To protect against link rot and disappearing content, I save local copies of web pages I like into my web archive. Ideally, this archive is completely self-contained. I want a local copy of each page’s HTML and all the assets it depends on (CSS, images, scripts, and so on). I don’t want my archived pages fetching resources from the original website, which could disappear at any time. Initially, I was verifying self-containedness by hand. I’d open each saved page in a web browser, and check the browser dev tools. I’d check the Console tab for errors and failed requests, and watch the Network tab for requests going out to the internet. That approach was tedious and error-prone, and I wanted something better. I now automate the process using Playwright, a testing framework that allows you to control web browsers. I open each page using Playwright, and check if it’s making external requests. Detecting external requests in Playwright When you open a page in Playwright, you can subscribe to request events which tell you when the page issues a request. In this Python snippet, we start Playwright, start a WebKit browser, and open a new page. We tell the page to append external requests to a list, and then load example.com. We see that example.com makes two requests on page load: one for https://example.com, and one for a JavaScript file. from contextlib import closing from playwright.sync_api import sync_playwright with ( sync_playwright() as p, closing(p.webkit.launch()) as browser, closing(browser.new_page()) as page, ): requests = [] page.on("request", lambda msg: requests.append(msg)) page.goto("https://example.com") print("requests:", requests) # requests: [<Request url='https://example.com/' method='GET'>, # <Request url='https://example.com/s.js' method='GET'>] In a more complex scenario, we might use Playwright to interact with the page after loading – for example, clicking a button or filling in a form, and then we could see the requests triggered by each action. Once we have a list of requests made by the page, we can inspect the url property to check if any of them were fetching from an external website. I wrote a function is_external_request to carry out this check: from playwright.sync_api import Request import urllib.parse def is_external_request(req: Request, original_url: str) -> bool: """ Check if a request has a different origin than the original URL. A request is considered internal if: - it matches the original URL exactly - it has the same scheme, host, and port as the original URL - it uses a `blob:` scheme (for example, in-memory media representations) """ if req.url == original_url: return False # Compare the scheme/host of the request to that of the original URL. org_url = urllib.parse.urlsplit(original_url) req_url = urllib.parse.urlsplit(req.url) if org_url.scheme == req_url.scheme and org_url.netloc == req_url.netloc: return False # Ignore blob: URLs, which are in-memory representation of some data, # and not an external request. # # Playwright detects blob: URLs when loading media files; for example # using the <audio> or <video> tag. if req_url.scheme == "blob": return False # Fail open: if we can't decide if a request was internal or external, # assume it was external. return True assert_page_is_self_contained(ctx, url) I’ve wrapped these two steps in a function assert_page_is_self_contained(ctx, url), which is part of my chives utility library. The function takes a Playwright Browser (or BrowserContext) and a URL, then: Opens a new page in the browser Loads the URL in the page Checks what happened when the page completes loading It throws an exception if there are any: Uncaught page exceptions or errors Console warnings or errors Non-GET requests Requests targeting external resources The function tries to trigger important network requests that might be skipped on the initial page load. For example, it evaluates the body’s offsetHeight to force a synchronous reflow, which causes the browser to request any lazily loaded images or fonts that might otherwise be missed. Here’s an example which verifies a file example.html from the working directory: from chives.browser_fixtures import assert_page_is_self_contained, file_uri from contextlib import closing from playwright.sync_api import sync_playwright with ( sync_playwright() as p, closing(p.webkit.launch()) as browser, ): assert_page_is_self_contained(browser, url=file_uri("example.html")) I originally passed a Page as the first argument, but changing the function to take a Browser or BrowserContext lets it manage its own pages. It gets a fresh page for every verification, and closes it when it’s done. (The first time I ran this, I created a new page for every verification, and never closed them – my computer ground to a crawl as my script opened hundreds of browser tabs.) If you create a new context, you can emulate different browser settings, like location, locale, and window size. Here’s an example where we use contexts to test the same page in light/dark mode: with ( sync_playwright() as p, closing(p.webkit.launch()) as browser, ): with closing(browser.new_context(color_scheme="light")) as light_ctx: assert_page_is_self_contained(light_ctx, url=file_uri("example.html")) with closing(browser.new_context(color_scheme="dark")) as dark_ctx: assert_page_is_self_contained(dark_ctx, url=file_uri("example.html")) In my web archive, I check every page in light/dark mode, and at 1×/2× pixel densities. It takes at least a second to run each verification, so I cache the results. As I save new pages, they’re added to a verification queue, and I’m warned of any missing assets. Unsurprisingly, the first run flagged many missing fonts and images, which I added to my local archive. Every saved page now passes cleanly. Browser is best-effort, but may still miss things Even with automated testing, it’s difficult to guarantee a completely self-contained archive. Different instances