<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: FF</title>
    <description>The latest articles on DEV Community by FF (@easy28).</description>
    <link>https://dev.to/easy28</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4161477%2F8ff0391b-afab-4e0c-ad26-d39bc7443dc6.png</url>
      <title>DEV Community: FF</title>
      <link>https://dev.to/easy28</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/easy28"/>
    <language>en</language>
    <item>
      <title>SVG favicons: what browsers actually read in 2026 (and when you still need an ICO)</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Thu, 12 Nov 2026 23:51:53 +0000</pubDate>
      <link>https://dev.to/easy28/svg-favicons-what-browsers-actually-read-in-2026-and-when-you-still-need-an-ico-3maa</link>
      <guid>https://dev.to/easy28/svg-favicons-what-browsers-actually-read-in-2026-and-when-you-still-need-an-ico-3maa</guid>
      <description>&lt;p&gt;SVG favicons are the dream: one crisp vector file, scales forever, tiny payload. And in 2026 most of that dream works — but "most" is doing a lot of heavy lifting. Depending on which browser, which platform, and which non-browser tool fetches your icon, the SVG you shipped may be read, ignored, or replaced by a request for &lt;code&gt;/favicon.ico&lt;/code&gt; you never made. Here's what actually happens, and the link stack that keeps every consumer happy.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who reads your SVG favicon
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Chrome, Firefox, Edge&lt;/strong&gt; — all read &lt;code&gt;&amp;lt;link rel="icon" type="image/svg+xml" href="/favicon.svg"&amp;gt;&lt;/code&gt; and prefer it over raster alternatives when declared correctly. Bonus you can't get from PNG: SVG favicons can embed CSS with media queries, which is exactly how the &lt;a href="https://favicon.one/blog/dark-mode-favicon-safari/" rel="noopener noreferrer"&gt;dark-mode Safari favicon trick&lt;/a&gt; swaps colors without JavaScript.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Safari on macOS&lt;/strong&gt; — supports SVG favicons, with quirks: it historically ignored the &lt;code&gt;type&lt;/code&gt; attribute and picked icons by &lt;code&gt;sizes&lt;/code&gt; and order instead. Declaring both a PNG and an SVG without thinking about order leads to Safari grabbing the "wrong" one. If your icon looks right in Chrome and off in Safari, this is usually why.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;iOS Safari&lt;/strong&gt; — reads none of this. No ICO, no SVG. It only ever looks at &lt;code&gt;apple-touch-icon&lt;/code&gt; (180×180, no transparency). If your fancy SVG "doesn't work on iPhone," it was never going to.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Not-a-browser consumers&lt;/strong&gt; — feed readers, crawlers, chat-link previews, some enterprise tooling: a surprising number still ignore your HTML entirely and GET &lt;code&gt;/favicon.ico&lt;/code&gt; by convention. Ship no ICO and those tools show a blank square or log 404 noise on every request.&lt;/p&gt;

&lt;h2&gt;
  
  
  The link stack that works everywhere
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"/favicon.ico"&lt;/span&gt; &lt;span class="na"&gt;sizes&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"32x32"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"/favicon.svg"&lt;/span&gt; &lt;span class="na"&gt;type&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"image/svg+xml"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"apple-touch-icon"&lt;/span&gt; &lt;span class="na"&gt;href&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"/apple-touch-icon.png"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;ICO is the floor every consumer can fall back to, SVG is the upgrade modern browsers take, apple-touch-icon covers iOS. Two practical notes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;SVG with a bitmap floor beats SVG alone.&lt;/strong&gt; The ICO costs a few KB and silences the direct &lt;code&gt;/favicon.ico&lt;/code&gt; fetchers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep the SVG small.&lt;/strong&gt; Embedded fonts or giant path data blow up your document weight for a 16×16 render. Simplify paths before shipping — &lt;a href="https://favicon.one/blog/favicon-blurry/" rel="noopener noreferrer"&gt;why favicons go blurry and how vector fixes it&lt;/a&gt; covers the sizing math.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Converting: the boring part everyone skips wrong
&lt;/h2&gt;

&lt;p&gt;You have an SVG. You still need the ICO floor and the PNG ladder (16/32/192/512). Two wrong ways to do it: screenshot your icon and call it a day, or upscale a 32px export to 512px and ship mush. The right way is generating each raster size from the original vector:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;SVG → multi-size ICO directly in the browser: &lt;a href="https://favicon.one/convert/svg-to-ico/" rel="noopener noreferrer"&gt;svg-to-ico converter&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;PNG → ICO when your source is already raster: &lt;a href="https://favicon.one/convert/png-to-ico/" rel="noopener noreferrer"&gt;png-to-ico converter&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;No SVG yet, just text or an emoji idea: the &lt;a href="https://favicon.one/svg-favicon-generator/" rel="noopener noreferrer"&gt;SVG favicon generator&lt;/a&gt; and &lt;a href="https://favicon.one/emoji-to-favicon/" rel="noopener noreferrer"&gt;emoji-to-favicon&lt;/a&gt; build the vector from scratch — all client-side, nothing uploaded.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The full size-vs-format matrix (what each platform actually reads in 2026) is in the &lt;a href="https://favicon.one/blog/svg-vs-png-favicon/" rel="noopener noreferrer"&gt;SVG vs PNG favicon breakdown&lt;/a&gt; — worth bookmarking next to your deploy checklist.&lt;/p&gt;

&lt;h2&gt;
  
  
  Verify before you blame the browser
&lt;/h2&gt;

&lt;p&gt;The fastest way to lose an hour is arguing with Safari about an icon your page never actually declared. Run the deployed URL through a &lt;a href="https://favicon.one/favicon-checker/" rel="noopener noreferrer"&gt;favicon checker&lt;/a&gt; — it fetches the live page, lists every declared icon, and shows what each one resolves to. If your SVG isn't in that list, the problem is your &lt;code&gt;&amp;lt;link&amp;gt;&lt;/code&gt; tag, not the browser.&lt;/p&gt;

&lt;p&gt;That's the whole game: vector for modern browsers, ICO floor for everything else, apple-touch for iOS, and a checker pass before you blame anyone. Ten minutes, once per project.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>html</category>
      <category>css</category>
      <category>performance</category>
    </item>
    <item>
      <title>Your video calls freeze on upload — the number your speed test hides</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Sun, 08 Nov 2026 01:40:00 +0000</pubDate>
      <link>https://dev.to/easy28/your-video-calls-freeze-on-upload-the-number-your-speed-test-hides-129l</link>
      <guid>https://dev.to/easy28/your-video-calls-freeze-on-upload-the-number-your-speed-test-hides-129l</guid>
      <description>&lt;p&gt;Your internet "feels fast." Download tests glow green. And yet, twice a week, your video call turns into a wax museum the moment a cloud backup, sync client, or large email attachment starts moving.&lt;/p&gt;

&lt;p&gt;Here's the thing almost everyone misses: &lt;strong&gt;a video call is an upload-heavy application.&lt;/strong&gt; Your camera stream has to travel &lt;em&gt;out&lt;/em&gt;. Download speed — the number every ISP advertises and every speed test leads with — is nearly irrelevant to how you look and sound to the other side.&lt;/p&gt;

&lt;h2&gt;
  
  
  What actually kills the call
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Upload saturation.&lt;/strong&gt; Most home plans are asymmetric — 500 Mbps down, 20 up. Your call needs ~3 Mbps of steady upload. The moment OneDrive decides to sync 10 GB, that 20 Mbps pipe is full, and your face is queued behind it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bufferbloat.&lt;/strong&gt; When the upload pipe saturates, latency on the &lt;em&gt;whole&lt;/em&gt; connection climbs — often from 10 ms to 300+ ms. That's why your screen share stutters even though "nothing else is happening."&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Jitter on the upstream.&lt;/strong&gt; Wi-Fi retransmissions plus a competing upload means your packets arrive in bursts. Codecs hate bursts more than they hate delay.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  The 25-second diagnosis
&lt;/h2&gt;

&lt;p&gt;Run a test that measures latency &lt;strong&gt;per direction while the line is saturated&lt;/strong&gt; — not just idle ping. &lt;a href="https://speedpulse.tech" rel="noopener noreferrer"&gt;SpeedPulse&lt;/a&gt; (free, no signup, no app, no cookies, all client-side — full disclosure: I built it) does exactly this:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;it keeps a &lt;strong&gt;ping train running during both the download and upload phases&lt;/strong&gt;, then reports &lt;code&gt;loaded − idle&lt;/code&gt; latency for each direction separately&lt;/li&gt;
&lt;li&gt;grades the worse direction A+ to F on the community bufferbloat scale&lt;/li&gt;
&lt;li&gt;measures &lt;strong&gt;jitter&lt;/strong&gt; (mean absolute difference between consecutive pings)&lt;/li&gt;
&lt;li&gt;translates results into plain-English verdicts — one of them is specifically for &lt;strong&gt;video calls&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;cross-checks &lt;strong&gt;8 independent CDN edges&lt;/strong&gt; afterward, so you can see if your ISP flatters the speed-test server itself&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If your download direction grades A but upload grades C or worse — congratulations, you found exactly why your camera freezes during backups.&lt;/p&gt;

&lt;h2&gt;
  
