eric-burel

# -*- coding: utf-8 -*-
"""LBKE Anthropic batch runner.ipynb

## Demo of calling Anthropic Batch API from a LangGraph graph

Improvements:
- an interrupt should be prefered to ending the chart when the work is pending => currently each new run to check if the batch is there will send a new message to the thread since we re-invoke the chart
- better processing responses, maybe using a MessageState/add reducer to treat the response as an assistant message added to a conversation
- as a bonus, polling the chart on a predefined frequence (useful for a quick run)
- test if it still works after the notebook is closed thanks to gdrive checkpointing

eric-burel

# Commented out IPython magic to ensure Python compatibility.
# %pip install -q langgraph
# %pip install -q langgraph-checkpoint-sqlite
# Demo notebook: https://colab.research.google.com/drive/16V9TpOsLMOuU655ryerni7Hc7mni6hcU?usp=sharing

import os

from google.colab import drive
DRIVE_MOUNT_POINT = "/content/drive"
drive.mount(DRIVE_MOUNT_POINT)

eric-burel

• Can't mix static layout and dynamic pages, this prevents define common static data requirements in the layout and then optionnally dynamic requirements in some pages below: vercel/next.js#44712 (comment)

• RSC might be a way to colocate graphql data requirements with components, but it's not very declarative

• i18n: since "FormattedMessage" tend to be a leaf component, we can't really server-side render it, all i18n messages thus need hydration

• 404 are delicate because route handlers can be mixed with routes : vercel/next.js#49696, and also when fetching files (you may conflate a file with the first route parameter of your app eg [lang])

eric-burel

const {EditorState} = require("prosemirror-state")
const {EditorView} = require("prosemirror-view")
const {Schema, DOMParser} = require("prosemirror-model")
const {schema} = require("prosemirror-schema-basic")
const {addListNodes} = require("prosemirror-schema-list")
const {exampleSetup} = require("prosemirror-example-setup")
const { inputRules, InputRule } = require("prosemirror-inputrules")


const starRule = new InputRule(/\*(.+)\*/, (state, match, start, end) => {

eric-burel

import fs from "fs";
import os from "os";
import path from "path";
const fsPromises = fs.promises;

// DATA STORE
// In a real project this belongs to the db server code
// We simulate a distant db by storing data in a file
const filePath = path.resolve(os.homedir(), ".mockdb");
async function updateData<TData=any>(data:TData) {

eric-burel

/**
 * Run this script manually in the console and copy-paste the result to generate
 * a TOC based on "section" layout, instead of markdown titles (that are not helpful here)
 *
 * NOTE: this script is loaded in index.html,
 * so you can call generateTocFromSections easily from the slidev app, in the browser console
 *
 * NOTE: this could also be adapted with Cheerio or an HTML/md parser to parse the markdown directly?
 */
function generateTocFromSections() {

eric-burel

const edits = [
  [
    "node_modules/ts-invariant/package.json",
    {
      type: "module",
      exports: {
        ".": "./lib/invariant.esm.js",
        "./process/index.js": "./process/index.js",
      },
    },

eric-burel

import { useIntlContext } from "@vulcanjs/i18n";
import { useEffect, useRef } from "react";
import { block } from "./block";
import debug from "debug";
const debugTransitions = debug("vn:route-transition");

/**
 * Can trigger an alert on unsaved changes
 *
 * @see https://github.com/ReactTraining/history/blob/master/docs/blocking-transitions.md

eric-burel

// in next.config.js Webpack field
config.plugins.push(
  new options.webpack.DefinePlugin({
    'process.env.NEXT_IS_SERVER': JSON.stringify(options.isServer.toString()),
  })
)

eric-burel

// in next.config.js
import packageJson from "./package.json"

module.exports = {
  env: {
    NEXT_PUBLIC_APP_VERSION: packageJson.version
  }
}