of a browser might load different assets, and a single page load might miss different variations: Responsive images – web pages can offer different sizes of an image, and the browser chooses a specific size based on the current window dimensions Appearance settings – a browser might load different fonts or images depending on light/dark mode Interaction-dependent assets – a web page might have assets that are only required on hover or user action You can mitigate some of this by testing with multiple contexts or adding post-load interactions, but every step makes the process take longer. Fortunately, most of the pages I save are simple, blog-style pages where I’ve stripped out the JavaScript. For my specific use case, initial load verification is more than good enough. Other uses for this code There are other projects where it could be useful to intercept and audit the requests made by a web page: Verifying your Content Security Policy (CSP). The CSP header restricts what resources can be loaded by a web page, and you could check that your configuration has the correct effect. Auditing login screens. You could check that sensitive pages don’t load any external scripts, which mitigates the risk of credential theft via a compromised third-party. Testing analytics and tracking. You could script user flows, like a shopping cart checkout, and confirm that tracking events fire correctly. I don’t do much front-end web development, but Playwright has become a key tool in my toolbox. I read a lot about people using AI agents and MCPs to interact with websites, but I prefer Playwright. It’s more fiddly and requires explicit instructions, but it’s constrained, deterministic, and I never have to worry about it doing something unexpected. [If the formatting of this post looks odd in your feed reader, visit the original article]

I listen to a lot of podcasts, and I like how they fit around other tasks. I press play, lock my phone, and put it down. I’m free to wash the dishes, fold the laundry, or shop for groceries. Unfortunately, more and more information is only published as a video. Technical talks, conference sessions, video essays – they don’t work in an audio-only podcast app. I could convert these videos to MP3 files, but that breaks down the moment a video isn’t pure spoken word. If a speaker says, “Look at this slide” or holds up a diagram, an audio-only file leaves me stranded. I don’t want to give up the podcast player I like, nor stare at a screen for an hour – but I do want the information in these videos. To solve this, I’m abusing my podcast player’s chapter support. This gives me the best of both worlds: I can listen to a video as audio-first, and glance at my lock screen if I need a moment of visual context. The idea: Chapters every few seconds MP3 files can have ID3 metadata, and ID3 metadata can include chapters. A chapter covers a particular time range, and it can have an associated title, description, and cover art. My podcast app of choice is Overcast, which can’t play videos, but it does have robust chapter support. I can jump between chapters, navigate a table of contents, and see per-chapter cover art. To get videos into Overcast, I’m creating MP3 files with a new chapter every few seconds, and the per-chapter cover art is a corresponding frame from the video. As I play the file, I get a slow, stop-motion-like rendition of the original video. If my phone is locked, I can glance at my lock screen and see the current frame in the Now Playing screen. Overcast is developed by Marco Arment, and I got this idea from Forecast, his app for adding chapters to podcasts. In particular, I was struck by its ability to create chapters that don’t display in the chapter list – ideal if I don’t want a table of contents with hundreds of entries. As I was developing my script, I compared my output to the output from Forecast to ensure I was creating the chapters correctly. The code: FFmpeg and Mutagen There are three steps in this process: Convert a video file to an MP3 Extract images from the video at a fixed interval Insert the images as hidden chapters in the MP3 file Let’s go through each in turn. 1. Convert a video file to an MP3 Converting a video file to an MP3 is a single FFmpeg command: ffmpeg -i video.mp4 audio.mp3 This is consistently the slowest step of the process, and I do wonder if I could use different settings or an alternative encoder to make it go faster – but it’s not slow enough to be worth further investigation. 2. Extract images from the video at a fixed interval Extracting images from a video needs a more complicated FFmpeg command: ffmpeg -i video.mp4 \ -vf 'fps=1/5,scale=iw*sar:ih,scale=min(iw\,945):min(ih\,945):force_original_aspect_ratio=decrease' \ thumbnail_%04d.jpg This extracts an image every 5 seconds, downscales any image larger than 945 pixels square (while preserving the original aspect ratio), and saves the results as sequentially numbered JPEG images (thumbnail_0001.png, thumbnail_0002.png, and so on). The key is the -vf flag, which defines two FFmpeg filters: The fps filter selects one frame every 5 seconds (fps=1/5). The first scale filter scales the width based on the sample aspect ratio (scale=iw*sar:ih). Without this filter, frames can be stretched and distorted. The second scale filter scales the input video, preserving the original aspect ratio (force_original_aspect_ratio=decrease), and ensuring the output images fit within 945×945px or the size of the input video, whichever is smaller. My limit is 945 pixels because that’s the largest size that cover art is shown on my iPhone. This filter still isn’t completely correct – it sometimes creates images from portrait videos that are smaller than I’m expecting – but it’s good enough. These are only thumbnails for glancing at, and if I want to change it later, I can always do the image resizing outside FFmpeg. 