Porting a macOS Clojure Dev Setup to Windows WSL2 | Hacker News Reader

Porting a macOS Clojure Dev Setup to Windows WSL2

blog.michielborkent.nl

93 pointsBorkdude86 comments

WSL 2 is pretty decent honestly.

I've been using WSL 1 + Docker for full time web development since early 2018 on Windows 10 Pro.

Upgraded to WSL 2 when it hit the stable release and it works quite well. The only thing that sucks is mount performance for files outside of WSL 2's file system but if your source code is inside of WSL 2's file system it's very solid.

This is with ~$850 worth of 6 year old hardware too (i5 3.2ghz, 16gb of RAM, 1st generation SSD, etc.). I never bothered to upgrade because everything I do around Flask / Phoenix / Rails / Node development + running Docker / VMs + etc. works great. I do video editing too.

If anyone is curious what it's like, I made a video where I went over all of the tools I use on Windows and WSL 2 to have an acceptable dev environment for Linux based development https://nickjanetakis.com/blog/a-linux-dev-environment-on-wi....

It covers both the Windows tools and how I have everything configured on the WSL 2 side of things (tmux, terminal Vim, etc.).

I'm really torn with WSL. WSL1 has far superior performance when performing I/O tasks on the mounted Windows drives, but slower than WSL2 locally. WSL2 is unbearably slow when performing I/O on the mounted Windows, using vim with a git pluggin or FZF is crippling. So I attempted to just use WSL2+vim on the linux local file system, but then when I try to run Windows npm scripts in a folder UNC paths cannot be understood by CMD.exe. It is the only thing preventing me from fully switching over.

Have you considered switching to GNU/Linux instead? Given that you do all your work in a Linux environment already...

If you have npm installed in WSL 2, what use cases are you running into where you need to run Windows npm scripts?

You could exclusively run npm commands inside of WSL 2 and never have to touch cmd.exe.

I have to run npm scripts in Windows land because they launch electron apps that require Windows.

I agree with the sentiment around I/O which is the last remaining pain point.

I only use it for Dropbox since my org files live there but I ended up installing the Dropbox CLI instead.

It kinda sucks though since WSL 2 doesn't have a proper init setup meaning I have to manually start Dropbox's daemon anytime I restart

folder UNC paths cannot be understood by CMD.exe

Does using Powershell instead of CMD solve this for you perhaps (I mean, it does wrt UNC, but I don't know what else you're doing with it)?

WSL 2 just makes VirtualBox unusable for me in practice so I can't really use it. And honestly if I wanted a VM I'd just use VirtualBox.

> Installing Ubuntu on new PC. The installer is surprisingly smooth, detects my wifi, even nvidia video card apparently.

This is the advantage of the development model behind Linux "drivers." People with serious commitment to code quality review them. They typically should work more reliably.

I usually don't complain about anything Windows, but certain OEMs (Dell, HP) clearly care only for hardware. I've been suffering through a comedy of awful drivers, culminating in using my phone to print.

I wish WHQL did more thorough testing.

https://en.wikipedia.org/wiki/WHQL_Testing is out of date, does WHQL still exist, after Microsoft firing a testing division in favor of Windows Insiders?

It's not spelled out clearly, but I'm understanding that the main rationale here is to:

  - have an Ubuntu dev environment, 
  - while benefiting from Win's RDP 
  - AND also having access to native Windows for testing.

Because Clojure runs perfectly well on regular Windows, it's not like WSL is required. So otherwise the author would have to have a separate Windows install just for testing.

Author here. This sums it up well. I want the WSL2 setup so I can also keep running the zsh setup and bash scripts I was already using on macOS. For work I need to run several Docker containers in addition to the Clojure/Java stuff. Since I had not used Docker on Windows prior to WSL2, I jumped to the combination immediately. I figured that running everything in WSL2 was easier than some things inside and some things outside.

