make depot if not restored (#91)

* Pass cache-name between save/restore

* Use save job name in cache-name

* Re-order test jobs

* Fix typo in registry warning

.github/workflows/CI.yml

@@ -20,115 +20,177 @@ permissions:
   contents: read
 
 jobs:
-  generate-key:
+  generate-prefix:
     runs-on: ubuntu-latest
     outputs:
-      cache-name: ${{ steps.name.outputs.cache-name }}
+      cache-prefix: ${{ steps.name.outputs.cache-prefix }}
     steps:
-    - name: Generate random file
-      shell: 'julia --color=yes {0}'
-      run: 'write("random.txt", string(rand(10)))'
-    - name: Set cache-name as output
+    - name: Generate random cache-prefix
       id: name
-      run: echo "cache-name=${{ hashFiles('random.txt') }}" >> $GITHUB_OUTPUT
+      run: |
+        cache_prefix=$(head -n 100 </dev/urandom | shasum -a 256 | cut -d ' ' -f 1)
+        echo "cache-prefix=$cache_prefix" >>"$GITHUB_OUTPUT"
 
   test-save:
-    needs: generate-key
+    needs: generate-prefix
+    runs-on: ${{ matrix.os }}
+    outputs:
+      cache-name: ${{ steps.cache-name.outputs.cache-name }}
     strategy:
       matrix:
-        dep:
-          - name: pandoc_jll
-            version: "3"
+        nested:
+          - name: matrix
+            invalid-chars: ","  # Use invalid characters in job matrix to ensure we escape them
+        version:
+          - "1.0"
+          - "1"
+          - "nightly"
         os:
           - ubuntu-latest
           - windows-latest
           - macOS-latest
+        exclude:
+          # Test Julia "1.0" on Linux only
+          - version: "1.0"
+            os: windows-latest
+          - version: "1.0"
+            os: macOS-latest
       fail-fast: false
-    runs-on: ${{ matrix.os }}
     env:
       JULIA_DEPOT_PATH: /tmp/julia-depot
     steps:
     - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+    - name: Set cache-name
+      id: cache-name
+      shell: bash
+      run: |
+        echo "cache-name=${{ needs.generate-prefix.outputs.cache-prefix }}-${{ github.job }}" >>"$GITHUB_OUTPUT"
+    - uses: julia-actions/setup-julia@v1
+      with:
+        version: ${{ matrix.version }}
     - name: Save cache
       id: cache
       uses: ./
       with:
-        cache-name: ${{ needs.generate-key.outputs.cache-name }}
+        cache-name: ${{ steps.cache-name.outputs.cache-name }}
+        delete-old-caches: required
     - name: Check no artifacts dir
-      shell: 'julia --color=yes {0}'
+      shell: julia --color=yes {0}
       run: |
         dir = joinpath(first(DEPOT_PATH), "artifacts")
         @assert !isdir(dir)
     - name: Install a small binary
-      shell: 'julia --color=yes {0}'
-      run: 'using Pkg; Pkg.add(PackageSpec(name="${{ matrix.dep.name }}", version="${{ matrix.dep.version }}"))'
+      shell: julia --color=yes {0}
+      run: |
+        using Pkg
+        if VERSION >= v"1.3"
+            Pkg.add(PackageSpec(name="pandoc_jll", version="3"))
+        else
+            Pkg.add(PackageSpec(name="Scratch", version="1"))
+            using Scratch
+            get_scratch!("test")
+        end
+
+  test-restore:
+    needs: test-save
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        nested:
+          - name: matrix
+            invalid-chars: ","  # Use invalid characters in job matrix to ensure we escape them
+        version:
+          - "1.0"
+          - "1"
+          - "nightly"
+        os:
+          - ubuntu-latest
+          - windows-latest
+          - macOS-latest
+        exclude:
+          # Test Julia "1.0" on Linux only
+          - version: "1.0"
+            os: windows-latest
+          - version: "1.0"
+            os: macOS-latest
+      fail-fast: false
+    env:
+      JULIA_DEPOT_PATH: /tmp/julia-depot
+    steps:
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+    - uses: julia-actions/setup-julia@v1
+      with:
+        version: ${{ matrix.version }}
+    - name: Restore cache
+      id: cache
+      uses: ./
+      with:
+        cache-name: ${{ needs.test-save.outputs.cache-name }}
+        # Cannot require a successful cache delete on forked PRs as the permissions for actions is limited to read
+        delete-old-caches: ${{ github.event.pull_request.head.repo.fork && 'false' || 'required' }}
+    - name: Test cache-hit output
+      shell: julia --color=yes {0}
+      run: |
+        @show ENV["cache-hit"]
+        @assert ENV["cache-hit"] == "true"
+      env:
+        cache-hit: ${{ steps.cache.outputs.cache-hit }}
+    - name: Check existance or emptiness of affected dirs
+      shell: julia --color=yes {0}
+      run: |
+        # These dirs should exist as they've been cached
+        artifacts_dir = joinpath(first(DEPOT_PATH), "artifacts")
+        if VERSION >= v"1.3"
+            @assert !isempty(readdir(artifacts_dir))
+        else
+            @assert !isdir(artifacts_dir)
+        end
+        packages_dir = joinpath(first(DEPOT_PATH), "packages")
+        @assert !isempty(readdir(packages_dir))
+        compiled_dir = joinpath(first(DEPOT_PATH), "compiled")
+        @assert !isempty(readdir(compiled_dir))
+        scratchspaces_dir = joinpath(first(DEPOT_PATH), "scratchspaces")
+        @assert !isempty(readdir(scratchspaces_dir))
+        logs_dir = joinpath(first(DEPOT_PATH), "logs")
+        @assert !isempty(readdir(logs_dir))
 