3. Insert the images as hidden chapters in the MP3 file Inserting the chapters into the MP3 file is more complicated. Although FFmpeg has basic support for ID3 metadata, as far as I know, it can’t insert chapters with per-chapter artwork. Instead, I’m going to reach for Python and the Mutagen library. Here’s the code to add a chapter to an MP3 file: from mutagen.id3 import APIC, CHAP, ID3, PictureType audio = ID3("audio.mp3") with open("thumbnail_0001.jpg", "rb") as f: img_data = f.read() image_frame = APIC(mime="image/jpeg", type=PictureType.OTHER, data=img_data) chapter_frame = CHAP( element_id="chp1", start_time=0, end_time=5 * 1000, sub_frames=[image_frame] ) audio.add(chapter_frame) audio.save() This creates a single chapter that lasts the first 5 seconds (0 to 5000 milliseconds), and the per-chapter cover art is thumbnail_0001.jpg. If we ran this in a loop, we could add images for every 5 second slice of the original video. This code is inserting two frames into the ID3 metadata: The CHAP (chapter) frame contains the timing information, and it can have subframes for metadata like title, chapter art, or associated URL. The APIC (attached picture) subframe contains information about a picture, which can either be a blob of image data or a URL to an image on the web. Normally, you’d also insert a CTOC frame which defines a table of contents, but I don’t want a TOC with hundreds of 5-second chapters, so I’m deliberately not doing this here. This is allowed by the ID3 spec – you’re not required to insert a CTOC frame if you’re using chapters, and you can have chapters that aren’t listed in your table of contents. To work out which frames I needed, I used Forecast to create some chapters by hand, and I inspected their frames. In particular, loading an MP3 and calling Mutagen’s pprint() method shows a human-readable list of frames, and then I could drill into the individual fields: from mutagen.id3 import ID3 audio = ID3("audio.mp3") print(audio.pprint()) I wrapped all this code in a project called glancecast, which allows you to convert a video file with a single command, with optional flags to set the frame length and chapter art size: $ python3 glancecast.py interesting_talk.mp4 interesting_talk.mp3 The process takes a minute or so to complete, most of which is spent transcoding the video file to MP3. The resulting MP3s are usually 40 to 50 MB in size, which is very reasonable. The outcome: How it looks in practice Here’s what one of these “glanceable” podcasts looks like in Overcast and on my lock screen: Maggie Appleton presented this talk over two years ago and it’s been on my “talks to watch” list ever since. Once I put it in Overcast? I listened to it in less than a day. It’s not a lot of extra information, but enough that I can quickly glance down and get the gist of what a speaker is saying. Both views update with a new frame every few seconds, or I can put my phone in my pocket and ignore the screen. I’ve used this approach for half a dozen videos so far, and I’m happy with the results. I expect to keep using it, because I have a long queue of videos I’ve been meaning to watch. If you’d like to try this, check out glancecast for the full code and instructions. [If the formatting of this post looks odd in your feed reader, visit the original article]

I wrote a post for the Tailscale blog about a long-running series of corruption incidents, and how they eventually led us to find an SQLite bug that predates my entire programming career. I’m incredibly proud of this, both the work and the blog post. Before Tailscale, I was coming from smaller teams where I didn’t get to tackle problems of this scale or complexity. This was exactly the sort of tricky, deep technical challenge I wanted to be part of (though I’d rather it hadn’t been quite so stressful)! I’m glad I got to play a small part in these incidents, and I learnt so much from the more experienced engineers I worked with. I never want to hear the words “SQLite corruption” again, but if I do, I’d want to have Tailscalars at my side. Writing the blog post has a blast, too. The piece transformed from a rough draft into a solid, engaging piece of writing, thanks to thoughtful feedback from many people at Tailscale. Most of my writing is self-edited, and it’s always a pleasure to work with a dedicated editor. Please check out the blog post if you haven’t read it already – I think it’s a fascinating technical story, and one readers of this site are bound to enjoy. [If the formatting of this post looks odd in your feed reader, visit the original article]

Yesterday at work, a customer spotted a typo in our UI: “you can use the use the Tailscale CLI”. After the typo was fixed, I wanted to find other cases of accidentally repeated words or phrases. I used two regular expressions to search every codebase for unnecessary repetition. The first regex finds repeated words: \b([A-Za-z]+) \1\b Backfill product data from from Stripe Learn more about about inviting users Argument must be be one of host name, IP set name, IP prefix, or IP There’s a capturing group for a single word made up of letters ([A-Za-z]+), a space, then a backreference to the group. That expression is surrounded by word boundary assertions \b, which check that I’m at the start/end of a word – this avoids finding repeated character sequyences that within longer words, like “with the reason”. The second regex finds repeated phrases: \b([A-Za-z]+ [A-Za-z]+) \1\b Follow the steps in the in the "How to" section Log in to in to your account To configure federated identities federated identities using the Go SDK I’ve changed the capturing group, so now it looks for two words separated by a space. Sometimes repetition is useful, like when I really really went to emphasise a point, but often it’s just a typo. Cleaning up these mistakes has been a fun Friday cleanup task. [If the formatting of this post looks odd in your feed reader, visit the original article]