Well if you don't strictly need Ubuntu chances are your scripts would work with Cygwin. Which operates under the regular Windows paradigm. Then you can use native Emacs too[0].

I've been doing Clojure stuff on a pure Windows setup for quite a while and honestly I liked it more than my current Linux setup (only switched because of some linux kernel features I needed).

[0] note that ansi-term is (or used to be) broken on Windows tough , other than that pretty much everything works

I think Cygwin and msys are essentially deprecated at this point. WSL is easier to setup and more compatible. I was big Cygwin fan for a long time, but I've definitely run into edge cases that it couldn't handle. WSL works pretty flawlessly for me so far and I can just use apt for package management instead of the Cygwin setup tool.

Depends if you care about doing anything natively on Windows or not. Say Windows has perfectly good native support for NodeJS. If you go with WSL you'll be running WSL's Node. Many may not want that.

Java/Clojure same deal. There's no problem with them on Windows. Running everything under WSL sounds like a big compromise to make to me. Though for Java/Clojure I could imagine you could still for most projects start your software from the normal java.exe as jars generally don't have platform specific code. But in my codebase I do actually use a few libraries that are system specific.

You can run windows apps within WSL too though.

(I guess maybe not WSL2)

Well the thing is if you npm install a lot of stuff using the WSL version there may be packages that include platform specific stuff. Then trying to run it with the native node might not work. But thinking of it I guess you could have all the WSL scripts use the native version. Not sure if there would be issues.

Or let's say you have to compile something that you want to use as a child process in your software. If you compile that under WSL (correct me if I'm wrong[0]) it's not going to be easily accessible to normal Windows apps.

[0] I see that now there's some way to invoke wsl commands from regular Windows but not sure how that works exactly.

Magit performance alone is enough to make me never want to run Emacs on Windows natively. I encounter various other issues as well, for example inferior modes missing prompts.

How is it on WSL2? I tried magit on WSL1 and it was still terribly slow.

Native speed.

Lots of bits and pieces very helpful eg the one about memory. Thanks.

FWIW this is what I do. And I run stuff in WSL when I occasionally encounter stuff that won't run directly on Win10. Been working for me for about a year now.

Why not run Linux natively? If you need Windows, you can run it in VM.

I don't know any legitimate mainstream reason why you would run Linux when WSL2 exists. It runs everything Linux does perfectly, but also with an operating system that isn't built around an 11-year old somehow still "experimental" display server that leaves fundamental use-cases like screen sharing to "an exercise for the reader"

WSL 2 has issues: https://github.com/Microsoft/WSL/issues

For most of the users doing a lightweight work, it might not be a problem but my repository is extremely huge and building on WSL 2 takes at least 5X more than natives Widnows or MacOS.

https://github.com/microsoft/WSL/issues/4197

edit: My use-case for WSL 2 is that I have codebase that only runs on windows / Linux. Switching between the two is cumbersome and VM takes a hit on performance. If I didn't need Windows then definitley no need to use WSL 2 and just go for Linux/MacOS

If Wayland doesn't work for you, you can still use Xorg, that has worked fine for millions of people in the last few decades. I like Wayland, but I won't switch until it gets to the same level of features and stability of Xorg.

Also if we must be fair, Windows still has no way to reliably install Windows updates through SSH, because you apparently cannot elevate yourself, without even if you are logged in as administrator. I still need to RDP into my Windows instances while every other OS can conveniently be updated using just a few shell commands.

Windows updates can be installed remotely/using only command line tools with Powershell, though I forget if it requires admin.

I tried almost everything and I can't for pete's sake find a way to get them to install when connected through SSH. Chocolatey works fine, though

Also, have you ever tried to do any serious in C++ development work under Windows? Downloading 10 gigabytes of SDK just to run the compiler is not fun.