   # Do tests with no matrix also given the matrix is auto-included in cache key
   test-save-nomatrix:
-    needs: generate-key
+    needs: generate-prefix
     runs-on: ubuntu-latest
+    outputs:
+      cache-name: ${{ steps.cache-name.outputs.cache-name }}
     steps:
     - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+    - name: Set cache-name
+      id: cache-name
+      run: |
+        echo "cache-name=${{ needs.generate-prefix.outputs.cache-prefix }}-${{ github.job }}" >>"$GITHUB_OUTPUT"
     - name: Save cache
       id: cache
       uses: ./
       with:
-        cache-name: ${{ needs.generate-key.outputs.cache-name }}
+        cache-name: ${{ steps.cache-name.outputs.cache-name }}
+        delete-old-caches: required
     - name: Check no artifacts dir
-      shell: 'julia --color=yes {0}'
+      shell: julia --color=yes {0}
       run: |
         dir = joinpath(first(DEPOT_PATH), "artifacts")
         @assert !isdir(dir)
     - name: Install a small binary
-      shell: 'julia --color=yes {0}'
-      run: 'using Pkg; Pkg.add("pandoc_jll")'
-
-  test-restore:
-    needs: [generate-key, test-save]
-    strategy:
-      matrix:
-        dep:
-          - name: pandoc_jll
-            version: "3"
-        os:
-          - ubuntu-latest
-          - windows-latest
-          - macOS-latest
-      fail-fast: false
-    runs-on: ${{ matrix.os }}
-    env:
-      JULIA_DEPOT_PATH: /tmp/julia-depot
-    steps:
-    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
-    - name: Restore cache
-      id: cache
-      uses: ./
-      with:
-        cache-name: ${{ needs.generate-key.outputs.cache-name }}
-    - name: Test cache-hit output
-      shell: 'julia --color=yes {0}'
+      shell: julia --color=yes {0}
       run: |
-        @show ENV["cache-hit"]
-        @assert ENV["cache-hit"] == "true"
-      env:
-        cache-hit: ${{ steps.cache.outputs.cache-hit }}
-    - name: Check existance or emptiness of affected dirs
-      shell: 'julia --color=yes {0}'
-      run: |
-        # These dirs should exist as they've been cached
-        artifacts_dir = joinpath(first(DEPOT_PATH), "artifacts")
-        @assert !isempty(readdir(artifacts_dir))
-        packages_dir = joinpath(first(DEPOT_PATH), "packages")
-        @assert !isempty(readdir(packages_dir))
-        compiled_dir = joinpath(first(DEPOT_PATH), "compiled")
-        @assert !isempty(readdir(compiled_dir))
-        scratchspaces_dir = joinpath(first(DEPOT_PATH), "scratchspaces")
-        @assert !isempty(readdir(scratchspaces_dir))
-        logs_dir = joinpath(first(DEPOT_PATH), "logs")
-        @assert !isempty(readdir(logs_dir))
+        using Pkg
+        if VERSION >= v"1.3"
+            Pkg.add(PackageSpec(name="pandoc_jll", version="3"))
+        else
+            Pkg.add(PackageSpec(name="Scratch", version="1"))
+            using Scratch
+            get_scratch!("test")
+        end
 
