Added llnotes experiment

.github/workflows/llnotes-release.yml

@@ -0,0 +1,31 @@
+name: Release llnotes
+
+on:
+  push:
+    tags:
+      - 'llnotes/v*'
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    permissions:
+      id-token: write
+      contents: write
+    steps:
+      - uses: actions/checkout@v3
+
+      - name: Setup Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: 1.21
+
+      - name: Build
+        run: /bin/bash .github/workflows/build.sh llnotes
+
+      - name: Compress
+        run: /bin/bash .github/workflows/compress.sh
+
+      - name: Create release
+        uses: softprops/action-gh-release@v1
+        with:
+          files: dist/*

.github/workflows/push-llnotes.yml

@@ -0,0 +1,51 @@
+name: build-llnotes
+
+on:
+  pull_request:
+    types: [opened, synchronize]
+    paths: ['llnotes/**']
+
+permissions:
+  id-token: write
+  contents: read
+  pull-requests: write
+
+jobs:
+  tests:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Setup Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: 1.21
+      - name: Install Tools
+        run: go install gotest.tools/gotestsum@latest
+      - name: Test
+        working-directory: llnotes
+        run: make test
+
+  fmt:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Setup Go
+        uses: actions/setup-go@v4
+        with:
+          go-version: 1.21.0
+      - name: Install Tools
+        run: |
+          go install golang.org/x/tools/cmd/goimports@latest
+          go install honnef.co/go/tools/cmd/staticcheck@latest
+      - name: Run make fmt
+        working-directory: llnotes
+        run: make fmt
+      - name: Run go mod tidy
+        working-directory: llnotes
+        run: go mod tidy
+      - name: Fail on differences
+        run: |
+          # Exit with status code 1 if there are differences (i.e. unformatted files)
+          git diff --exit-code

Makefile

@@ -35,4 +35,7 @@ fmt-go-libs:
 fmt-acceptance:
 	cd acceptance && make fmt
 
-fmt: fmt-acceptance fmt-go-libs
+fmt-llnotes:
+	cd llnotes && make fmt
+
+fmt: fmt-acceptance fmt-go-libs fmt-llnotes

NOTICE

@@ -10,4 +10,11 @@ This Software contains code from the following projects, licensed under the BSD-
 
 go-github
 Copyright 2013 The go-github AUTHORS. All rights reserved.
-License - https://github.com/google/go-github/blob/master/LICENSE
+License - https://github.com/google/go-github/blob/master/LICENSE
+
+This Software contains code from the following projects, licensed under the MIT license:
+
+go-diff
+Copyright (c) 2014 Sourcegraph, Inc.
+Copyright (c) 2012 Matias Bordese
+https://github.com/sourcegraph/go-diff/blob/master/LICENSE

go-libs/Makefile

@@ -16,6 +16,7 @@ coverage: test
 
 vendor:
 	@echo "✓ Filling vendor folder with library code ..."
+	@go mod tidy
 	@go mod vendor
 
 .PHONY: build vendor coverage test fmt

go-libs/github/github.go

@@ -211,6 +211,13 @@ func (c *GitHubClient) ListCommits(ctx context.Context, org, repo string, req *L
 	})
 }
 
+func (c *GitHubClient) GetCommit(ctx context.Context, org, repo string, sha string) (*RepositoryCommit, error) {
+	var res RepositoryCommit
+	path := fmt.Sprintf("%s/repos/%s/%s/commits/%s", gitHubAPI, org, repo, sha)
+	err := c.api.Do(ctx, "GET", path, httpclient.WithResponseUnmarshal(&res))
+	return &res, err
+}
+
 func (c *GitHubClient) CompareCommits(ctx context.Context, org, repo, base, head string) listing.Iterator[RepositoryCommit] {
 	type response struct {
 		Commits []RepositoryCommit `json:"commits,omitempty"`

go-libs/go.mod

@@ -8,6 +8,7 @@ require (
 	github.com/fatih/color v1.16.0
 	github.com/google/go-querystring v1.1.0
 	github.com/nwidger/jsoncolor v0.3.2
+	github.com/sourcegraph/go-diff v0.7.0
 	github.com/spf13/cobra v1.8.0
 	github.com/spf13/pflag v1.0.5
 	github.com/spf13/viper v1.18.2