  
  The fixes
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Cap your upload ~10–20% below your plan rate&lt;/strong&gt; in your router's QoS/SQM settings. On OpenWrt, SQM with &lt;code&gt;cake&lt;/code&gt; does this in two minutes. The backup still finishes (it just takes 10% longer), and your calls stop dying.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Schedule sync-heavy apps&lt;/strong&gt; (Backblaze, iCloud, Google Drive) outside meeting hours, or throttle them in their own settings.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Wire the workstation.&lt;/strong&gt; One less source of upstream jitter you can't control.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If you can only fix one direction, fix upload.&lt;/strong&gt; It's the thin end of every asymmetric connection.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The counter-intuitive part: making your internet &lt;em&gt;slightly slower on purpose&lt;/em&gt; is what makes calls feel instant. Queue discipline beats raw bandwidth for real-time apps — every time.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why most speed tests won't show you this
&lt;/h2&gt;

&lt;p&gt;Idle ping and download throughput are easy; &lt;strong&gt;loaded upload latency&lt;/strong&gt; requires saturating your upstream while pinging through it, per direction, without ICMP. In a browser you time HTTPS round trips with &lt;code&gt;fetch(url, { mode: 'no-cors' })&lt;/code&gt; and &lt;code&gt;performance.now()&lt;/code&gt; — opaque responses, but full network round trips, which is all timing needs. If you're curious about the exact methodology (and the limitations we're honest about): &lt;a href="https://speedpulse.tech/blog/how-speedpulse-measures-jitter-and-bufferbloat/" rel="noopener noreferrer"&gt;How SpeedPulse measures jitter, bufferbloat and 8 CDN latencies&lt;/a&gt;.&lt;/p&gt;




&lt;p&gt;More background if you want it: &lt;a href="https://speedpulse.tech/blog/internet-speed-for-video-calls/" rel="noopener noreferrer"&gt;how much internet speed video calls actually need&lt;/a&gt; · &lt;a href="https://speedpulse.tech/blog/what-is-bufferbloat/" rel="noopener noreferrer"&gt;what bufferbloat is&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Run the test before your next meeting and check the &lt;strong&gt;upload direction grade&lt;/strong&gt; — drop it in the comments if it surprises you: &lt;a href="https://speedpulse.tech" rel="noopener noreferrer"&gt;speedpulse.tech&lt;/a&gt;&lt;/p&gt;

</description>
      <category>networking</category>
      <category>productivity</category>
      <category>performance</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Your ping is fine — your jitter isn't. The number that ruins your aim</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Sun, 08 Nov 2026 00:10:44 +0000</pubDate>
      <link>https://dev.to/easy28/your-ping-is-fine-your-jitter-isnt-the-number-that-ruins-your-aim-1ai8</link>
      <guid>https://dev.to/easy28/your-ping-is-fine-your-jitter-isnt-the-number-that-ruins-your-aim-1ai8</guid>
      <description>&lt;p&gt;Your speed test says 15 ms ping. Your game feels like it's running through syrup anyway: shots that clearly connected don't register, your character teleports half a step, peeker's advantage feels rigged in every fight. Nothing is "wrong" — except one number your speed test never showed you.&lt;/p&gt;

&lt;h2&gt;
  
  
  Ping is an average. Games live in the spikes.
&lt;/h2&gt;

&lt;p&gt;Ping (latency) tells you how long one round trip takes. &lt;strong&gt;Jitter measures how much that time *wobbles&lt;/strong&gt;* between consecutive packets — and real-time games are far more sensitive to wobble than to the average.&lt;/p&gt;

&lt;p&gt;Game netcode runs on assumptions. Interpolation buffers, rollback windows, hit registration timing — all sized around "packets arrive every ~15 ms, give or take 2". When jitter hits 40–80 ms, packets arrive in bursts: 5 ms, 90 ms, 12 ms, 120 ms. Your opponent's movement model gets starved, then flooded. The server sees a different fight than you do. That's the "I shot him first" death recap.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where the wobble comes from:&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Wi-Fi&lt;/strong&gt; — retransmissions, airtime contention from every other device in the house, power-save bursts. Wired vs Wi-Fi is routinely a 3–10× jitter difference.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bufferbloat&lt;/strong&gt; — when someone in the house downloads/uploads, your packets sit in a bloated router queue &lt;em&gt;sometimes&lt;/em&gt; (that "why is it only bad at 7 pm" pattern). This is jitter with a schedule.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cheap ISP routing&lt;/strong&gt; — congestion on the path, or peering games between your ISP and the game server's network.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Measure it in 25 seconds
&lt;/h2&gt;

&lt;p&gt;Any speed test shows you ping. Very few show you jitter and latency-under-load. &lt;a href="https://speedpulse.tech" rel="noopener noreferrer"&gt;SpeedPulse&lt;/a&gt; — which I built, free, no signup, no app, no cookies, all client-side — measures in one run:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Idle ping&lt;/strong&gt; (median of 8 samples — one DNS hiccup won't skew it)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Jitter&lt;/strong&gt; — the RFC 3550-style mean absolute difference between consecutive ping samples&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bufferbloat&lt;/strong&gt; — a ping train keeps running &lt;em&gt;while the line is saturated&lt;/em&gt; by download and upload streams; the tool reports &lt;code&gt;loaded − idle&lt;/code&gt; latency per direction and grades the worse one A+ to F&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Plain-English verdicts&lt;/strong&gt; for gaming, streaming, calls and browsing&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;An 8-CDN cross-check&lt;/strong&gt; — after the test it pings 8 independent CDN edges (Cloudflare, Fastly, Google, GitHub, AWS…). If the speed-test server is dramatically faster than the pack, your ISP may be treating test traffic specially — the classic "test is fine but my game is not" explanation&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  What the numbers should look like
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Good&lt;/th&gt;
&lt;th&gt;Hurts&lt;/th&gt;
&lt;th&gt;Broken&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Idle ping&lt;/td&gt;
&lt;td&gt;&amp;lt; 30 ms&lt;/td&gt;
&lt;td&gt;50–80 ms&lt;/td&gt;
&lt;td&gt;100 ms+&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Jitter&lt;/td&gt;
&lt;td&gt;&amp;lt; 5 ms&lt;/td&gt;
&lt;td&gt;15–30 ms&lt;/td&gt;
&lt;td&gt;40 ms+&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Bufferbloat grade&lt;/td&gt;
&lt;td&gt;A / A+&lt;/td&gt;
&lt;td&gt;C / D&lt;/td&gt;
&lt;td&gt;F&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;If your ping looks fine but jitter reads 30 ms+, you've found the ghost in your aim. (Grade thresholds follow the Waveform / bufferbloat.org scale so results compare across tools.)&lt;/p&gt;

&lt;h2&gt;
  
  
  The fixes, in order of effectiveness
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Go wired.&lt;/strong&gt; Ethernet is the single biggest jitter fix that exists. If you can't run cable, at least move the work device to 5 GHz and away from the microwave.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Enable SQM (fq_codel/CAKE) on the router.&lt;/strong&gt; Trivial on OpenWrt; many stock firmwares have a QoS page. It keeps buffers short so latency stays flat even at full line utilization — the cure for the 7 pm problem.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cap upload slightly below your plan.&lt;/strong&gt; Upstream bloats first; your outgoing game packets are the casualties.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Rule out ISP-side path issues.&lt;/strong&gt; Compare jitter to multiple destinations, not just the test server — the 8-CDN panel is a cheap proxy for this. Persistently bad across all edges = your ISP's routing, worth a support ticket with numbers attached.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  For the nerds: how you measure jitter without ICMP
&lt;/h2&gt;

&lt;p&gt;Browsers can't send ICMP, but they can time HTTPS round trips: &lt;code&gt;fetch(url, { mode: 'no-cors' })&lt;/code&gt; completes a full DNS+TCP+TLS+HTTP round trip even though the response is opaque. Sample it 8 times, take the mean absolute difference between consecutive samples — that's jitter. Latency uses the median; jitter uses the consecutive differences (deviation-from-mean hides bursty behavior, which is exactly the thing that kills game feel). Full methodology write-up, limitations included: &lt;a href="https://speedpulse.tech/blog/how-speedpulse-measures-jitter-and-bufferbloat/" rel="noopener noreferrer"&gt;How SpeedPulse measures jitter, bufferbloat and 8 CDN latencies&lt;/a&gt;.&lt;/p&gt;




&lt;p&gt;If you want the deeper background: &lt;a href="https://speedpulse.tech/blog/what-is-jitter/" rel="noopener noreferrer"&gt;what jitter is, in plain English&lt;/a&gt; · &lt;a href="https://speedpulse.tech/blog/good-ping-for-gaming/" rel="noopener noreferrer"&gt;what a good ping for gaming actually is&lt;/a&gt; · &lt;a href="https://speedpulse.tech/blog/how-to-fix-high-ping/" rel="noopener noreferrer"&gt;how to fix high ping&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Run the test on your own line and drop your jitter number in the comments: &lt;a href="https://speedpulse.tech" rel="noopener noreferrer"&gt;speedpulse.tech&lt;/a&gt; — happy to help interpret.&lt;/p&gt;

