[HN Gopher] How Compiler Explorer Works in 2025
       ___________________________________________________________________
        
       How Compiler Explorer Works in 2025
        
       Author : vitaut
       Score  : 205 points
       Date   : 2025-06-04 17:32 UTC (5 days ago)
        
 (HTM) web link (xania.org)
 (TXT) w3m dump (xania.org)
        
       | unconed wrote:
       | Putting AI disclaimers at the end of your post seems like the
       | wrong way to do it. If you feel the need to put a disclaimer, put
       | it at the top. Otherwise, what's the point?
        
         | YetAnotherNick wrote:
         | AI disclaimer is both in the top and bottom. First box
         | contains:
         | 
         | > Written with LLM assistance. Details at end.
        
           | unconed wrote:
           | Yep I totally didn't notice it because it's off to the side
           | and looks like some kind of promo box.
        
             | thesnide wrote:
             | i would argue that it might actually be a kind of promo box
             | :-)
        
         | arkj wrote:
         | Will anything be ever written in future without a little help
         | from llm?
        
         | windward wrote:
         | They're not even really disclaimers. There's no legal
         | liability. It's just disclosure.
        
       | quaintdev wrote:
       | Is it just me or something is wrong with HN
       | https://ibb.co/0RwqjZvP
        
         | arkj wrote:
         | An interesting glitch. A few more refreshes and got the site
         | unavailable message. It's fixed now.
        
       | arkj wrote:
       | Maybe it's a client side error but I see three links of this post
       | on the homepage.
        
       | psnehanshu wrote:
       | The tool is called Compiler Explorer, but is hosted at
       | godbolt.org. There's also compliler-explorer.com, which is the
       | same thing. Why not retire the first domain? Just redirect to the
       | namesake domain to prevent link rot.
        
         | krackers wrote:
         | Because most people colloquially know it as "Godbolt".
        
         | MeetingsBrowser wrote:
         | Most people I know, including myself, refer to it as godbolt
         | and not compiler explorer.
         | 
         | I didn't know it was hosted anywhere other than godbolt.org
        
         | 0xTJ wrote:
         | When I want to visit Compiler Explorer, I start typing godbolt,
         | then his Enter when the browser brings up the correct item from
         | my History. "compiler-explorer.com" is also much longer (if you
         | want to type out the whole things).
        
         | jevndev wrote:
         | Funny enough, he has talked about this exact problem on his
         | podcast "Two's complement"; Specifically the episode "The
         | future of compiler explorer". Commenters below are correct that
         | it's just about how heavily associated his name is with the
         | tool. I just figured I'd also drop this source here because he
         | has a lot of interesting things to say about his involvement
         | with the project
        
           | 57473m3n7Fur7h3 wrote:
           | For anyone else wanting to listen to the episode, this site
           | worked well for me:
           | 
           | https://podtail.com/en/podcast/two-s-complement/the-
           | future-o...
           | 
           | It does have ads, but they were not too intrusive. Scroll
           | down if there's an ad on first click and there's a play
           | button that plays the episode.
           | 
           | For me the ads it showed were only text and images, not audio
           | interrupting ads.
           | 
           | You can also listen to it on YouTube:
           | 
           | https://www.youtube.com/watch?v=2QXo5c7cUKQ
           | 
           | But since it's audio only, I preferred listening to it via
           | the aforementioned podcast website.
        
         | gpderetta wrote:
         | The godbolt name is so iconic that at $WORK our locally hosted
         | instance of compiler explorer is under the go/godbolt shortcut.
        
           | porridgeraisin wrote:
           | The go/godbolt shortcut form tells me that WORK=google
           | 
           | Is it right?
        
             | gpderetta wrote:
             | I had the go/ shortcut at at least two different jobs.
             | Neither was G. But it is possible that xooglers set it up.
        
       | wild_pointer wrote:
       | How about compiling client-side with wasm, fully or partially?
       | Feasible? Was it considered?
        
         | petabyt wrote:
         | Some of the compilers they use (msvc) are not open source and
         | are not allowed to run in wasm. And porting entire compiler
         | toolchains to wasm would be a gargantuan task.
        
           | charcircuit wrote:
           | WASM has x86 emulators.
        
