Query GitLab v4 API endpoints with curl and jq; handles record pagination

gitlapi

#!/usr/bin/env bash
##
##  Query a GitLab v4 API endpoint, with pagination
##
##  Author:    Kevin Ernst <ernstki -at- mail.uc.edu>
##  License:   WTFPL
##  Date:      22 May 2024
##  Requires:  jq (https://github.com/jqlang/jq)
##  Homepage:  https://gist.github.com/ernstki/3707675c8a4ddb06d128154947c49e29
##
ME=$(basename "${BASH_SOURCE[0]}")

_urlencode() (
    # Author:   Chris Down (https://gist.github.com/cdown/1163649)
    #           with to support multiple arguments by me
    # License:  Unknown
    LC_COLLATE=C
    while (( $# )); do
        for (( i = 0; i < ${#1}; i++ )); do
            c=${1:i:1}
            case $c in
                [a-zA-Z0-9.~_-]) printf "$c" ;;
                *)               printf '%%%02X' "'$c" ;;
            esac
        done
        if (( $# > 1 )); then printf '+'; fi  # delimit separate args w/ +'s
        shift
    done
)

_gitlab_api() (
    # set TRACE=1 in the environment to enable execution tracing
    if (( TRACE )); then set -x; fi
    set -u
    : ${GITLAB_URL:?Please define GITLAB_URL as the base URL for your GitLab instance}
    : ${GITLAB_TOKEN:?Please define GITLAB_TOKEN with your private token for the GitLab API}
    local api=$GITLAB_URL/api/v4
    local endpoint=/search
    local searchterms=()
    local curlargs=(
        --silent
        --header "Authorization: Bearer $GITLAB_TOKEN"
    )
    local perpage=20
    local queryargs= count= totals= all= pages= wantheader=

    while (( $# )); do
        case $1 in
            -h|--help|--flags|-\?)
                echo "
  $ME - query GitLab v4 API endpoints with pagination

  usage:
    $ME [-h|--help]
    $ME [-c|--count] [-t|--totals] { /endpoint | TERM [TERM…] }
    $ME [-a|--all] [-p|--pages INT] [-pp|--per-page INT]
    ${ME//?/ } { /endpoint | TERM [TERM…] } [&qs_arg1[&qs_arg2…]]

  where:
    -h, --help       shows this help
    -c, --count      just prints the number of results and returns
    -t, --totals     prints HTTP headers for # pages, # per page, total results
    -a, --all        returns all records instead of just the first page
    -p, --pages      limit results to this many pages (default: 1)
    -pp, --per-page  specifies page size (default: $perpage)

   …and other options starting with a dash are passed through to \`curl\`

  examples:
    $ $ME search terms           # do a code search for 'search' and 'terms'
    $ $ME --all '\"exact phrase\"  # search for an exact phrase, all results
    $ $ME -I '\"search phrase\"'   # see HTTP headers for the above
    $ $ME --count /projects      # count how many projects

  homepage:
    https://gist.github.com/ernstki/3707675c8a4ddb06d128154947c49e29
"
                return
                ;;
            -c|--count)
                count=1
                ;;
            -t|--totals)
                totals=1
                ;;
            -a|--all)
                all=1
                ;;
            -p|--pages)
                shift
                pages=$1
                ;;
            -pp|--pp|--per-page)
                shift
                perpage=$1
                ;;
            -*)
                # FIXME: should probably _only_ accept -I / --head
                if [[ $1 =~ -(i|-include) ]]; then
                    # because it intersperses headers into JSON output which
                    # `jq` can't handle
                    echo 'Ignoring unsupported `-i` / `--include` curl option.' >&2
                else
                    curlargs+=("$1")
                fi
                ;;
            /*)
                endpoint=$1
                ;;
            \&*)
                queryargs+=$1
                ;;
            *)
                searchterms+=("$1")
                ;;
        esac
        shift
    done

    if [[ ${#searchterms[*]} -gt 0 ]]; then
        if [[ $endpoint != /search ]]; then
            echo 'ERROR: Bare search terms only accepted for `/search` endpoint.' >&2
            return 1
        fi
        # otherwise
        queryargs+="&scope=blobs&search=$(set +x; _urlencode "${searchterms[@]}")"
        (( ${TRACE:-} )) && declare -p queryargs
    fi