   test-restore-nomatrix:
-    needs: [generate-key, test-save-nomatrix]
+    needs: test-save-nomatrix
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
@@ -136,20 +198,26 @@ jobs:
       id: cache
       uses: ./
       with:
-        cache-name: ${{ needs.generate-key.outputs.cache-name }}
+        cache-name: ${{ needs.test-save-nomatrix.outputs.cache-name }}
+        # Cannot require a successful cache delete on forked PRs as the permissions for actions is limited to read
+        delete-old-caches: ${{ github.event.pull_request.head.repo.fork && 'false' || 'required' }}
     - name: Test cache-hit output
-      shell: 'julia --color=yes {0}'
+      shell: julia --color=yes {0}
       run: |
         @show ENV["cache-hit"]
         @assert ENV["cache-hit"] == "true"
       env:
         cache-hit: ${{ steps.cache.outputs.cache-hit }}
     - name: Check existance or emptiness of affected dirs
-      shell: 'julia --color=yes {0}'
+      shell: julia --color=yes {0}
       run: |
         # These dirs should exist as they've been cached
         artifacts_dir = joinpath(first(DEPOT_PATH), "artifacts")
-        @assert !isempty(readdir(artifacts_dir))
+        if VERSION >= v"1.3"
+            @assert !isempty(readdir(artifacts_dir))
+        else
+            @assert !isdir(artifacts_dir)
+        end
         packages_dir = joinpath(first(DEPOT_PATH), "packages")
         @assert !isempty(readdir(packages_dir))
         compiled_dir = joinpath(first(DEPOT_PATH), "compiled")
@@ -159,3 +227,54 @@ jobs:
         logs_dir = joinpath(first(DEPOT_PATH), "logs")
         @assert !isempty(readdir(logs_dir))
 
+  test-save-cloned-registry:
+    needs: generate-prefix
+    runs-on: ubuntu-latest
+    outputs:
+      cache-name: ${{ steps.cache-name.outputs.cache-name }}
+    steps:
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+    - name: Set cache-name
+      id: cache-name
+      run: |
+        echo "cache-name=${{ needs.generate-prefix.outputs.cache-prefix }}-${{ github.job }}" >>"$GITHUB_OUTPUT"
+    - name: Save cache
+      uses: ./
+      with:
+        cache-name: ${{ steps.cache-name.outputs.cache-name }}
+        # Cannot require a successful cache delete on forked PRs as the permissions for actions is limited to read
+        delete-old-caches: ${{ github.event.pull_request.head.repo.fork && 'false' || 'required' }}
+    - name: Add General registry clone
+      shell: julia --color=yes {0}
+      run: |
+        using Pkg
+        Pkg.Registry.add("General")
+      env:
+        JULIA_PKG_SERVER: ""
+    # Set the registry worktree to an older state to simulate the cache storing an old version of the registry.
+    - name: Use outdated General worktree
+      run: git -C ~/.julia/registries/General reset --hard HEAD~20
+
+  test-restore-cloned-registry:
+    needs: test-save-cloned-registry
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+    - name: Add General registry clone
+      shell: julia --color=yes {0}
+      run: |
+        using Pkg
+        Pkg.Registry.add("General")
+      env:
+        JULIA_PKG_SERVER: ""
+    - name: Restore cache
+      uses: ./
+      with:
+        cache-name: ${{ needs.test-save-cloned-registry.outputs.cache-name }}
+        # Cannot require a successful cache delete on forked PRs as the permissions for actions is limited to read
+        delete-old-caches: ${{ github.event.pull_request.head.repo.fork && 'false' || 'required' }}
+    - name: Test registry is not corrupt
+      shell: julia --color=yes {0}
+      run: |
+        using Pkg
+        Pkg.Registry.update()