             | tempay wrote:
             | It would be a lot of data to download and would likely be
             | uncomfortably slow for many cases.
             | 
             | Also, as is mentioned in the article in a few places, the
             | website is optimised for low effort rather than low cost.
        
         | jenadine wrote:
         | That would imply compiling each compiler to wasm, which might
         | not be supported. Plus challenges like making a virtual file
         | system to get the header files and other files needed for the
         | compilation.
         | 
         | Edit: also, you couldn't execute the resulting binary
        
       | galkk wrote:
       | tl;dr:                   * Right now, Compiler Explorer costs
       | around $3000 a month (including AWS, monitoring, Sentry for
       | errors, Grafana, and other expenses).          * nsjail for
       | security/isolation         * 3.9 terabytes of compilers,
       | libraries, and tools         * Up to 30+ EC2 instances (EC2
       | instances are virtual machines)         * 4,724 compiler versions
       | * 1,982,662 short links saved (and as of recently, ~14k ex-goo.gl
       | links)         * 1.8 million compilations per week
       | 
       | If my napkin math is correct, it's around 3 compilations/sec. and
       | their cost is 0.0004 cents per compilation. Fascinating. If
       | anyone asked me about ballpark estimate of compiler explorer
       | cost, I'd be wrong at least at magnitude. Like - they must be
       | heavy cpu/io/network bound, and this is like the worst scenario
       | for cloud use.
       | 
       | This and lichess
       | (https://news.ycombinator.com/item?id=41922928#41928953) shows
       | that you really can handle very serious loads for quite cheap.
        
         | dieortin wrote:
         | > $3,00015
         | 
         | That's a very weird place to put a comma. I have no idea what
         | number this is representing (since I doubt it's $3 a month)
        
           | detaro wrote:
           | the 15 is a copy-paste error (footnote)
        
             | sapiogram wrote:
             | Thank you, the comment has value now.
        
         | maccard wrote:
         | I'd love to see a breakdown of their ec2 costs vs everything
         | else. I'd expect that if they really wanted to a single bare
         | metal machine would probably save them a decent chunk of money.
         | 
         | But $3k/mo is surprisingly lean!
        
           | mattgodbolt wrote:
           | I'll do a follow-up post sometime on where the money goes.
        
       | tomcam wrote:
       | > 8. Your browser renders the assembly, and you go "ooh, how
       | clever is this compiler!"
       | 
       | Incorrect. I go "ooh, the appropriately named Mr. Godbolt is a
       | fucking beast!"
        
         | mattgodbolt wrote:
         | So sorry to disappoint you, but I'm really not :D
        
       | gpderetta wrote:
       | > I really don't know how many developers use Compiler Explorer:
       | we purposefully don't have the kind of tracking that could tell
       | us. But, it's at least in the thousands I think
       | 
       | I think this a large underestimate :D
        
         | knowitnone wrote:
         | Why do you say that? I've bookmarked the site but I've never
         | used it. If my code compiles, I'm good. Not many people dig
         | into the assembly to find that extra 1% performance boost. I'm
         | not saying this tool is not useful because it is but how often
         | do you use it?
        
           | JavierFlores09 wrote:
           | I seldom use it myself, but I frequently see people sharing
           | compiler explorer snippets to compare generated assembly on
           | discord conversations about LLVM in just about any native
           | language discord (but mainly the Rust official one), which
           | may sound like a rare occasion, but it happens more often
           | than you'd think. My impression is that it is definitely in
           | the thousands, but it being a service that I've only seen
           | used for quick comparisons rather than in-depth analysis, I
           | can't tell whether it is much more than that
        
           | gpderetta wrote:
           | While it might have been started to inspect the assembly,
           | these days I mostly see it getting used to share compilable
           | code and test it against multiple compilers.
           | 
           | Sometimes is just a quick way to test minimalist code snipets
        
       | pfdietz wrote:
       | I wonder how many compiler crashes they experience. That's a lot
       | of test cases.
        
       | cerisier wrote:
       | For the author: are there any plans to make the compilers builds
       | available ? Prebuilt gccs are a rare and valuable spice !
        
         | mattgodbolt wrote:
         | They are already free and available. Check out our "infra" repo
         | and the ce_install tool and/or just hack the S3 URLs
        
       ___________________________________________________________________
       (page generated 2025-06-09 23:02 UTC)