</description>
      <category>networking</category>
      <category>gamedev</category>
      <category>performance</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Your JSON isn't broken — your strings are: escapes, BOMs and the 2^53 lie</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Fri, 06 Nov 2026 15:35:42 +0000</pubDate>
      <link>https://dev.to/easy28/your-json-isnt-broken-your-strings-are-escapes-boms-and-the-253-lie-115b</link>
      <guid>https://dev.to/easy28/your-json-isnt-broken-your-strings-are-escapes-boms-and-the-253-lie-115b</guid>
      <description>&lt;p&gt;A few weeks ago we published a 30-second audit for spotting JSON tools that leak your payloads. This post is about the other kind of pain: the times JSON "breaks" and it's not the JSON's fault at all — it's the strings, the invisible characters, or a number that silently lied to you.&lt;/p&gt;

&lt;p&gt;These are the four failure modes we see most, and what to actually do about them.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. There are only eight legal escapes — everything else is a landmine
&lt;/h2&gt;

&lt;p&gt;JSON strings allow exactly: \" \ \/ \b \f \n \r \t and \uXXXX. That's it.&lt;/p&gt;

&lt;p&gt;Two gotchas hide in plain sight:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Astral characters need surrogate pairs.&lt;/strong&gt; &lt;code&gt;\uXXXX&lt;/code&gt; is one UTF-16 code unit. To escape 𝄞 (U+1D11E) you need &lt;code&gt;\uD834\uDD1E&lt;/code&gt;. A lone surrogate like &lt;code&gt;\uD834&lt;/code&gt; is invalid JSON — yet some parsers accept it and crash later, far from the cause.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Real newlines and tabs are illegal inside strings.&lt;/strong&gt; Pasting a literal tab from a terminal output, or a multi-line string from a code editor, breaks strict parsers with a confusing "Unexpected token".&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Test escapes round-trip instead of eyeballing: the &lt;a href="https://jsonformatter.app/tools/json-escape" rel="noopener noreferrer"&gt;JSON escape tool&lt;/a&gt; has a minimal mode and an ASCII-safe &lt;code&gt;\uXXXX&lt;/code&gt; mode, and the &lt;a href="https://jsonformatter.app/tools/json-unescape" rel="noopener noreferrer"&gt;unescape tool&lt;/a&gt; validates what comes out instead of silently returning garbage.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. The invisible characters: BOM, U+00A0, and friends
&lt;/h2&gt;

&lt;p&gt;A file that looks perfectly fine in your editor but fails with &lt;code&gt;Unexpected token ﻿&lt;/code&gt; is almost always a &lt;strong&gt;UTF-8 BOM&lt;/strong&gt;. The spec (RFC 8259) says producers must not add one — and explicitly lets parsers ignore it. Plenty of parsers don't.&lt;/p&gt;

&lt;p&gt;Similarly, content copied from PDFs or Word documents picks up typographic quotes, en-dashes and non-breaking spaces (&lt;code&gt;\u00A0&lt;/code&gt;) that are invisible in most editors but structurally fatal. A tree viewer that can display raw escapes makes these obvious in seconds — our &lt;a href="https://jsonformatter.app/tools/json-viewer" rel="noopener noreferrer"&gt;JSON viewer&lt;/a&gt; does exactly that, and it all runs in your browser.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. The 2^53 lie: when parsing silently changes your data
&lt;/h2&gt;

&lt;p&gt;This is the nastiest one because &lt;strong&gt;there's no error&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;JavaScript numbers are IEEE-754 doubles, so any integer above &lt;code&gt;2^53 - 1&lt;/code&gt; (9007199254740991) can't be represented exactly. Parse &lt;code&gt;9007199254740993&lt;/code&gt; with &lt;code&gt;JSON.parse&lt;/code&gt; and you silently get &lt;code&gt;9007199254740992&lt;/code&gt;. No exception. No warning.&lt;/p&gt;

&lt;p&gt;Where this bites in real life:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Snowflake IDs (Twitter, Discord) — deliberately generated above 2^53&lt;/li&gt;
&lt;li&gt;Payment/transaction IDs from payment providers&lt;/li&gt;
&lt;li&gt;Millisecond timestamps far into the future (rare, but possible)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The fix is boring but non-negotiable: big IDs travel as &lt;strong&gt;strings&lt;/strong&gt;. If you're auditing an existing integration, paste a real response into a &lt;a href="https://jsonformatter.app/tools/json-validator" rel="noopener noreferrer"&gt;strict validator&lt;/a&gt; — ours flags lossy integers before they reach your code.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. JSON-in-JSON: the double-escape onion
&lt;/h2&gt;

&lt;p&gt;The most confusing sight in a log: a value that looks like &lt;code&gt;"{\"a\": 1}"&lt;/code&gt;. That's a JSON document serialized as a &lt;em&gt;string inside&lt;/em&gt; another JSON document — logs, webhooks and message queues do this constantly.&lt;/p&gt;

&lt;p&gt;Strip one level of escaping, then parse again. For real payloads use something like &lt;code&gt;fromjson&lt;/code&gt; in &lt;a href="https://jsonformatter.app/tools/jq-online" rel="noopener noreferrer"&gt;jq running in your browser&lt;/a&gt; — no terminal, no uploads, described in &lt;a href="https://jsonformatter.app/news/json-transform-jq-jsonpath" rel="noopener noreferrer"&gt;our post on client-side jq&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why we care so much about this
&lt;/h2&gt;

&lt;p&gt;Every tool linked above is 100% client-side: paste your payload, and nothing leaves your browser tab. After the CodeBeautify leak (user-pasted payloads exposed through a public feed), "the tool never sees your data" stopped being a nice-to-have for us — it's the product. The full story is in &lt;a href="https://jsonformatter.app/news/codebeautify-data-leak-explained" rel="noopener noreferrer"&gt;the leak explained&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If you want the reference version of all of this — the full escape table, the JSON vs JSON5 vs JSONC vs NDJSON comparison, and the base64 alphabet traps — we put together &lt;a href="https://easy28.github.io/json-debugging-checklist/json-cheatsheet.html" rel="noopener noreferrer"&gt;a one-page cheat sheet&lt;/a&gt; you can keep open next to your editor.&lt;/p&gt;

&lt;p&gt;What's the weirdest JSON failure you've debugged? Wrong-position &lt;code&gt;Unexpected token&lt;/code&gt; stories especially welcome.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>json</category>
      <category>tutorial</category>
      <category>programming</category>
    </item>
    <item>
      <title>Your AI-Generated SVG Won't Render: The Same 8 Defects Every Time</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Mon, 02 Nov 2026 00:45:19 +0000</pubDate>
      <link>https://dev.to/easy28/your-ai-generated-svg-wont-render-the-same-8-defects-every-time-5e66</link>
      <guid>https://dev.to/easy28/your-ai-generated-svg-wont-render-the-same-8-defects-every-time-5e66</guid>
      <description>&lt;p&gt;You asked a model for an SVG. The code &lt;em&gt;looks&lt;/em&gt; right — tags, paths, colors, all plausible. You paste it into a file, open it in a browser, and get... a blank rectangle. No error, no warning, nothing.&lt;/p&gt;

&lt;p&gt;If this happens to you regularly, it's not bad luck. SVG is XML, and XML is unforgiving in a way HTML never is. A browser reading HTML will patch over a missing quote or an unclosed tag and render what it can. A browser reading SVG does none of that: one syntax slip and the whole document is rejected — usually silently.&lt;/p&gt;

&lt;p&gt;Language models generate SVG the way they generate everything else: token by token, plausibly. They pattern-match on millions of &lt;em&gt;examples&lt;/em&gt; without running an XML validator or a geometry engine. So they fail on exactly the details a validator would have caught — and they fail in the same ways, over and over. Here are the eight defects behind almost every broken AI-generated SVG I've seen.&lt;/p&gt;

&lt;h2&gt;
  
  
  The 8 defects
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;1. Missing &lt;code&gt;xmlns&lt;/code&gt; attribute.&lt;/strong&gt; The #1 killer. Without &lt;code&gt;xmlns="http://www.w3.org/2000/svg"&lt;/code&gt; the browser treats the file as generic XML. It may render inline in an HTML page, then fail as a standalone file or inside an &lt;code&gt;&amp;lt;img&amp;gt;&lt;/code&gt; tag — which makes it look like a caching bug. Models forget it constantly.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Missing or invented &lt;code&gt;viewBox&lt;/code&gt;.&lt;/strong&gt; No &lt;code&gt;viewBox&lt;/code&gt;, and the browser has no idea which part of the coordinate system to show. Or the model writes &lt;code&gt;viewBox="0 0 24 24"&lt;/code&gt; and then draws at &lt;code&gt;x="150"&lt;/code&gt; — every shape lands off-canvas and you get an empty frame.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Shapes translated out of view.&lt;/strong&gt; AI loves decorative &lt;code&gt;transform="translate(350, 280)"&lt;/code&gt; guesswork. The markup is valid, the preview is blank, and nothing tells you the artwork is sitting just outside the frame.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. A raw &lt;code&gt;&amp;amp;&lt;/code&gt; in a label.&lt;/strong&gt; SVG is strict XML. A label like &lt;code&gt;Tom &amp;amp; Jerry&lt;/code&gt; with a bare unescaped &lt;code&gt;&amp;amp;&lt;/code&gt; invalidates the &lt;em&gt;entire document&lt;/em&gt;. One character, whole file dead.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. NaN and Infinity coordinates.&lt;/strong&gt; Models do arithmetic in public: &lt;code&gt;d="M 10 10 L NaN 50"&lt;/code&gt; or a radius of &lt;code&gt;Infinity&lt;/code&gt;. The path is syntactically present and renders as nothing.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;6. Unclosed or misnested tags.&lt;/strong&gt; Truncated output, a &lt;code&gt;&amp;lt;g&amp;gt;&lt;/code&gt; that never closes, an attribute missing its quote. XML parsers are all-or-nothing — one structural slip and nothing renders at all.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;7. Chat prose and code fences.&lt;/strong&gt; "Sure! Here's your logo:" ... &lt;code&gt;&lt;/code&gt;`&lt;code&gt;svg&lt;/code&gt; ... "Let me know if you'd like adjustments!" Paste that into a file and you have prose, fences, and markup layered together. Browsers don't strip any of it for you.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;8. Everything &lt;code&gt;fill="none"&lt;/code&gt;.&lt;/strong&gt; The model styles each shape with &lt;code&gt;fill="none"&lt;/code&gt; and forgets to add a stroke. The file parses, the shapes exist, the image is empty. This one fools people the most because the SVG technically "works".&lt;/p&gt;