go-libs/go.sum

@@ -111,8 +111,12 @@ github.com/sagikazarmark/locafero v0.4.0 h1:HApY1R9zGo4DBgr7dqsTH/JJxLTTsOt7u6ke
 github.com/sagikazarmark/locafero v0.4.0/go.mod h1:Pe1W6UlPYUk/+wc/6KFhbORCfqzgYEpgQ3O5fPuL3H4=
 github.com/sagikazarmark/slog-shim v0.1.0 h1:diDBnUNK9N/354PgrxMywXnAwEr1QZcOr6gto+ugjYE=
 github.com/sagikazarmark/slog-shim v0.1.0/go.mod h1:SrcSrq8aKtyuqEI1uvTDTK1arOWRIczQRv+GVI1AkeQ=
+github.com/shurcooL/go v0.0.0-20180423040247-9e1955d9fb6e/go.mod h1:TDJrrUr11Vxrven61rcy3hJMUqaf/CLWYhHNPmT14Lk=
+github.com/shurcooL/go-goon v0.0.0-20170922171312-37c2f522c041/go.mod h1:N5mDOmsrJOB+vfqUK+7DmDyjhSLIIBnXo9lvZJj3MWQ=
 github.com/sourcegraph/conc v0.3.0 h1:OQTbbt6P72L20UqAkXXuLOj79LfEanQ+YQFNpLA9ySo=
 github.com/sourcegraph/conc v0.3.0/go.mod h1:Sdozi7LEKbFPqYX2/J+iBAM6HpqSLTASQIKqDmF7Mt0=
+github.com/sourcegraph/go-diff v0.7.0 h1:9uLlrd5T46OXs5qpp8L/MTltk0zikUGi0sNNyCpA8G0=
+github.com/sourcegraph/go-diff v0.7.0/go.mod h1:iBszgVvyxdc8SFZ7gm69go2KDdt3ag071iBaWPF6cjs=
 github.com/spf13/afero v1.11.0 h1:WJQKhtpdm3v2IzqG8VMqrr6Rf3UYpEF239Jy9wNepM8=
 github.com/spf13/afero v1.11.0/go.mod h1:GH9Y3pIexgf1MTIWtNGyogA5MwRIDXGUr+hbWNoBjkY=
 github.com/spf13/cast v1.6.0 h1:GEiTHELF+vaR5dhz3VqZfFSzZjYbgeKDpBxQVS4GYJ0=

go-libs/llnotes/announce.go

@@ -0,0 +1,59 @@
+package llnotes
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"github.com/databricks/databricks-sdk-go/listing"
+	"github.com/databricks/databricks-sdk-go/logger"
+)
+
+var blogPrompt = MessageTemplate(`Do not hallucinate.
+You are professional technical writer and you receive draft release notes for {{.version}} of project called {{.repo.Name}} in a markdown format from multiple team members. 
+Project can be described as "{{.repo.Description}}"
+
+You write a long post announcement that takes at least 5 minutes to read, summarize the most important features, and mention them on top. Keep the markdown links when relevant.
+Do not use headings. Write fluent paragraphs, that are at least few sentences long. Blog post title should nicely summarize the feature increments of this release.
+
+Don't abuse lists. paragraphs should have at least 3-4 sentences. The title should be one-sentence summary of the incremental updates for this release
+
+You aim at driving more adoption of the project on Medium.`)
+
+func (lln *llNotes) versionNotes(ctx context.Context, newVersion string) ([]string, error) {
+	versions, err := listing.ToSlice(ctx, lln.gh.Versions(ctx, lln.org, lln.repo))
+	if err != nil {
+		return nil, fmt.Errorf("versions: %w", err)
+	}
+	if newVersion == "" {
+		newVersion = versions[0].Version
+	}
+	prevVersion := "v0.0.0"
+	for i, v := range versions {
+		if v.Version == newVersion {
+			prevVersion = versions[i+1].Version
+			break
+		}
+	}
+	return lln.ReleaseNotesDiff(ctx, prevVersion, newVersion)
+}
+
+func (lln *llNotes) Announce(ctx context.Context, newVersion string) (History, error) {
+	notes, err := lln.versionNotes(ctx, newVersion)
+	if err != nil {
+		return nil, fmt.Errorf("parallel: %w", err)
+	}
+	repo, err := lln.gh.GetRepo(ctx, lln.org, lln.repo)
+	if err != nil {
+		return nil, fmt.Errorf("get repo: %w", err)
+	}
+	rawNotes := strings.Join(notes, "\n")
+	logger.Debugf(ctx, "Raw notes: %s", rawNotes)
+	return lln.Talk(ctx, History{
+		blogPrompt.AsSystem(map[string]any{
+			"version": newVersion,
+			"repo":    repo,
+		}),
+		UserMessage(rawNotes),
+	})
+}