action.yml

@@ -69,19 +69,32 @@ runs:
         else
             depot="~/.julia"
         fi
-        echo "depot=$depot" >> $GITHUB_OUTPUT
-        [ "${{ inputs.cache-artifacts }}" = "true" ] && A_PATH="${depot}/artifacts"
-        echo "artifacts-path=$A_PATH" >> $GITHUB_OUTPUT
-        [ "${{ inputs.cache-packages }}" = "true" ] && P_PATH="${depot}/packages"
-        echo "packages-path=$P_PATH" >> $GITHUB_OUTPUT
-        [ "${{ inputs.cache-registries }}" = "true" ] && R_PATH="${depot}/registries"
-        echo "registries-path=$R_PATH" >> $GITHUB_OUTPUT
-        [ "${{ inputs.cache-compiled }}" = "true" ] && PCC_PATH="${depot}/compiled"
-        echo "compiled-path=$PCC_PATH" >> $GITHUB_OUTPUT
-        [ "${{ inputs.cache-scratchspaces }}" = "true" ] && S_PATH="${depot}/scratchspaces"
-        echo "scratchspaces-path=$S_PATH" >> $GITHUB_OUTPUT
-        [ "${{ inputs.cache-logs }}" = "true" ] && L_PATH="${depot}/logs"
-        echo "logs-path=$L_PATH" >> $GITHUB_OUTPUT
+        echo "depot=$depot" | tee -a "$GITHUB_OUTPUT"
+
+        cache_paths=()
+        artifacts_path="${depot}/artifacts"
+        [ "${{ inputs.cache-artifacts }}" = "true" ] && cache_paths+=("$artifacts_path")
+        packages_path="${depot}/packages"
+        [ "${{ inputs.cache-packages }}" = "true" ] && cache_paths+=("$packages_path")
+        registries_path="${depot}/registries"
+        if [ "${{ inputs.cache-registries }}" = "true" ]; then
+            if [ ! -d "${registries_path/#\~/$HOME}" ]; then
+                cache_paths+=("$registries_path")
+            else
+                echo "::warning::Julia depot registries already exist. Skipping restoring of cached registries to avoid potential merge conflicts when updating. Please ensure that \`julia-actions/cache\` precedes any workflow steps which add registries."
+            fi
+        fi
+        compiled_path="${depot}/compiled"
+        [ "${{ inputs.cache-compiled }}" = "true" ] && cache_paths+=("$compiled_path")
+        scratchspaces_path="${depot}/scratchspaces"
+        [ "${{ inputs.cache-scratchspaces }}" = "true" ] && cache_paths+=("$scratchspaces_path")
+        logs_path="${depot}/logs"
+        [ "${{ inputs.cache-logs }}" = "true" ] && cache_paths+=("$logs_path")
+        {
+            echo "cache-paths<<EOF"
+            printf "%s\n" "${cache_paths[@]}"
+            echo "EOF"
+        } | tee -a "$GITHUB_OUTPUT"
       shell: bash
       env:
         PATH_DELIMITER: ${{ runner.OS == 'Windows' && ';' || ':' }}
@@ -95,6 +108,9 @@ runs:
           matrix_key=$(echo "$MATRIX_JSON" | jq 'paths(type != "object") as $p | ($p | join("-")) + "=" + (getpath($p) | tostring)' | jq -rs 'join(";") | . + ";"')
         fi
         restore_key="${{ inputs.cache-name }};os=${{ runner.os }};${matrix_key}"
+        # URL encode any restricted characters:
+        # https://github.com/actions/toolkit/blob/5430c5d84832076372990c7c27f900878ff66dc9/packages/cache/src/cache.ts#L38-L43
+        restore_key=$(sed 's/,/%2C/g' <<<"${restore_key}")
         key="${restore_key}run_id=${{ github.run_id }};run_attempt=${{ github.run_attempt }}"
         echo "restore-key=${restore_key}" >> $GITHUB_OUTPUT
         echo "key=${key}" >> $GITHUB_OUTPUT
@@ -106,20 +122,17 @@ runs:
       id: cache
       with:
         path: |
-          ${{ steps.paths.outputs.artifacts-path }}
-          ${{ steps.paths.outputs.packages-path }}
-          ${{ steps.paths.outputs.registries-path }}
-          ${{ steps.paths.outputs.scratchspaces-path }}
-          ${{ steps.paths.outputs.compiled-path }}
-          ${{ steps.paths.outputs.logs-path }}
-
+          ${{ steps.paths.outputs.cache-paths }}
         key: ${{ steps.keys.outputs.key }}
         restore-keys: ${{ steps.keys.outputs.restore-key }}
         enableCrossOsArchive: false
 