&lt;p&gt;Defects 1–6 kill the file outright at parse time. 7–8 are sneakier: the file &lt;em&gt;renders&lt;/em&gt;, it just renders empty — worse to debug because there's no error message to search for.&lt;/p&gt;

&lt;h2&gt;
  
  
  The 30-second diagnosis
&lt;/h2&gt;

&lt;p&gt;Before fixing anything, figure out which defect you have:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Open the file and search for &lt;code&gt;xmlns&lt;/code&gt;. Missing? That's defect 1.&lt;/li&gt;
&lt;li&gt;Check the &lt;code&gt;viewBox&lt;/code&gt; against the largest coordinates in the file. Drawing at 400 but viewing 0–24? Defect 2 or 3.&lt;/li&gt;
&lt;li&gt;Look for &lt;code&gt;NaN&lt;/code&gt;, &lt;code&gt;Infinity&lt;/code&gt;, or a bare &lt;code&gt;&amp;amp;&lt;/code&gt; outside entities. Defects 4 and 5.&lt;/li&gt;
&lt;li&gt;Text at the top of the file that isn't markup? Defect 7.&lt;/li&gt;
&lt;li&gt;Everything has &lt;code&gt;fill="none"&lt;/code&gt; and no &lt;code&gt;stroke&lt;/code&gt;? Defect 8.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Or skip the manual pass: paste the raw chat output — prose and fences and all — into &lt;a href="https://svgeditor.dev/svg-repair" rel="noopener noreferrer"&gt;SVG Repair&lt;/a&gt;. It extracts the real markup, repairs the XML, adds the missing &lt;code&gt;xmlns&lt;/code&gt; and &lt;code&gt;viewBox&lt;/code&gt;, zeroes out NaN coordinates, expands the canvas to include off-screen shapes, makes invisible shapes visible, and shows you an itemized report of every defect next to a before/after preview. It actually renders the repaired file and checks it paints a non-empty image before claiming success. Everything runs client-side — the file never leaves your browser.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prevent it at the source
&lt;/h2&gt;

&lt;p&gt;Repairing is faster than regenerating, but the cheapest fix is an SVG that isn't broken. Four lines added to the prompt eliminate most defects:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;"Output only the raw SVG markup, no explanation, no code fences." — kills defect 7 entirely.&lt;/li&gt;
&lt;li&gt;"Include &lt;code&gt;xmlns&lt;/code&gt; and a &lt;code&gt;viewBox&lt;/code&gt; that matches the coordinate range you draw in." — kills defects 1 and 2.&lt;/li&gt;
&lt;li&gt;"Draw all shapes inside &lt;code&gt;0 0 512 512&lt;/code&gt;. Use no transforms." — kills defect 3.&lt;/li&gt;
&lt;li&gt;"No &lt;code&gt;&amp;lt;script&amp;gt;&lt;/code&gt;, no CSS classes, no text — convert text to paths." — kills the surprise below.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  One more thing AI SVGs smuggle in: scripts
&lt;/h2&gt;

&lt;p&gt;Models occasionally emit &lt;code&gt;&amp;lt;script&amp;gt;&lt;/code&gt; tags, &lt;code&gt;onload&lt;/code&gt; handlers, or &lt;code&gt;foreignObject&lt;/code&gt; blocks — sometimes because you asked for animation, sometimes out of nowhere. That matters beyond rendering: an SVG carrying scripts becomes a live XSS vector the moment someone opens it directly or inlines it into a page. If you're accepting SVGs from users, or embedding AI output into anything public, strip active content before it ships. The repair pipeline above strips it by default.&lt;/p&gt;

&lt;h2&gt;
  
  
  When it renders again, keep going
&lt;/h2&gt;

&lt;p&gt;A repaired file is a working file, not a finished one. AI output carries redundant precision and empty groups that a proper &lt;a href="https://svgeditor.dev/svg-optimizer" rel="noopener noreferrer"&gt;SVG optimizer&lt;/a&gt; typically shaves 30–60%. Whole icon sets can be &lt;a href="https://svgeditor.dev/batch" rel="noopener noreferrer"&gt;repaired and converted in batch&lt;/a&gt;. And when the geometry is &lt;em&gt;almost&lt;/em&gt; right, a &lt;a href="https://svgeditor.dev/" rel="noopener noreferrer"&gt;free browser-based SVG editor&lt;/a&gt; lets you fix it visually — select and move shapes, edit text, adjust fills and strokes — without installing anything or uploading the file anywhere.&lt;/p&gt;

&lt;p&gt;The full write-up with more prompt examples is here: &lt;a href="https://svgeditor.dev/blog/ai-generated-svg-not-rendering" rel="noopener noreferrer"&gt;Why your AI-generated SVG doesn't render&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>ai</category>
      <category>tutorial</category>
      <category>chatgpt</category>
    </item>
    <item>
      <title>The 2026 Favicon Checklist: every size your site actually needs</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Sun, 01 Nov 2026 10:36:52 +0000</pubDate>
      <link>https://dev.to/easy28/the-2026-favicon-checklist-every-size-your-site-actually-needs-31m3</link>
      <guid>https://dev.to/easy28/the-2026-favicon-checklist-every-size-your-site-actually-needs-31m3</guid>
      <description>&lt;p&gt;A few years ago "add a favicon" meant dropping one &lt;code&gt;favicon.ico&lt;/code&gt; in your site root. In 2026 that same site will look broken in at least four places: blurry in a retina browser tab, a generic globe in Google search results, a white square when an Android user saves it to their home screen, and a black blob on an iPhone.&lt;/p&gt;

&lt;p&gt;The fix isn't hard — but the requirements have quietly multiplied. Here's the complete size set that actually matters in 2026, why each one exists, and a copy-paste HTML block at the end.&lt;/p&gt;

&lt;h2&gt;
  
  
  The 2026 favicon checklist
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Size&lt;/th&gt;
&lt;th&gt;File&lt;/th&gt;
&lt;th&gt;Where it's used&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;16×16&lt;/td&gt;
&lt;td&gt;&lt;code&gt;favicon-16x16.png&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Classic browser tabs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;32×32&lt;/td&gt;
&lt;td&gt;&lt;code&gt;favicon-32x32.png&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Retina tabs, Windows taskbar&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;48×48&lt;/td&gt;
&lt;td&gt;inside &lt;code&gt;favicon.ico&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Windows site icons, older browsers&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;180×180&lt;/td&gt;
&lt;td&gt;&lt;code&gt;apple-touch-icon.png&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;iOS home screen, Safari pinned tabs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;192×192&lt;/td&gt;
&lt;td&gt;&lt;code&gt;android-chrome-192x192.png&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Android home screen, Chrome PWA&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;512×512&lt;/td&gt;
&lt;td&gt;&lt;code&gt;android-chrome-512x512.png&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;PWA splash screen&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;vector&lt;/td&gt;
&lt;td&gt;&lt;code&gt;favicon.svg&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Modern browsers, dark-mode adaptive icons&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Six files. Not optional if you care about how your site presents on mobile and in search.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why each size survives
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;16×16 + 32×32 — the browser tab.&lt;/strong&gt; 16×16 is the legacy default; 32×32 is what actually renders on high-DPI displays. Ship both. A tab icon that's only 16×16 gets smeared by the browser's upscaler on any modern laptop.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;48×48 — inside the .ico.&lt;/strong&gt; The &lt;code&gt;.ico&lt;/code&gt; container can bundle multiple sizes in one file (16/32/48 is the common bundle), which is why a well-built &lt;code&gt;favicon.ico&lt;/code&gt; is still the most compatible single file you can ship. Windows site pins and plenty of crawlers read it first.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;180×180 — Apple's tax.&lt;/strong&gt; iOS ignores your favicon entirely. When a user taps "Add to Home Screen", Safari reads exactly &lt;code&gt;apple-touch-icon.png&lt;/code&gt; at exactly 180×180. One more trap: iOS composites transparency onto black — export on your brand background or your logo gets a black box.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;192×192 + 512×512 — Android &amp;amp; PWA.&lt;/strong&gt; These are read from &lt;code&gt;site.webmanifest&lt;/code&gt;. 192 is the Android home-screen icon; 512 powers the full-screen splash when your PWA launches. Ship a 512 source and downscale — never upscale, it shows.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;SVG — the modern bonus.&lt;/strong&gt; Every evergreen browser takes &lt;code&gt;favicon.svg&lt;/code&gt;, it scales everywhere, and it can embed a &lt;code&gt;prefers-color-scheme&lt;/code&gt; media query so your icon adapts to dark browser themes (worth its own post).&lt;/p&gt;