    queryargs+="&per_page=$perpage"

    if (( all && pages )); then
        echo 'ERROR: The `--all` and `--pages` options are mutually-exclusive.' >&2
        return 1
    fi

    if (( count )); then
        # HTTP headers end with CR+LF, so make sure to get _only_ the digits
        curl --head "${curlargs[@]}" "$api$endpoint?$queryargs" \
          | sed -n 's/X-Total: \([[:digit:]][[:digit:]]*\).*/\1/p'
        return

    elif (( totals )); then
        # print summary of results using HTTP request headers
        curl --head "${curlargs[@]}" "$api$endpoint?$queryargs" \
          | sed -nE '/X-(Page|Per-Page|Total|Total-Pages):/p'
        return

    elif (( all )); then
        # get the total number of pages
        pages=$(
            curl --head "${curlargs[@]}" "$api$endpoint?$queryargs" \
              | sed -n 's/X-Total-Pages: \([[:digit:]][[:digit:]]*\).*/\1/p'
        )
        if ! [[ $pages =~ [[:digit:]]* ]]; then
            echo "ERROR: Problem fetching total pages; try TRACE=1." >&2
            return 1
        fi

    else
        if (( !pages )); then pages=1; fi
    fi

    if [[ "${curlargs[*]}" =~ -(I|-head[^e]) ]]; then
        # only need first page of results; don't pipe through `jq` because
        # it'll be confused by the HTTP headers
        curl "${curlargs[@]}" "$api$endpoint?$queryargs"
    else
        # the first unwraps each results array, the second combines all results
        # back into an array
        for (( p=1; p<=pages; p++ )); do
            if (( pages > 1 )); then echo "Fetching page ${p} of results…" >&2; fi
            # the `[^e]` is so the regexp doesn't match `--header` (a default from above)
            curl "${curlargs[@]}" "$api$endpoint?$queryargs&page=$p" | jq '.[]'
        done | jq --slurp .
    fi
)


# https://stackoverflow.com/a/28776166/785213
# works because you can't `return` from a script
(return 0 2>/dev/null) && sourced=1 || sourced=0

if (( !sourced )); then
    _gitlab_api "$@"
fi

Borne of the need to return all the results from a GitLab code search, and run some simple summary stats on those results. This seems like something jq or a competing utility would do, but searching the InterWebs, I turned up empty-handed. The usual suggestions were just use a for loop in shell script.

Probably Xidel has some kind of support for pagination, but the way Xidel works is sometimes difficult to reason about. Shell script, though, I can do.

Tested on macOS, with the *BSD version of sed, but I don't think I've done anything there that won't work on Linux. Feedback welcome.

Installation

Make sure you have jq available in your search path.

Create a personal access token in your GitLab settings with the read_api scope.

Then:

GIST=https://gist.github.com/ernstki/3707675c8a4ddb06d128154947c49e29
mkdir -p ~/bin
( curl -L $GIST/raw || wget -O - $GIST/raw ) > ~/bin/gitlapi
chmod a+x ~/bin/gitlapi

# define these in your login scripts, or the current shell session
export GITLAB_URL=https://url.to.your/gitlab
export GITLAB_TOKEN='personal access token for read-only API'

# make sure it works
gitlapi --help

Your ~/bin is typically already in your $PATH for most modern Unixes. You may need to log out and back in if your ~/.profile or similar checks for the existence of ~/bin on login, though.

Examples

$ gitlapi search terms            # do a code search for 'search' and 'terms'
$ gitlapi --all '"exact phrase"'  # search for an exact phrase, all results
$ gitlapi -I '"search phrase"'    # see HTTP headers for the above
$ gitlapi --count /projects       # count how many projects

Bugs and misfeatures

There is no error handling if you mess up the GITLAB_URL (remember to include e.g. the /gitlab part of the URL if not served from the root) or your GITLAB_TOKEN is wrong. Here's how you can troubleshoot that, though:

TRACE=1 gitlapi -I [other options]

That is curl's -I / --head option. Other curl options like -f / --fail may work, too.

References

• Fetch multiple pages of json with jq - stackoverflow.com
• GitLab REST API - docs.gitlab.com
  • Other pagination headers