Also, most if not all compilers (GCC is a number 1 offender here) are much slower than Linux or macOS under Windows. It all depends on what you do, what kind of user you are and what kind of job you are trying to do. No one uses something worse on purpose, so there sure must be reasons why people still say installing Linux is worth it. WSL is good for those who would have run Linux in a VM anyway, but if you need Linux or if you're just used to it, it's still not enough.

I haven't noticed GCC being slower, though I admittedly haven't measured. Which build of GCC do you use? MinGW64? or MSYS or something else? If you use the versions that are Cygwin derivatives, they will probably be slower, but that's not really due to GCC.

No, it's a native mingw-based GCC toolchain(cross compiler for an embedded platform), and everything is noticeably slower than on Linux or even macOS. I think it's because GCC is designed as relying on a multitude of separate binaries instead of being a single process. For each file it compiles it spawns a truckload of separate processes (collect2, cc1plus, as, ld...) and if there's something that's slow as heck on Windows, is spawning processes.

You're right about the slowness! I just noticed it now. I'm confused why though -- I don't actually see it spawning a lot of processes, and the slowness is far beyond what could be explained by that. Even a single process call appears to run much more slowly. I haven't dug into it but maybe the I/O pattern is funny?

Downloading 10 gigabytes of SDK just to run the compiler is not fun.

This is not very accurate. Just the compiler/linker (which doesn't need the SDK to build and run programs, i.e. equivalent of getting just gcc) is about 300mb to download. Then, assuming you're talking about the Windows SDK, which as far as I'm aware you only need when targetting OS-specifics is abut 700mb.

Why do you need WSL2? I don't want to touch Windows, unless there is no choice. WSL2 is just a bait MS are using to ensure Windows usage won't decline among developers. I'm not buying it.

If you are forced to use Windows (often corp IT are not competent enough or are simply not willing to spend extra resources on Linux support and are forcing Windows on users), then WLS2 is an option for running Linux. But I'd never pick it as a first choice.

Maybe I need systemd to manage services. Is it still broken under WSL2?

My notes. Many yaks were shaved due to Windows hating my graphics card. https://github.com/chadbrewbaker/CompilingOnAWSLambda/blob/m...

It’s a little bit off topic but does any one make WSL2 working on Surface Pro with the LTE connection? Everything works fine un WiFi but nothing works with LTE.

Usually adapter protocol bindings related. I have the same problem on my desktop with WSL2. It turns IPv4 off for some reason!

Why not use Linux directly? Set up one PC for games, and one for dev; I can't believe anyone uses Windows for anything but 'games.' But that's fair; just a subjective comment.

I've never worked somewhere that uses Linux dev machines. Almost always Mac these days, but otherwise it's been Windows. At least a good option exists for *nix tools on Windows now.

And WSL2 is good enough for personal development that I use it on my desktop as well -- I don't want to waste time fucking around with drivers or dual booting for something I do infrequently anyways.

I've never understood what drivers you people are "fucking around with". On Windows I have to go to netgears website and download a driver or use the included CD to get my WiFi dongle working. On Linux it works out of the box.

I've always found Windows to be the huge pain in the rear end, not Linux.

Lucky you? Usually it's my bluetooth adapter, or my AMD graphics card, or my thunderbolt, or my 4 button mouse... or.. or... the list goes on and on. For a normal experience yeah it generally could work out of the box if I just want chrome and facebook like normal people. But none of us here are normal are we?

If you want chrome and facebook you go to an american shopping building and get a chromebook or a rebranded mass-produced bottom-shelf el cheapo laptop with whatever it has preinstalled. Heck, normal users don't see a device in terms of hardware and software, it's just "the flip-open-thing that gets me cat pictures".

- Dunno what bluetooth you're using, but cheapo USB ones off Amazon work as well as my laptop's Intel built-ins.

- `radeon` is in the box (for Fedora, I assume Ubuntu) and has worked on half a dozen different cards over the last decade. I've never installed `amdgpu` but it looks no more difficult than nVidia's proprietary drivers?