&lt;h2&gt;
  
  
  The Google search angle people forget
&lt;/h2&gt;

&lt;p&gt;Google shows a favicon next to your result on mobile (and increasingly desktop). If your icon declaration is broken or too small, Google shows a generic globe — and search results with a real icon get measurably more clicks than the globe. Google recommends a square image ≥48×48, ideally multiples of 48, reachable at the root. This is one of the few branding fixes that directly affects CTR.&lt;/p&gt;

&lt;h2&gt;
  
  
  The one source-of-truth rule
&lt;/h2&gt;

&lt;p&gt;Maintain &lt;strong&gt;one square master ≥512×512&lt;/strong&gt; (1024×1024 ideal) and derive everything from it. Hand-maintained icon folders are where stale sizes and 404s come from — treat favicon files as build output, not design artifacts.&lt;/p&gt;

&lt;h2&gt;
  
  
  Copy-paste block
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/favicon.ico"&lt;/span&gt; &lt;span class="na"&gt;sizes=&lt;/span&gt;&lt;span class="s"&gt;"48x48"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/favicon.svg"&lt;/span&gt; &lt;span class="na"&gt;type=&lt;/span&gt;&lt;span class="s"&gt;"image/svg+xml"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/favicon-16x16.png"&lt;/span&gt; &lt;span class="na"&gt;sizes=&lt;/span&gt;&lt;span class="s"&gt;"16x16"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/favicon-32x32.png"&lt;/span&gt; &lt;span class="na"&gt;sizes=&lt;/span&gt;&lt;span class="s"&gt;"32x32"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"apple-touch-icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/apple-touch-icon.png"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"manifest"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/site.webmanifest"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;meta&lt;/span&gt; &lt;span class="na"&gt;name=&lt;/span&gt;&lt;span class="s"&gt;"theme-color"&lt;/span&gt; &lt;span class="na"&gt;content=&lt;/span&gt;&lt;span class="s"&gt;"#ffffff"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  The 30-second audit
&lt;/h2&gt;

&lt;p&gt;Not sure what your site currently serves? Two moves:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Check what's declared vs. what resolves&lt;/strong&gt; — the &lt;a href="https://favicon.one/favicon-checker/" rel="noopener noreferrer"&gt;Favicon Checker&lt;/a&gt; reads your live page and reports every icon it finds.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Generate the full set from one image&lt;/strong&gt; — upload one square PNG to &lt;a href="https://favicon.one/" rel="noopener noreferrer"&gt;favicon.one&lt;/a&gt; and get a ZIP with every size above, the &lt;code&gt;.ico&lt;/code&gt;, the SVG, the manifest, and ready-to-paste HTML. Free, no signup, all in-browser.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If you only have a JPG or an odd-size PNG, the &lt;a href="https://favicon.one/convert/png-to-ico/" rel="noopener noreferrer"&gt;PNG → ICO converter&lt;/a&gt; handles the legacy part.&lt;/p&gt;

&lt;p&gt;Complete reference with per-platform details: &lt;a href="https://favicon.one/blog/favicon-size-guide/" rel="noopener noreferrer"&gt;Favicon Size Guide: Every Dimension You Need in 2026&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>html</category>
      <category>performance</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>I stopped pasting API responses into online JSON formatters — here's the 30-second audit I run instead</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Sun, 01 Nov 2026 01:44:28 +0000</pubDate>
      <link>https://dev.to/easy28/i-stopped-pasting-api-responses-into-online-json-formatters-heres-the-30-second-audit-i-run-2774</link>
      <guid>https://dev.to/easy28/i-stopped-pasting-api-responses-into-online-json-formatters-heres-the-30-second-audit-i-run-2774</guid>
      <description>&lt;p&gt;In November 2025, watchTowr Labs disclosed that CodeBeautify.org — one of the largest developer-tool sites on the internet — had been &lt;strong&gt;exposing user-pasted data through a public "Recent Links" feed&lt;/strong&gt;. Not a hypothetical: API keys, SSH private keys, and cloud credentials that developers pasted in to "just format this quickly" were reachable by anyone who knew where to look.&lt;/p&gt;

&lt;p&gt;If you've ever pasted a production payload into an online formatter, this one's for you.&lt;/p&gt;

&lt;h2&gt;
  
  
  The uncomfortable truth about "online tools"
&lt;/h2&gt;

&lt;p&gt;Most web-based formatters are not actually processing your data in your browser. Many POST your payload to a backend, format it server-side, and hand it back. Some store pastes to make "shareable links" — and shareable means enumerable.&lt;/p&gt;

&lt;p&gt;We put together a full technical breakdown of what happened, what was exposed, and the mechanics behind it:&lt;br&gt;
&lt;strong&gt;&lt;a href="https://jsonformatter.app/news/codebeautify-data-leak-explained" rel="noopener noreferrer"&gt;The CodeBeautify data leak, explained&lt;/a&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The 30-second audit (run this on any web tool)
&lt;/h2&gt;

&lt;p&gt;Before trusting any online tool with a real payload:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Open DevTools → &lt;strong&gt;Network&lt;/strong&gt; tab, filter to Fetch/XHR&lt;/li&gt;
&lt;li&gt;Clear the log&lt;/li&gt;
&lt;li&gt;Paste your JSON and hit format&lt;/li&gt;
&lt;li&gt;Watch the network panel&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;Any request whose payload contains your data = it left your machine.&lt;/strong&gt; Doesn't matter what the landing page says about privacy. The network panel doesn't lie.&lt;/p&gt;

&lt;p&gt;Second check: load the page, turn off Wi-Fi (or use DevTools' offline throttling), and see if the tool still works. A genuinely client-side tool keeps working with zero connectivity. One that "phone home" dies on the spot.&lt;/p&gt;

&lt;h2&gt;
  
  
  What a real client-side pipeline looks like
&lt;/h2&gt;

&lt;p&gt;For the past year I've been using &lt;a href="https://jsonformatter.app" rel="noopener noreferrer"&gt;JSON Formatter App&lt;/a&gt; precisely because you can do this audit on it and come up clean — after the initial page load, everything runs in the browser: formatting, validation, transforms, the lot. No account, no cookies required, paste and go.&lt;/p&gt;

&lt;p&gt;The parts I actually use daily:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Validation that names the error.&lt;/strong&gt; Chrome 114+ changed how V8 reports JSON syntax errors, which silently broke a bunch of tools that parsed V8's error strings. A scanner written against the spec instead survives browser updates, and gives you plain-English errors with exact line/column plus jump-to-error: &lt;a href="https://jsonformatter.app/news/json-validator-error-locations" rel="noopener noreferrer"&gt;why JSON validators should name the error&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Real jq in the browser.&lt;/strong&gt; Not a jq-&lt;em&gt;lookalike&lt;/em&gt; — actual jq compiled to WebAssembly: &lt;code&gt;map&lt;/code&gt;, &lt;code&gt;select&lt;/code&gt;, &lt;code&gt;sort_by&lt;/code&gt;, &lt;code&gt;group_by&lt;/code&gt;, object construction, with live preview as you type: &lt;a href="https://jsonformatter.app/news/json-transform-jq-jsonpath" rel="noopener noreferrer"&gt;jq + JSONPath transform panel&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;One-click auto-repair&lt;/strong&gt; for the trailing-comma / single-quote / comment mistakes we all make when hand-editing config&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;And if you want the general checklist for evaluating whether a formatter is safe to use with sensitive payloads — the offline test, the network audit, what to look for in the privacy policy: &lt;strong&gt;&lt;a href="https://jsonformatter.app/news/secure-json-formatter-online" rel="noopener noreferrer"&gt;what makes a JSON formatter actually secure&lt;/a&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The habit worth keeping
&lt;/h2&gt;

&lt;p&gt;It's not "never use online tools" — it's &lt;strong&gt;know which tier of data you're pasting&lt;/strong&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Never&lt;/strong&gt;: credentials, tokens, customer data, anything from prod → local tool or editor only&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gray zone&lt;/strong&gt;: internal payloads, staging responses → only tools that pass the network audit&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fine&lt;/strong&gt;: public API samples, tutorial snippets, your kid's Minecraft JSON → whatever's fastest&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The CodeBeautify leak wasn't malice — it was a "shareable links" feature meeting real-world data. Every tool with a paste box and a server is one config change away from the same outcome. Audit before you paste, not after.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Full disclosure of the incident and what was exposed: &lt;a href="https://jsonformatter.app/news/codebeautify-data-leak-explained" rel="noopener noreferrer"&gt;CodeBeautify data leak explained&lt;/a&gt; — and yes, run the audit on our tool too. That's the point.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>privacy</category>
      <category>tutorial</category>
      <category>security</category>
    </item>
    <item>
      <title>Your internet isn't slow — it's bufferbloat. Diagnose it in 25 seconds.</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Sat, 31 Oct 2026 07:26:03 +0000</pubDate>
      <link>https://dev.to/easy28/your-internet-isnt-slow-its-bufferbloat-diagnose-it-in-25-seconds-b9a</link>
      <guid>https://dev.to/easy28/your-internet-isnt-slow-its-bufferbloat-diagnose-it-in-25-seconds-b9a</guid>
      <description>&lt;p&gt;You pay for 500 Mbps. At 2 a.m. everything is instant. At 7 p.m., when someone starts a game update, your video call turns into a slideshow — and the speed test you run right after still cheerfully reports 500 Mbps.&lt;/p&gt;