-    - name: list restored depot directory sizes
-      if: ${{ steps.cache.outputs.cache-hit == 'true' }}
-      run: du -shc ${{ steps.paths.outputs.depot }}/* || true
+    # if it wasn't restored make the depot anyway as a signal that this action ran
+    # for other julia actions to check, like https://github.com/julia-actions/julia-buildpkg/pull/41
+    - name: make depot if not restored, then list depot directory sizes
+      run: |
+        mkdir -p ${{ steps.paths.outputs.depot }}
+        du -shc ${{ steps.paths.outputs.depot }}/* || true
       shell: bash
 
     # github and actions/cache doesn't provide a way to update a cache at a given key, so we delete any
@@ -127,21 +140,21 @@ runs:
 
     # Not windows
     - uses: pyTooling/Actions/with-post-step@adef08d3bdef092282614f3b683897cefae82ee3
-      if: ${{ inputs.delete-old-caches == 'true' && runner.OS != 'Windows' }}
+      if: ${{ inputs.delete-old-caches != 'false' && runner.OS != 'Windows' }}
       with:
         # seems like there has to be a `main` step in this action. Could list caches for info if we wanted
         # main:  julia ${{ github.action_path }}/handle_caches.jl "${{ github.repository }}" "list"
         main: echo ""
-        post: julia $GITHUB_ACTION_PATH/handle_caches.jl "${{ github.repository }}" "rm" "${{ steps.keys.outputs.restore-key }}"
+        post: julia $GITHUB_ACTION_PATH/handle_caches.jl rm "${{ github.repository }}" "${{ steps.keys.outputs.restore-key }}" "${{ github.ref }}" "${{ inputs.delete-old-caches != 'required' }}"
       env:
         GH_TOKEN: ${{ inputs.token }}
 
     # Windows (because this action uses command prompt on windows)
     - uses: pyTooling/Actions/with-post-step@adef08d3bdef092282614f3b683897cefae82ee3
-      if: ${{ inputs.delete-old-caches == 'true' && runner.OS == 'Windows' }}
+      if: ${{ inputs.delete-old-caches != 'false' && runner.OS == 'Windows' }}
       with:
         main: echo ""
-        post: cd %GITHUB_ACTION_PATH% && julia handle_caches.jl "${{ github.repository }}" "rm" "${{ steps.keys.outputs.restore-key }}"
+        post: cd %GITHUB_ACTION_PATH% && julia handle_caches.jl rm "${{ github.repository }}" "${{ steps.keys.outputs.restore-key }}" "${{ github.ref }}" "${{ inputs.delete-old-caches != 'required' }}"
       env:
         GH_TOKEN: ${{ inputs.token }}
 

handle_caches.jl

@@ -1,58 +1,67 @@
 using Pkg, Dates
 function handle_caches()
-    repo = ARGS[1]
-    func = ARGS[2]
-    restore_key = get(ARGS, 3, "")
+    subcommand = ARGS[1]
 