- Thunderbolt is a continuing problem, though TB3 is significantly improved IME. Docks and multi-monitor over DisplayPort MST are fine. Haven't tried PCIe extensions.

- Many-buttoned mice work better than they do on Macs, I'll tell you that much.

AMD cards now with AMDGPU are vastly more convenient than Nvidia ones under Linux. AMDGPU is fully open source, so everything JustWorks™ without installing anything. Everything is shipped in Mesa and the kernel itself, and it does not require anything else.

Cool, good to know. I didn't realize that it was more like `nouveau`. That's awesome!

None of these cards need any additional driver under Linux, unless it's extremely obscure hardware made by some Chinese company. Damn AMD has even rewrote its entire Linux graphic drivers from scratch, now they are fully open and work fantastically out of the box. I bought an RX 580 specifically because now I can just install Linux and have everything working perfectly without caring about even installing a single package.

I bought a WiFi + Bluetooth PCIe card last month; it was one of those cheap Chinese no-brand cards that consist in an Intel WiFi chip slapped on a board. It worked instantly without installing anything, with just the drivers shipped in the Linux source tree that every distribution has. My mouse is a Logitech mouse with 5 buttons and it too works out of the box.

How long has it passed since the last time you installed Linux? It has changed a lot in the last 10 years; lots of things work flawlessly now. I haven't had to meddle with any driver since at least 2010, when my shitty ATi Radeon X1600 still only worked under FGLRX or when I tried to use a crappy Chinese WiFi dongle whose drivers were only distributed as source on a CD (damn you realtek)

Windows:

Sometimes Windows Update can't automatically find an appropriate driver the first time a device is connected and I have to install one manually. Time lost: fifteen minutes, tops.

Linux:

Last time I bought a USB wifi dongle, I bought one with a Linux penguin on the box. It turned out that the driver was a realtek fork and I had to manually apply their patches and rebuild every time my kernel updated. Could never get DKMS working quite right. Time lost: two hours.

I have dual monitors, one running on an Intel integrated card, one running on NVidia. Either one can run on the open source drivers alone, but I cannot activate both at the same time. The Intel drivers output hot garbage when Nouveau is running. There's a "flicker free" mode which eliminates a similar problem, but the Intel driver appears to ignore my config in this duel head configuration. Time lost: eight hours.

My experience:

WiFi device comes with CD but I have no CD-ROM. Then I try to google this no name brand and find the official website but fail.

I plug the device in Linux and figure out exactly the real chip used (many devices reuse same chips and put their own antenna and branding on top).

Now I know the real chip name but the guys that make it do not offer driver downloads to regular users, so I need to Google and find some driver.

Then the next problem, what website that offers driver downloads should I trust? I never heard of any of them, in the end I remember that many years ago I was downloading stuff from softpedia, decided that I already spent too much time on this and since it was my son PC I can risk it and grabbed a few potential drivers from there.

After that I needed to refresh my memory how to install drivers when you don't have a wizard and read a quick tutorial and follow it.

Conclusion, there are tons of hardware , as a Linux users in more then 15 years I can list only this issues(I am not using 4K or multiple screens setups):

- a few years back AMD catalyst driver dropped support for my 1 year old laptop (maybe the chip was older) forcing me to decide if I use older software or the crappy at that time open source software ( this costed AMD sicne I bought NVIDIA next time)

- for latest NVIDIA driver I just need to google , find and enable and official repository and now I can install from a list of drivers versions, I do not need to make an NVIDIA account to get updates.

- my Cannon printer was detect fine, I had to select a similar model from a dropdown and change a setting from RGM to CMYK (I had to google this)

- I had issues with microphones on cheap headphones, my solution was to buy USB headphones - they just work (including Microsoft ones) no driver installs

I never had to install an audio or network driver on Linux, on Windows I had issues with same hardware where installing drivers from Windows and not from the original CD would make the audio and network conflict (both were Realteck chips and Windows would get confused).

