15/25: Run either awk or pure bash code for slow days

einarjon · einarjon · commit 5acc2fdf9e6e · 2024-03-07T15:05:39.000Z
PUREBASH environment variable can be set in order to run bash (50-100
times slower than awk).
Day 15 optimized to run in under 20 minutes.
diff --git a/15.sh b/15.sh
@@ -1,33 +1,42 @@
 #! /usr/bin/env bash
 A=($(<"${1:-15.txt}"))
-GIVEUP=${2:-10}
-#A=(0 3 6)
-B=(); i=0; l=""
+PUREBASH=${2:-$PUREBASH}
+declare -a B=()
+i=0; l=""
 for a in "${A[@]}"; do B[a]=$((++i)); done
 while [ $i -lt 2020 ]; do
     n=$((i-${l:-$i})); l=${B[n]}; B[n]=$((++i))
 done
 echo "15A: $n"
 
+# eval is slow, but large arrays are slower
+# Split the higher numbers into smaller arrays
 sharded_swap() { # syntax: l=B[$1]; B[$1]=$2
-    local x="B$(($1>>9))[$1&511]"
+    local x="B$(($1>>8))[$(($1&255))]"
     #local x="B$(($1>>6))[$1&63]"
     eval "l=\${$x}; $x=$2"
 }
-SECONDS=0
-trap 'echo "$i: $SECONDS sec $((i/300000))%"' USR1 EXIT
-i=0; for a in "${A[@]}"; do sharded_swap "$a" $((++i)); done; l="";
-while [[ $i -lt 30000000 && $SECONDS -le $GIVEUP ]]; do
-    n=$((i-${l:-$i})); sharded_swap $n $((++i))
-done
-if [ $i = 30000000 ]; then
+
+if [ -n "$PUREBASH" ]; then
+    trap 'echo "Giving up (i=$i, $SECONDS seconds)"; exit 1' TERM INT
+    echo "This could take 15-20 minutes. Ctrl-C or 'kill $$' to stop."
+    declare -a B{1..1000}
+    for N in $(seq 1000000 1000000 30000000 ); do
+        while ((i < N)); do
+            n=$((i-${l:-$i}));
+            if (( n < 2048 )); then
+                l=${B[n]}; B[n]=$((++i))
+            else
+                sharded_swap $n $((++i))
+            fi
+        done
+        echo "$i took $SECONDS seconds" ;
+    done
     echo "15B: $n"
 else
-    # The code above is painfully slow. Switch to awk
-    echo "15B(round $i): $n (Gave up after $SECONDS seconds)"
     printf "%s\n" "${A[@]}" | awk '{B[$0]=++i;}
-END {
-    while (i<30000000) { n=l?(i-l):0; l=B[n]; B[n]=++i; }
-    print "15B(awk):", n;
-}'
+    END {
+        while (i<30000000) { n=l?(i-l):0; l=B[n]; B[n]=++i; }
+        print "15B(awk):", n;
+    }'
 fi
diff --git a/25.sh b/25.sh
@@ -1,12 +1,17 @@
 #! /usr/bin/env bash
 A=($(< "${1:-25.txt}"))
-declare -i k=1 e=1 key1=${A[0]} key2=${A[1]}
-until ((k==key1)); do
-    ((k=k*7 % 20201227, e=e*key2 % 20201227))
-done
-echo "25A: $e"
-echo "${A[@]}" | awk '{key1=$1; key2=$2; k=1; e=1}
-END {
-while (k != key1) { k = k*7 % 20201227; e = e*key2 % 20201227 }
-print "25A(awk):", e
-}'
+PUREBASH=${2:-$PUREBASH}
+if [ -n "$PUREBASH" ]; then
+    trap 'echo "Giving up after $SECONDS seconds"; exit 1' INT TERM
+    echo "This could take a minute. Ctrl-C or 'kill $$' to exit"
+    declare -i k=1 e=1 key1=${A[0]} key2=${A[1]}
+    until ((k==key1)); do
+        ((k=k*7 % 20201227, e=e*key2 % 20201227))
+    done
+    echo "25A: $e"
+else
+    echo "${A[@]}" | awk '{key1=$1; key2=$2; k=1; e=1
+        while (k != key1) { k = k*7 % 20201227; e = e*key2 % 20201227 }
+        print "25A(awk):", e
+    }'
+fi
diff --git a/README.md b/README.md
@@ -4,6 +4,7 @@ https://adventofcode.com/2020/
 
 Input not included, but can be given on the command line.
 Defaults to *number*.txt in the same folder (no leading 0).
+Setting env variable PUREBASH to any value (or giving a second argument) will use slow bash instead of using faster awk in days 15 and 25.
 
 Description of what I'm doing. Contains spoilers....
 
@@ -74,9 +75,9 @@ Description of what I'm doing. Contains spoilers....
 
 ### 15.sh
  1. Store the last known position in a sparse array. Calculate next position until done.
- 2. Finally bash fails. Sparse arrays in bash seem to be linked lists, so accessing/adding an item is O\[Nlog(N)\] or worse.
+ 2. Finally bash fails. Sparse arrays in bash seem to be linked lists, so accessing/adding an item is O\[N\] or worse.
     Gave up after 20 hours getting to item 6.1M, and solved it in 10-12 seconds with awk.
-    Using a sharded-set https://github.com/romkatv/advent-of-code-2019 solves this in about half an hour.
+    Using a sharded-set https://github.com/romkatv/advent-of-code-2019 solves this in about 20 minutes.
 
 ### 16.sh
  1. Find the min/max of the classes, then loop through every ticket to find illegal values.
@@ -98,9 +99,8 @@ Description of what I'm doing. Contains spoilers....
  2. Cheat a bit and use regex{n} to denote repeated matches.
 
 ### 20.sh
- 1. Slice and dice all permutations of the edges into arrays. Grep each tile against all permutations of edges.
-    The ones with only 6 matches are corners. The ones with 7 are edges.
- 2. Tile matching 144 pieces. Nope...
+ 1. Slice and dice all permutations of the edges into arrays. Grep each tile against all permutations of edges. Convert line number to index/rotation. The ones with only 2 matches are corners. The ones with 3 are edges. Use that to assemble the image.
+ 2. Not done yet: Once the picture is assembled, use a multi-line regex to find the monsters. Then count.
 
 ### 21.sh
  Grep, sort, count
@@ -121,5 +121,5 @@ Description of what I'm doing. Contains spoilers....
 
 ### 25.sh
  1. A couple of while loops. Yay.
- 2. I need to complete 17, 20-2 and 24-2 to get this one. Nope.
+ 2. I need to complete 20-2 to get this one. Nope.
 