go-libs/llnotes/chain.go

@@ -0,0 +1,115 @@
+package llnotes
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/databricks/databricks-sdk-go/service/serving"
+	"github.com/databrickslabs/sandbox/go-libs/sed"
+)
+
+type message interface {
+	ChatMessage() serving.ChatMessage
+}
+
+type SystemMessage string
+
+func (m SystemMessage) ChatMessage() serving.ChatMessage {
+	return serving.ChatMessage{
+		Role:    serving.ChatMessageRoleSystem,
+		Content: string(m),
+	}
+}
+
+type UserMessage string
+
+func (m UserMessage) ChatMessage() serving.ChatMessage {
+	return serving.ChatMessage{
+		Role:    serving.ChatMessageRoleUser,
+		Content: string(m),
+	}
+}
+
+type AssistantMessage string
+
+func (m AssistantMessage) ChatMessage() serving.ChatMessage {
+	return serving.ChatMessage{
+		Role:    serving.ChatMessageRoleAssistant,
+		Content: string(m),
+	}
+}
+
+type History []message
+
+func (h History) Messages() (out []serving.ChatMessage) {
+	for _, v := range h {
+		out = append(out, v.ChatMessage())
+	}
+	return
+}
+
+func (h History) messageTokens(m message) int {
+	// this is good enough approximation of message token count
+	content := m.ChatMessage().Content
+	return len(strings.Split(content, " "))
+}
+
+func (h History) truncated(m message, maxTokens int) message {
+	// this is good enough approximation of message token count
+	cm := m.ChatMessage()
+	tokens := strings.Split(cm.Content, " ")
+	if len(tokens) < maxTokens {
+		return m
+	}
+	switch m.(type) {
+	case UserMessage:
+		return UserMessage(strings.Join(tokens[:maxTokens-100], " "))
+	case SystemMessage:
+		return SystemMessage(strings.Join(tokens[:maxTokens-100], " "))
+	case AssistantMessage:
+		return AssistantMessage(strings.Join(tokens[:maxTokens-100], " "))
+	}
+	panic("cannot truncate message")
+}
+
+func (h History) totalTokens() int {
+	totalTokens := 0
+	for _, m := range h {
+		totalTokens += h.messageTokens(m)
+	}
+	return totalTokens
+}
+
+func (h History) With(m message) History {
+	maxContextSize := 32768
+	increment := h.messageTokens(m)
+	if increment > maxContextSize {
+		m = h.truncated(m, 32000)
+	}
+	return append(h, m)
+}
+
+func (h History) Last() string {
+	return h[len(h)-1].ChatMessage().Content
+}
+
+func (h History) Excerpt(n int) string {
+	var out []string
+	oneLine := sed.Rule(`\n|\s+`, ` `)
+	for i, v := range h {
+		m := v.ChatMessage()
+		out = append(out, fmt.Sprintf("(%d/%d) %s: %s",
+			i+1, len(h),
+			strings.ToUpper(m.Role.String()),
+			h.onlyNBytes(oneLine.Apply(m.Content), n)))
+	}
+	return strings.Join(out, "\n")
+}
+
+func (h History) onlyNBytes(j string, numBytes int) string {
+	diff := len([]byte(j)) - numBytes
+	if diff > 0 {
+		return fmt.Sprintf("%s... (%d more bytes)", j[:numBytes], diff)
+	}
+	return j
+}