🍫

src/export-book-pdf.ts

@@ -1,19 +1,17 @@
 #!/usr/bin/env node
-/* eslint-disable no-process-env */
 import 'dotenv/config'
 
 import fs from 'node:fs'
 import fsp from 'node:fs/promises'
 import path from 'node:path'
 
-import type { Metadata } from 'playwright/test'
 import PDFDocument from 'pdfkit'
 
-import type { ContentChunk } from './types'
-import { assert } from './utils'
+import type { BookMetadata, ContentChunk } from './types'
+import { assert, getEnv } from './utils'
 
 async function main() {
-  const asin = process.env.ASIN
+  const asin = getEnv('ASIN')
   assert(asin, 'ASIN is required')
 
   const outDir = path.join('out', asin)
@@ -23,20 +21,20 @@ async function main() {
   ) as ContentChunk[]
   const metadata = JSON.parse(
     await fsp.readFile(path.join(outDir, 'metadata.json'), 'utf8')
-  ) as Metadata
+  ) as BookMetadata
   assert(content.length, 'no book content found')
   assert(metadata.meta, 'invalid book metadata: missing meta')
   assert(metadata.toc?.length, 'invalid book metadata: missing toc')
 
   const title = metadata.meta.title
-  const author = metadata.meta.authorList.join('\n')
+  const authors = metadata.meta.authorList
 
   const doc = new PDFDocument({
     autoFirstPage: true,
     displayTitle: true,
     info: {
       Title: title,
-      Author: author
+      Author: authors.join(', ')
     }
   })
   const stream = doc.pipe(fs.createWriteStream(path.join(outDir, 'book.pdf')))
@@ -50,10 +48,10 @@ async function main() {
     doc.text(title, { align: 'center' })
     const w = doc.widthOfString(title)
 
-    const byline = `By ${author}`
+    const byline = `By ${authors.join(',\n')}`
 
     doc.fontSize(20)
-    doc.y -= 10
+    doc.y -= doc.heightOfString(byline) / 2
     doc.text(byline, {
       align: 'center',
       indent: w - doc.widthOfString(byline)

src/extract-kindle-book.ts

@@ -7,7 +7,7 @@ import { input } from '@inquirer/prompts'
 import delay from 'delay'
 import { chromium, type Locator } from 'playwright'
 
-import type { Info, Meta, Metadata, PageChunk } from './types'
+import type { BookInfo, BookMeta, BookMetadata, PageChunk } from './types'
 import {
   assert,
   deromanize,
@@ -55,8 +55,8 @@ async function main() {
   })
   const page = await context.newPage()
 
-  let info: Info | undefined
-  let meta: Meta | undefined
+  let info: BookInfo | undefined
+  let meta: BookMeta | undefined
 
   page.on('response', async (response) => {
     try {
@@ -342,7 +342,7 @@ async function main() {
     } while (true)
   } while (true)
 
-  const result: Metadata = { info: info!, meta: meta!, toc, pages }
+  const result: BookMetadata = { info: info!, meta: meta!, toc, pages }
   await fs.writeFile(
     path.join(outDir, 'metadata.json'),
     JSON.stringify(result, null, 2)

src/transcribe-book-content.ts

@@ -21,7 +21,7 @@ async function main() {
 
   const openai = new OpenAIClient()
 
-  const results: ContentChunk[] = (
+  const content: ContentChunk[] = (
     await pMap(
       pageScreenshots,
       async (screenshot) => {
@@ -40,6 +40,7 @@ async function main() {
         )
 
         try {
+          const maxRetries = 20
           let retries = 0
 
           do {
@@ -78,13 +79,19 @@ Do not include any additional text, descriptions, or punctuation. Ignore any emb
 
             if (!text) continue
             if (text.length < 100 && /i'm sorry/i.test(text)) {
+              if (retries >= maxRetries) {
+                throw new Error(
+                  `Model refused too many times (${retries} times): ${text}`
+                )
+              }
+
               // Sometimes the model refuses to generate text for an image
               // presumably if it thinks the content may be copyrighted or
-              // otherwise inappropriate. I haven't seen this from "gpt-4o",
-              // but I have seen it more regularly from "gpt-4o-mini", so in
-              // this case we'll retry with a higher temperature and cross our
-              // fingers.
-              console.warn(`retrying refusal...`, { index, text, screenshot })
+              // otherwise inappropriate. I've seen this both "gpt-4o" and
+              // "gpt-4o-mini", but it seems to happen more regularly with
+              // "gpt-4o-mini". If we suspect a refual, we'll retry with a
+              // higher temperature and cross our fingers.
+              console.warn('retrying refusal...', { index, text, screenshot })
               continue
             }
 
@@ -108,9 +115,9 @@ Do not include any additional text, descriptions, or punctuation. Ignore any emb
 
   await fs.writeFile(
     path.join(outDir, 'content.json'),
-    JSON.stringify(results, null, 2)
+    JSON.stringify(content, null, 2)
   )
-  console.log(JSON.stringify(results, null, 2))
+  console.log(JSON.stringify(content, null, 2))
 }
 
 await main()

src/types.ts

@@ -19,7 +19,7 @@ export interface PageChunk {
   screenshot: string
 }
 
-export interface Meta {
+export interface BookMeta {
   ACR: string
   asin: string
   authorList: Array<string>
@@ -42,7 +42,7 @@ export interface Meta {
   endPosition: number
 }
 
-export interface Info {
+export interface BookInfo {
   clippingLimit: number
   contentChecksum: any
   contentType: string
@@ -69,9 +69,9 @@ export interface Info {
   srl: number
 }
 
-export interface Metadata {
-  info: Info
-  meta: Meta
+export interface BookMetadata {
+  info: BookInfo
+  meta: BookMeta
   toc: TocItem[]
   pages: PageChunk[]
 }