-    if func == "list"
+    if subcommand == "list"
+        repo = ARGS[2]
         println("Listing existing caches")
         run(`gh cache list --limit 100 --repo $repo`)
-    elseif func == "rm"
-        caches = String[]
-        failed = String[]
-        for _ in 1:5 # limit to avoid accidental rate limiting
-            hits = split(strip(read(`gh cache list --limit 100 --repo $repo`, String)), keepempty=false)
-            search_again = length(hits) == 100
-            filter!(startswith(restore_key), hits)
-            isempty(hits) && break
-            # We can delete everything that matches the restore key because the new cache is saved later.
-            for c in hits
+    elseif subcommand == "rm"
+        repo, restore_key, ref = ARGS[2:4]
+        allow_failure = ARGS[5] == "true"
+
+        endpoint = "/repos/$repo/actions/caches"
+        page = 1
+        per_page = 100
+        escaped_restore_key = replace(restore_key, "\"" => "\\\"")
+        query = ".actions_caches[] | select(.key | startswith(\"$escaped_restore_key\")) | .id"
+
+        deletions = String[]
+        failures = String[]
+        while 1 <= page <= 5 # limit to avoid accidental rate limiting
+            cmd = `gh api -X GET $endpoint -F ref=$ref -F per_page=$per_page -F page=$page --jq $query`
+            ids = split(read(cmd, String); keepempty=false)
+            page = length(ids) == per_page ? page + 1 : -1
+
+            # We can delete all cache entries on this branch that matches the restore key
+            # because the new cache is saved later.
+            for id in ids
                 try
-                    run(`gh cache delete $(split(c)[1]) --repo $repo`)
-                    push!(caches, c)
+                    run(`gh cache delete $id --repo $repo`)
+                    push!(deletions, id)
                 catch e
                     @error e
-                    push!(failed, c)
+                    push!(failures, id)
                 end
             end
-            search_again || break
         end
-        if isempty(failed) && isempty(caches)
-            println("No existing caches found for restore key `$restore_key`")
+        if isempty(failures) && isempty(deletions)
+            println("No existing caches found on ref `$ref` matching restore key `$restore_key`")
         else
-            if !isempty(failed)
-                println("Failed to delete $(length(failed)) existing caches for restore key `$restore_key`")
-                println.(failed)
+            if !isempty(failures)
+                println("Failed to delete $(length(failures)) existing caches on ref `$ref` matching restore key `$restore_key`")
+                println.(failures)
                 @info """
                     To delete caches you need to grant the following to the default `GITHUB_TOKEN` by adding
-                    this to your yml:
+                    this to your workflow:
                     ```
                     permissions:
-                        actions: write
-                        contents: read
+                      actions: write
+                      contents: read
                     ```
                     (Note this won't work for fork PRs but should once merged)
                     Or provide a token with `repo` scope via the `token` input option.
                     See https://cli.github.com/manual/gh_cache_delete
                     """
+                allow_failure || exit(1)
             end
-            if !isempty(caches)
-                println("$(length(caches)) existing caches deleted that match restore key `$restore_key`:")
-                println.(caches)
+            if !isempty(deletions)
+                println("Deleted $(length(deletions)) caches on ref `$ref` matching restore key `$restore_key`")
+                println.(deletions)
             end
         end
     else
-        throw(ArgumentError("Unexpected second argument: $func"))
+        throw(ArgumentError("Unexpected subcommand: $subcommand"))
     end
 end