TMP[i-y]}${AA[i]:X+1}; done
     done < <(sed "s/turn.//" "$input")
     #ANS=$(printf "%s" "${AA[@]//0}" | grep -o . | sort | uniq -c | sed "s/^ */+/;s/ /*/")
     ANS=$(printf "%s" "${AA[@]//0}" | sed "s/./+&/g" )

2015/README.md

@@ -29,6 +29,9 @@ This is where bash sucks. The best I could do was an array of strings, and subst
  1. Switch to select what to do. Use substring index magic. Fixed strings of 1000 ones and zeros to make life easier. Swap is a bother, and uses string substitution.
  2. A million countable items in bash are always a bother. Using $(tr ...) allows me to store 62 values per character in a string, but it's extremely slow. Awk is around 30 times faster.
 
+*Update:* By collecting all the changes in an array and doing single $(tr ...) subshell call per input line, part 1 is over 2x faster.
+Part 2 is 40x faster, and outperforms awk.
+
 ### 07.sh
  1. After setting up, loop through all values and evaluate it to a number if all parameters have been evaluated.
 Takes about 100 iterations through all values. Run in a subshell to keep the namespace clean for part 2.