&lt;p&gt;Nothing is broken. Your speed test just doesn't measure the thing that's actually hurting you.&lt;/p&gt;

&lt;h2&gt;
  
  
  The number that speed tests don't show you
&lt;/h2&gt;

&lt;p&gt;Raw bandwidth answers "how much data can move per second". It says nothing about &lt;strong&gt;how long each tiny packet waits when the line is busy&lt;/strong&gt; — and that latency-under-load number is what real-time apps live and die by.&lt;/p&gt;

&lt;p&gt;That problem has a name: &lt;strong&gt;bufferbloat&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Modern routers and modems buffer packets instead of dropping them. That's polite under light load, but when a download saturates your link, outgoing packets — your video call frames, your game inputs — get stuck at the back of a huge queue. Your ping goes from 8 ms to 300+ ms while the download runs. The call stutters. The game desyncs. When the download finishes, everything is magically fine again, which is why it feels haunted.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Jitter&lt;/strong&gt; is the sibling problem: the &lt;em&gt;variation&lt;/em&gt; in your ping between packets. Real-time codecs can hide high latency better than they can hide inconsistent latency.&lt;/p&gt;

&lt;h2&gt;
  
  
  The 25-second diagnosis
&lt;/h2&gt;

&lt;p&gt;You need a test that measures latency &lt;strong&gt;while the line is saturated&lt;/strong&gt;, not just at idle. Waveform's bufferbloat test is the well-known option.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://speedpulse.tech" rel="noopener noreferrer"&gt;SpeedPulse&lt;/a&gt; is the one I built — free, no signup, no app, no cookies, everything runs client-side. One ~25-second run measures:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Download / upload / idle ping&lt;/strong&gt; — the usual suspects&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Jitter&lt;/strong&gt; (RFC 3550-style mean difference between consecutive pings)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bufferbloat&lt;/strong&gt; — a ping train keeps running &lt;em&gt;during&lt;/em&gt; the download and upload phases; the tool reports &lt;code&gt;loaded latency − idle latency&lt;/code&gt; per direction and grades the worse one from A+ to F&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Plain-English verdicts&lt;/strong&gt; for 4K streaming, gaming, video calls and browsing, so you don't have to interpret raw numbers&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;An 8-CDN cross-check&lt;/strong&gt; — after each test it pings 8 independent CDN edges (Cloudflare, jsDelivr, Fastly, GitHub, Google, AWS...). If the speed-test server is dramatically faster than the pack, your ISP may be prioritizing speed-test traffic — the classic reason "the test says I'm fine but everything feels slow"&lt;/li&gt;
&lt;li&gt;CSV export and a local history chart, all client-side&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Reading the grade
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Grade&lt;/th&gt;
&lt;th&gt;Extra latency under load&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;A+&lt;/td&gt;
&lt;td&gt;&amp;lt; 5 ms&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A&lt;/td&gt;
&lt;td&gt;&amp;lt; 30 ms&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;B&lt;/td&gt;
&lt;td&gt;&amp;lt; 60 ms&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;C&lt;/td&gt;
&lt;td&gt;&amp;lt; 200 ms&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;D&lt;/td&gt;
&lt;td&gt;&amp;lt; 400 ms&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;F&lt;/td&gt;
&lt;td&gt;≥ 400 ms&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;If you score C or worse, you've found your problem.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Full disclosure: I built SpeedPulse. The grade scale is the same A+–F scheme Waveform and the bufferbloat.org community use, so results are comparable across tools — I'd rather share a standard than invent a flattering one.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The fixes that actually work
&lt;/h2&gt;

&lt;p&gt;In order of effectiveness:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Enable SQM with fq_codel/CAKE&lt;/strong&gt; on your router (trivial on OpenWrt; many stock firmwares ship a basic "QoS"). Smart queue management keeps buffers short so latency stays flat even at full load. This is &lt;em&gt;the&lt;/em&gt; fix — A+ grades come from this.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cap your speeds ~5–10% below your ISP rate&lt;/strong&gt; in the SQM settings. Counter-intuitive but correct: queues then form inside your router, where SQM controls them, instead of inside the ISP's modem, where they don't.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fix the upload direction first.&lt;/strong&gt; Skinny upstreams bloat first — video calls die on &lt;em&gt;upload&lt;/em&gt; saturation (cloud backup + call = slideshow). If you can only cap one side, cap upload.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Go wired for the machine that matters.&lt;/strong&gt; Wi-Fi retransmissions and airtime contention add jitter no router setting can undo.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Check for speed-test favoritism.&lt;/strong&gt; If only the test's server is fast while every other CDN is slow, the 8-CDN panel makes the mismatch visible — worth raising with your ISP.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  For the nerds: how you measure this in a browser
&lt;/h2&gt;

&lt;p&gt;No ICMP, no raw sockets. A browser test times HTTPS round trips with &lt;code&gt;fetch(url, { mode: 'no-cors' })&lt;/code&gt; and &lt;code&gt;performance.now()&lt;/code&gt;. &lt;code&gt;no-cors&lt;/code&gt; returns an opaque response — no body, no status — but the full network round trip (DNS, TCP, TLS, HTTP) completes, which is all timing needs. Jitter is the mean absolute difference between consecutive samples; bufferbloat is the median loaded latency minus the idle median, measured per direction, graded on the worse of the two.&lt;/p&gt;

&lt;p&gt;Full methodology write-up, including the limitations we're honest about: &lt;a href="https://speedpulse.tech/blog/how-speedpulse-measures-jitter-and-bufferbloat/" rel="noopener noreferrer"&gt;How SpeedPulse measures jitter, bufferbloat and 8 CDN latencies&lt;/a&gt;.&lt;/p&gt;




&lt;p&gt;Run the test on your line and drop your grade in the comments — happy to help interpret results: &lt;a href="https://speedpulse.tech" rel="noopener noreferrer"&gt;speedpulse.tech&lt;/a&gt;. Want the long-form bufferbloat explainer (router queue theory, CAKE vs fq_codel)? &lt;a href="https://speedpulse.tech/blog/what-is-bufferbloat/" rel="noopener noreferrer"&gt;What is bufferbloat?&lt;/a&gt;&lt;/p&gt;

</description>
      <category>networking</category>
      <category>tutorial</category>
      <category>performance</category>
      <category>webdev</category>
    </item>
    <item>
      <title>How to Create Custom Emojis for Discord and Slack for Free (No Design Skills)</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Fri, 30 Oct 2026 17:44:31 +0000</pubDate>
      <link>https://dev.to/easy28/how-to-create-custom-emojis-for-discord-and-slack-for-free-no-design-skills-4efe</link>
      <guid>https://dev.to/easy28/how-to-create-custom-emojis-for-discord-and-slack-for-free-no-design-skills-4efe</guid>
      <description>&lt;p&gt;Every server admin and community manager hits the same wall eventually: you need a reaction emoji that simply doesn't exist. The inside joke lands, everyone types "we NEED that as an emoji," and then... nothing, because making custom emoji images sounds like a design job.&lt;/p&gt;

&lt;p&gt;It isn't. Here are two free ways to make original emojis in under a minute, plus how to upload them to Discord and Slack.&lt;/p&gt;

&lt;h2&gt;
  
  
  Option 1: Fuse two emojis into a new one
&lt;/h2&gt;

&lt;p&gt;The fastest route is an emoji "mixer" — it blends any two existing emojis into a brand-new one, no drawing required. &lt;a href="https://emojimagic.io/emoji-mixer" rel="noopener noreferrer"&gt;EmojiMagic.io's Emoji Mixer&lt;/a&gt; does this instantly in the browser:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Pick emoji A (say 🐱)&lt;/li&gt;
&lt;li&gt;Pick emoji B (say ☕)&lt;/li&gt;
&lt;li&gt;Get the fused result — cat + coffee usually lands somewhere adorable&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Every pair works, and you can regenerate until something clicks. No signup, no watermark, works on your phone too.&lt;/p&gt;

&lt;h2&gt;
  
  
  Option 2: Generate from a text prompt
&lt;/h2&gt;

&lt;p&gt;When you need something no two emojis can produce ("a capybara wearing a hard hat"), use an AI emoji generator instead. On &lt;a href="https://emojimagic.io" rel="noopener noreferrer"&gt;emojimagic.io&lt;/a&gt; you type the idea and get four styles back: Apple style, sticker, cute 3D, and flat. Free daily credits, no account needed.&lt;/p&gt;

&lt;h2&gt;
  
  
  Export a transparent PNG
&lt;/h2&gt;

&lt;p&gt;Both routes give you a &lt;strong&gt;transparent PNG&lt;/strong&gt; download — this matters, because a white box behind your emoji looks terrible on dark mode. The site handles transparency automatically; just hit download.&lt;/p&gt;

&lt;h2&gt;
  
  
  Upload to Discord
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Open your server → &lt;strong&gt;Server Settings → Emoji&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;Click &lt;strong&gt;Upload Emoji&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;Pick your PNG, give it a name in lowercase (e.g. &lt;code&gt;:catcoffee:&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;Everyone in the server can now use it — including Nitro users, who can carry it to other servers&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Discord resizes anything you throw at it, but square images around 128px look crispest.&lt;/p&gt;

&lt;h2&gt;
  
