ci/request-reviews/get-reviewers.sh

   1 #!/usr/bin/env bash
   2
   3 # Get the code owners of the files changed by a PR,
   4 # suitable to be consumed by the API endpoint to request reviews:
   5 # https://docs.github.com/en/rest/pulls/review-requests?apiVersion=2022-11-28#request-reviewers-for-a-pull-request
   6
   7 set -euo pipefail
   8
   9 log() {
  10     echo "$@" >&2
  11 }
  12
  13 if (( "$#" < 7 )); then
  14     log "Usage: $0 GIT_REPO OWNERS_FILE BASE_REPO BASE_REF HEAD_REF PR_NUMBER PR_AUTHOR"
  15     exit 1
  16 fi
  17
  18 gitRepo=$1
  19 ownersFile=$2
  20 baseRepo=$3
  21 baseRef=$4
  22 headRef=$5
  23 prNumber=$6
  24 prAuthor=$7
  25
  26 tmp=$(mktemp -d)
  27 trap 'rm -rf "$tmp"' exit
  28
  29 git -C "$gitRepo" diff --name-only --merge-base "$baseRef" "$headRef" > "$tmp/touched-files"
  30 readarray -t touchedFiles < "$tmp/touched-files"
  31 log "This PR touches ${#touchedFiles[@]} files"
  32
  33 # Get the owners file from the base, because we don't want to allow PRs to
  34 # remove code owners to avoid pinging them
  35 git -C "$gitRepo" show "$baseRef":"$ownersFile" > "$tmp"/codeowners
  36
  37 # Associative array with the user as the key for easy de-duplication
  38 # Make sure to always lowercase keys to avoid duplicates with different casings
  39 declare -A users=()
  40
  41 for file in "${touchedFiles[@]}"; do
  42     result=$(codeowners --file "$tmp"/codeowners "$file")
  43
  44     read -r file owners <<< "$result"
  45     if [[ "$owners" == "(unowned)" ]]; then
  46         log "File $file is unowned"
  47         continue
  48     fi
  49     log "File $file is owned by $owners"
  50
  51     # Split up multiple owners, separated by arbitrary amounts of spaces
  52     IFS=" " read -r -a entries <<< "$owners"
  53
  54     for entry in "${entries[@]}"; do
  55         # GitHub technically also supports Emails as code owners,
  56         # but we can't easily support that, so let's not
  57         if [[ ! "$entry" =~ @(.*) ]]; then
  58             warn -e "\e[33mCodeowner \"$entry\" for file $file is not valid: Must start with \"@\"\e[0m" >&2
  59             # Don't fail, because the PR for which this script runs can't fix it,
  60             # it has to be fixed in the base branch
  61             continue
  62         fi
  63         # The first regex match is everything after the @
  64         entry=${BASH_REMATCH[1]}
  65
  66         if [[ "$entry" =~ (.*)/(.*) ]]; then
  67             # Teams look like $org/$team
  68             org=${BASH_REMATCH[1]}
  69             team=${BASH_REMATCH[2]}
  70
  71             # Instead of requesting a review from the team itself,
  72             # we request reviews from the individual users.
  73             # This is because once somebody from a team reviewed the PR,
  74             # the API doesn't expose that the team was already requested for a review,
  75             # so we wouldn't be able to avoid rerequesting reviews
  76             # without saving some some extra state somewhere
  77
  78             # We could also consider implementing a more advanced heuristic
  79             # in the future that e.g. only pings one team member,
  80             # but escalates to somebody else if that member doesn't respond in time.
  81             gh api \
  82                 --cache=1h \
  83                 -H "Accept: application/vnd.github+json" \
  84                 -H "X-GitHub-Api-Version: 2022-11-28" \
  85                 "/orgs/$org/teams/$team/members" \
  86                 --jq '.[].login' > "$tmp/team-members"
  87             readarray -t members < "$tmp/team-members"
  88             log "Team $entry has these members: ${members[*]}"
  89
  90             for user in "${members[@]}"; do
  91                 users[${user,,}]=
  92             done
  93         else
  94             # Everything else is a user
  95             users[${entry,,}]=
  96         fi
  97     done
  98
  99 done
 100
 101 # Cannot request a review from the author
 102 if [[ -v users[${prAuthor,,}] ]]; then
 103     log "One or more files are owned by the PR author, ignoring"
 104     unset 'users[${prAuthor,,}]'
 105 fi
 106
 107 gh api \
 108     -H "Accept: application/vnd.github+json" \
 109     -H "X-GitHub-Api-Version: 2022-11-28" \
 110     "/repos/$baseRepo/pulls/$prNumber/reviews" \
 111     --jq '.[].user.login' > "$tmp/already-reviewed-by"
 112
 113 # And we don't want to rerequest reviews from people who already reviewed
 114 while read -r user; do
 115     if [[ -v users[${user,,}] ]]; then
 116         log "User $user is a code owner but has already left a review, ignoring"
 117         unset 'users[${user,,}]'
 118     fi
 119 done < "$tmp/already-reviewed-by"
 120
 121 # Turn it into a JSON for the GitHub API call to request PR reviewers
 122 jq -n \
 123     --arg users "${!users[*]}" \
 124     '{
 125       reviewers: $users | split(" "),
 126     }'