shell memo

inc.sh

#!/bin/bash
# Very slow build step.
sleep 5
jq .+1

memo-demo.sh

#!/bin/bash
time bash -c 'echo 1 | MEMO=inc-memo memo.sh inc.sh' # Slow
time bash -c 'echo 1 | MEMO=inc-memo memo.sh inc.sh' # Fast

memo.sh

#!/bin/bash
# Memoizes a pure command, using git notes.
# $MEMO is the name of the git note.
INPUT=$(git hash-object -w --stdin)
OUTPUT=$(git notes --ref="$MEMO" show "$INPUT" 2>/dev/null) ||
OUTPUT=$(git cat-file blob "$INPUT" | "$@" | git hash-object -w --stdin) &&
git notes --ref="$MEMO" add -f -m "$OUTPUT" "$INPUT" 2>/dev/null
git cat-file blob "$OUTPUT"

gist-pro-gist https://gist.github.com/jcorbin/2b6be1f57f26e6068516e6bcb663f411 thoughts:

• control flow for mortals: there's only so much || and && chaining a body can tolerate ;-)
• your MEMO variable can just be computed by hashing the the command string
• there's several "obivous-ish" applications to trees/directories and alternate command calling conventions
• ...those aspects might probably be especially useful to a project build tool
• ...but maybe not too much actually since you'd want to memoize at appropriate boundaries

The ideas just keep coming:

# TODO support for argv_0 versioning: if the first command arg is a tracked
#      file under the current git repo, then use its sha in place of command
#      name to compute $memo_id