  
  Upload to Slack
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Workspace name → Customize workspace → Emoji&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Add custom emoji&lt;/strong&gt;, name it, upload the PNG&lt;/li&gt;
&lt;li&gt;It's instantly available to the whole workspace&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Two quick tips
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Test at small size.&lt;/strong&gt; Your emoji renders at 22px in most chat clients. If the detail disappears, simplify.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Name for discoverability.&lt;/strong&gt; &lt;code&gt;:catcoffee:&lt;/code&gt; beats &lt;code&gt;:image3:&lt;/code&gt; — people type names to find reactions.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That's the whole workflow: mix or generate, download the transparent PNG, upload. The emoji your community asked for doesn't need a designer — just a free browser tab.&lt;/p&gt;

</description>
      <category>discord</category>
      <category>tutorial</category>
      <category>beginners</category>
      <category>productivity</category>
    </item>
    <item>
      <title>How Emoji Semantic Search Works: Embeddings, Cosine Similarity, and 310k Vectors in Milliseconds</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Thu, 29 Oct 2026 16:56:52 +0000</pubDate>
      <link>https://dev.to/easy28/how-emoji-semantic-search-works-embeddings-cosine-similarity-and-310k-vectors-in-milliseconds-470d</link>
      <guid>https://dev.to/easy28/how-emoji-semantic-search-works-embeddings-cosine-similarity-and-310k-vectors-in-milliseconds-470d</guid>
      <description>&lt;p&gt;Type "cozy morning coffee" into our emoji search and it returns breakfast, warm-light and calm-mood emoji combinations — none of which share a single keyword with the query. No synonym table, no hand-tagging. That swap from keyword lookup to meaning-based matching is powered by the most boring-sounding stack in modern ML: text embeddings plus cosine similarity, scanned over ~310,000 precomputed vectors. This is the engineering tour of how it works and what we learned building it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keyword tables don't generalize
&lt;/h2&gt;

&lt;p&gt;Classic emoji search is a lookup table: "happy" maps to 😊, and if you type "joyful" you get nothing. Every synonym must be hand-tagged and every phrasing anticipated.&lt;/p&gt;

&lt;p&gt;Embedding-based search flips the problem. Instead of matching strings, you match points in a high-dimensional space:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Every query is converted into a vector of numbers.&lt;/li&gt;
&lt;li&gt;Every item in the catalog (for us, 310k+ emoji combinations) is also a vector, precomputed ahead of time.&lt;/li&gt;
&lt;li&gt;"Similar" simply means "geometrically close" — and geometry generalizes to words the system has never seen.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The embedding pipeline, step by step
&lt;/h2&gt;

&lt;p&gt;We use an open-source multilingual model (BGE-M3, 1024 dimensions). Every text — queries at runtime, the whole catalog offline — goes through the same four stages:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Tokenization&lt;/strong&gt; — the model sees subword tokens, not words: &lt;code&gt;"cozy morning coffee" → ["co", "##zy", "morning", "coffee"]&lt;/code&gt;. Rare words split into known pieces, so the model can embed text it has never seen.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Transformer encoding&lt;/strong&gt; — each token flows through self-attention layers, exchanging information with every other token. "coffee" absorbs the fact that "morning" is next to it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mean pooling&lt;/strong&gt; — one per-token vector each is averaged into a single 1024-dimension sentence vector.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;L2 normalization&lt;/strong&gt; — the vector is scaled to unit length, landing on the surface of a unit sphere.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That last step is load-bearing. Once every vector has length 1, comparing &lt;em&gt;directions&lt;/em&gt; is all that's left.&lt;/p&gt;

&lt;h2&gt;
  
  
  The math that ranks everything
&lt;/h2&gt;

&lt;p&gt;Cosine similarity measures the angle between two vectors, ignoring length — from −1 (opposite) through 0 (unrelated) to +1 (same direction):&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;cos(θ) = (q · c) / (|q| × |c|)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Because we normalized everything to unit length, the denominator is always 1 and the whole formula collapses to a plain &lt;strong&gt;dot product&lt;/strong&gt; — one multiply-and-add per dimension. There's also a neat identity that falls out:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;|q − c|² = 2 − 2 × cos(θ)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Smaller Euclidean distance and larger cosine similarity are the &lt;em&gt;same statement&lt;/em&gt; for unit vectors. Whether your nearest-neighbor code sorts by dot product, cosine, or Euclidean distance, the ranking is identical.&lt;/p&gt;

&lt;p&gt;Here's the entire mechanism in three dimensions, small enough to check by hand:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;numpy&lt;/span&gt; &lt;span class="k"&gt;as&lt;/span&gt; &lt;span class="n"&gt;np&lt;/span&gt;

&lt;span class="n"&gt;query&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;np&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;array&lt;/span&gt;&lt;span class="p"&gt;([&lt;/span&gt;&lt;span class="mf"&gt;0.80&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.60&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.00&lt;/span&gt;&lt;span class="p"&gt;])&lt;/span&gt;   &lt;span class="c1"&gt;# "coffee"
&lt;/span&gt;&lt;span class="n"&gt;combos&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;np&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;array&lt;/span&gt;&lt;span class="p"&gt;([&lt;/span&gt;
    &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mf"&gt;0.70&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.70&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.14&lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt;                &lt;span class="c1"&gt;# "hot drink"
&lt;/span&gt;    &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mf"&gt;0.20&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.30&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.93&lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt;                &lt;span class="c1"&gt;# "birthday party"
&lt;/span&gt;&lt;span class="p"&gt;])&lt;/span&gt;

&lt;span class="n"&gt;cosine&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;combos&lt;/span&gt; &lt;span class="o"&gt;@&lt;/span&gt; &lt;span class="n"&gt;query&lt;/span&gt; &lt;span class="o"&gt;/&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="n"&gt;np&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;linalg&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;norm&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;combos&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;axis&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="n"&gt;np&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;linalg&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;norm&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;query&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="nf"&gt;print&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;cosine&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;   &lt;span class="c1"&gt;# [0.98 0.34]  -&amp;gt; hot drink wins by a mile
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Scale that from 3 dimensions to 1024, and from 2 candidates to 310,000, and you have semantic emoji search.&lt;/p&gt;

&lt;h2&gt;
  
  
  The lifecycle of one search
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Offline, once:&lt;/strong&gt; embed every catalog item and store its vector alongside its text. Vectors never change until the catalog does, so this cost is paid once — not per search.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Online, every search:&lt;/strong&gt; embed the query with the &lt;em&gt;same&lt;/em&gt; model, compute the dot product against the catalog, sort, return the top 100. Mixing models between index time and query time is the classic bug that quietly destroys result quality — two models produce two incompatible geometries.&lt;/p&gt;

&lt;h2&gt;
  
  
  Exact scan vs. ANN index
&lt;/h2&gt;

&lt;p&gt;1024 dimensions × 310k vectors is ~318 million multiply-adds per query. That sounds scary and isn't: a tight vectorized loop finishes in well under a second, and an exact scan returns the &lt;em&gt;provably&lt;/em&gt; best matches. Most systems switch to approximate nearest-neighbor (ANN) indexes — graph or cluster structures that trade a sliver of recall for an order-of-magnitude speedup — only past a few million vectors. Know your threshold before you reach for the fancy index.&lt;/p&gt;

&lt;h2&gt;
  
  
  Lessons we learned the hard way
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Normalize your vectors&lt;/strong&gt; at write time — cosine becomes a dot product and rankings stay consistent.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Use one model everywhere.&lt;/strong&gt; Index and query vectors must share a geometry.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Precompute aggressively.&lt;/strong&gt; Only the live query should pay the embedding cost.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Watch p99, not the average.&lt;/strong&gt; Inference has cold starts; warm instances matter more than average speed.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Evaluate with real queries.&lt;/strong&gt; Cosine scores don't feel like relevance — the only test that counts is whether humans like what comes back.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If you want to see it ranking live, every query on &lt;a href="https://emojimagic.io/text-to-emoji" rel="noopener noreferrer"&gt;emojimagic.io/text-to-emoji&lt;/a&gt; runs exactly this pipeline — &lt;a href="https://emojimagic.io/blog/how-emoji-embeddings-work" rel="noopener noreferrer"&gt;this post on our blog&lt;/a&gt; has the full worked example, and the site itself is free with no signup if you'd rather throw queries at it than read about them.&lt;/p&gt;

</description>
      <category>machinelearning</category>
      <category>nlp</category>
      <category>python</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Why SVG Files from Figma Are 50–80% Dead Weight (and What to Strip Out)</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Tue, 27 Oct 2026 23:52:48 +0000</pubDate>
      <link>https://dev.to/easy28/why-svg-files-from-figma-are-50-80-dead-weight-and-what-to-strip-out-g3o</link>
      <guid>https://dev.to/easy28/why-svg-files-from-figma-are-50-80-dead-weight-and-what-to-strip-out-g3o</guid>
      <description>&lt;p&gt;Every project has at least one multi-kilobyte SVG icon that could have been a few hundred bytes. Design tools write SVG for &lt;em&gt;their&lt;/em&gt; convenience, not for the web — and the result is routinely 3–5× larger than it needs to be.&lt;/p&gt;

&lt;p&gt;Here are the 7 things making your SVGs bigger than they should be.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Editor metadata
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;&amp;lt;metadata&amp;gt;&lt;/code&gt; blocks, &lt;code&gt;rdf:RDF&lt;/code&gt; payloads, generator comments — Figma, Illustrator and Inkscape all pad files with data only they read. Browsers never render it.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Coordinate precision
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;74.98399992065430&lt;/code&gt; renders identically to &lt;code&gt;74.984&lt;/code&gt;. Dropping to 2–3 decimals is the single biggest win on path-heavy artwork — often half the file.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Empty and nested groups
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;&amp;lt;g&amp;gt;&lt;/code&gt; wrappers that mirror the editor's layer panel but apply no transform or style.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Unused defs
&lt;/h2&gt;

&lt;p&gt;Gradients, filters, masks and symbols defined but never referenced by any shape.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Editor IDs and attributes
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;id="Layer_1"&lt;/code&gt;, &lt;code&gt;data-name&lt;/code&gt;, &lt;code&gt;xmlns:serif&lt;/code&gt;, style attributes with no visual effect.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Whitespace
&lt;/h2&gt;

&lt;p&gt;Pretty-printed indentation and newlines — great for humans, dead weight for machines.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. Embedded raster images (the trap)
&lt;/h2&gt;

&lt;p&gt;A base64 PNG hiding inside &lt;code&gt;&amp;lt;image&amp;gt;&lt;/code&gt; can be 95% of the file size, and &lt;strong&gt;no SVG optimizer can shrink it&lt;/strong&gt;. If your "SVG" is mostly an embedded bitmap, the honest fix is either re-exporting the artwork as true vectors or shipping an optimized WebP/JPEG instead.&lt;/p&gt;

&lt;h2&gt;
  
  
  How much can you actually save?
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Illustrator exports: &lt;strong&gt;50–80%&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;Sketch: 40–70%&lt;/li&gt;
&lt;li&gt;Figma: 30–60%&lt;/li&gt;
&lt;li&gt;Hand-written SVGs: 5–15%&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;And an optimized SVG is plain text, so gzip/Brotli squeezes another 50–70% on the wire. The two stack — do both.&lt;/p&gt;

&lt;h2&gt;
  
  
  Automate it
&lt;/h2&gt;

&lt;p&gt;If the same SVGs ship in an app, make optimization a compile step, exactly like minifying JavaScript:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Vite:&lt;/strong&gt; vite-plugin-svgo&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Webpack:&lt;/strong&gt; svgo-loader&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Any CI:&lt;/strong&gt; &lt;code&gt;npx svgo -f src/icons -o dist/icons&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Keep editable originals in the design repo, let the pipeline emit optimized files, and nobody has to remember anything again.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;The full guide — with a browser-based optimizer (SVGO running client-side, no upload) and a batch processor for whole icon sets — is at &lt;a href="https://svgeditor.dev/svg-optimizer" rel="noopener noreferrer"&gt;svgeditor.dev&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>svg</category>
      <category>webdev</category>
      <category>performance</category>
      <category>frontend</category>
    </item>
    <item>
      <title>Favicon not showing? Here's how to fix it (10 causes)</title>
      <dc:creator>FF</dc:creator>
      <pubDate>Tue, 27 Oct 2026 03:37:08 +0000</pubDate>
      <link>https://dev.to/easy28/favicon-not-showing-heres-how-to-fix-it-10-causes-ehb</link>
      <guid>https://dev.to/easy28/favicon-not-showing-heres-how-to-fix-it-10-causes-ehb</guid>
      <description>&lt;p&gt;A missing or broken favicon is one of the most common questions on Stack Overflow — and the answer is almost always one of ten things. Work through this checklist top to bottom and you'll find it.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Browser caching (the #1 cause)
&lt;/h2&gt;

&lt;p&gt;Browsers cache favicons extremely aggressively — far more than regular pages. Even after you've fixed the issue, the old (broken) icon can persist for days. Test in this order:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Hard refresh:&lt;/strong&gt; Ctrl/Cmd + Shift + R&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Incognito / private window&lt;/strong&gt; — if it shows there, it's just your cache&lt;/li&gt;
&lt;li&gt;Visit the file directly: open &lt;code&gt;https://yourdomain.com/favicon.ico&lt;/code&gt;. If you see the image, the file is fine and it's a cache/link issue.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  2. Wrong file path
&lt;/h2&gt;

&lt;p&gt;The href in your &lt;code&gt;&amp;lt;link&amp;gt;&lt;/code&gt; must point to where the file actually is. A leading slash (&lt;code&gt;/favicon.ico&lt;/code&gt;) means site root.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="c"&gt;&amp;lt;!-- ❌ wrong if file is in /assets/ --&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"favicon.ico"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;

&lt;span class="c"&gt;&amp;lt;!-- ✅ absolute path from site root --&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/favicon.ico"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;

&lt;span class="c"&gt;&amp;lt;!-- ✅ or the real location --&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/assets/favicon.ico"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  3. Missing &lt;code&gt;&amp;lt;link&amp;gt;&lt;/code&gt; tag
&lt;/h2&gt;

&lt;p&gt;Some browsers fall back to &lt;code&gt;/favicon.ico&lt;/code&gt; automatically, but you should never rely on that. Always declare icons explicitly in &lt;code&gt;&amp;lt;head&amp;gt;&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;type=&lt;/span&gt;&lt;span class="s"&gt;"image/x-icon"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/favicon.ico"&lt;/span&gt; &lt;span class="nt"&gt;/&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"icon"&lt;/span&gt; &lt;span class="na"&gt;type=&lt;/span&gt;&lt;span class="s"&gt;"image/png"&lt;/span&gt; &lt;span class="na"&gt;sizes=&lt;/span&gt;&lt;span class="s"&gt;"32x32"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/favicon-32x32.png"&lt;/span&gt; &lt;span class="nt"&gt;/&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;link&lt;/span&gt; &lt;span class="na"&gt;rel=&lt;/span&gt;&lt;span class="s"&gt;"apple-touch-icon"&lt;/span&gt; &lt;span class="na"&gt;sizes=&lt;/span&gt;&lt;span class="s"&gt;"180x180"&lt;/span&gt; &lt;span class="na"&gt;href=&lt;/span&gt;&lt;span class="s"&gt;"/apple-touch-icon.png"&lt;/span&gt; &lt;span class="nt"&gt;/&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  4. Wrong MIME type
&lt;/h2&gt;

&lt;p&gt;If your server serves &lt;code&gt;.ico&lt;/code&gt; as &lt;code&gt;image/png&lt;/code&gt; (or vice versa) some browsers will refuse to render it. &lt;code&gt;.ico&lt;/code&gt; should be &lt;code&gt;image/x-icon&lt;/code&gt; or &lt;code&gt;image/vnd.microsoft.icon&lt;/code&gt;; PNG should be &lt;code&gt;image/png&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. File not deployed
&lt;/h2&gt;

&lt;p&gt;Check that the file actually made it to production. Open the file URL in a new tab. A 404 means it's not there — rebuild and redeploy, and confirm your build pipeline copies favicons into the output directory.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Non-square source image
&lt;/h2&gt;

&lt;p&gt;Favicons must be square. If your source image isn't 1:1, browsers may show a stretched or blank icon. Crop first.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. Framework-specific gotchas
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Next.js App Router:&lt;/strong&gt; icon files go in &lt;code&gt;app/&lt;/code&gt; (not &lt;code&gt;public/&lt;/code&gt;) for auto-detection.&lt;br&gt;
&lt;strong&gt;Next.js Pages Router:&lt;/strong&gt; files go in &lt;code&gt;public/&lt;/code&gt;, tags go in &lt;code&gt;pages/_document.tsx&lt;/code&gt;.&lt;br&gt;
&lt;strong&gt;Webpack / Vite:&lt;/strong&gt; make sure &lt;code&gt;favicon.ico&lt;/code&gt; is in your &lt;code&gt;public/&lt;/code&gt; or &lt;code&gt;static/&lt;/code&gt; directory and referenced in your HTML template.&lt;/p&gt;

&lt;h2&gt;
  
  
  8. HTTPS / mixed content
&lt;/h2&gt;

&lt;p&gt;If your site is HTTPS but the favicon is referenced over HTTP, browsers will block it as mixed content. Always use HTTPS URLs or relative paths.&lt;/p&gt;

&lt;h2&gt;
  
  
  9. Service worker caching an old version
&lt;/h2&gt;

&lt;p&gt;If you have a PWA with a service worker, it may be serving a cached HTML that references an old favicon. Update and re-register the service worker.&lt;/p&gt;

&lt;h2&gt;
  
  
  10. The SVG favicon isn't supported
&lt;/h2&gt;

&lt;p&gt;SVG favicons aren't supported in every browser (notably older Chrome on Android). Always provide a PNG/ICO fallback alongside an SVG.&lt;/p&gt;

&lt;h2&gt;
  
  
  Still stuck? Automate the diagnosis
&lt;/h2&gt;

&lt;p&gt;Paste your URL into the &lt;a href="https://favicon.one/favicon-checker" rel="noopener noreferrer"&gt;Favicon Checker&lt;/a&gt; and it will tell you exactly which icons and tags are missing or broken — no guessing.&lt;/p&gt;




&lt;p&gt;Need a clean favicon package to replace a broken one? &lt;a href="https://favicon.one/" rel="noopener noreferrer"&gt;Generate one free at Favicon.one&lt;/a&gt; — text, emoji, SVG or image, every size and format, all processed locally in your browser.&lt;/p&gt;

</description>
      <category>beginners</category>
    </item>
  </channel>
</rss>