Btw I am not saying your experience is not happening, I am sure that some new hardware will be missing support and some other hardware might experience bugs, I added my experience to put things in balance.

Windows: Still has no package manager. Setting all software up takes time. Hardware was done by the laptop manufacturer, don't know how much work that would be. Works fine for the most part. Keyboard, mouse, wacom etc require a bit of fiddling.

Linux: Software setup is really easy. All hardware worked out of the box.

MacOS: Reading instructions, preparing installer etc. Got it "working" in a VM but without graphics accel, which makes it not really useful.

Time lost: A few hours more for windows, but quite comparable.

There technically is a “Windows package manager” https://github.com/microsoft/winget-cli, though it doesn't manage installed packages yet (that seems to be planned for v1.0 sometime next year https://github.com/microsoft/winget-cli/blob/master/doc/wind...).

Windows has like six 3rd party package managers. Chocolatey and Scoop have large repos and have been around since Windows 8. Modern Windows devices ship with the specific drivers required for the hardware and download drivers for unknown hardware automatically. Building my gaming PC from scratch, I didn't have to manually install drivers for anything. I've never even heard of a keyboard/mouse that didn't work out of the box.

Laptop suspend, ALSA sound, Bluetooth, and WiFi are almost always sources of constant frustration. I loved Linux but until they get that ironed out, I won't use it for a desktop solution.

I hate getting jerked around by Apple but not as much as I hate driver issues.

I won't buy non-Thinkpad laptops because I'm reasonably assured that they'll work with Fedora, but IME desktop hardware has a much higher chance of Just Working.

Who are "they", and why would they care about someone's demands? Don't get me wrong, I got bitten by some of these issues occasionally, too, but Linux is not an actual product I pay for. It's a bunch of free code. I can choose to contribute or not, but I can't expect that anyone is obliged to care about my demands.

This is a straw man argument. I wasn't making demands, never claimed Linux wasn't free, and never said anyone was expected to care about my situation.

I just said until they (contributors, obviously) have a better solution, I'll choose to pay for a desktop environment that suits my needs.

But why pay Apple tax for a Mac when you can pay $0 extra for a Thinkpad/Thinkstation that's certified tested to run multiple GNU/Linux distros well, or a Dell XPS “Developer Edition” or System76, Purism, or similar laptop with one preinstalled, and not be 'jerked around by' a single company?

Usually when a new series of GPUs is first released the driver support on Linux is pretty bad and even if it works its unstable.

The vast majority of corporate shops do not have a story for IT management of Linux machines. Accordingly, all employee PCs need to be either Windows or Mac -- and until WSL, for webdevs they were mostly Mac.

> The vast majority of corporate shops do not have a story for IT management of Linux machines.

Oh, that is changing and very quickly.

Even Microsoft is on board, for some reason.

Most corporations don't even support Macs and very often also don't support iPhones for mobile device management. Even in the IT department, you are very lucky if you can be "on your own" and use what gets the job done best/ you are most comfortable with. Oh and Bring Your Own Device isn't much of a thing here in middle Europe.

I can think of a few reasons not to, as someone who tried to use Linux first when switching away from macOS. The most notable of which is using Linux with a high-density 4k+ display is nightmarish at best; everything is either too big or too small, and at least in Ubuntu, fractional scaling is too broken to use on a daily basis – also requiring a ton of manual config and tweaking to even get usable. Windows, by contrast, works out of the box really well with HiDPI displays, and generally has a really well-refined experience these days, particularly between WSL2 and the Windows Terminal experience. It's the best of both worlds, without the hassle of dual-boot or manual config, IMO

Fractional scaling on Plasma Desktop works just fine on Ubuntu 20.04. I wouldn't use it if that wasn't the case.

A few Windows HiDPI problems I've experienced with an external 4k display: all the task bar status icons are blurred nearly unrecognizably after a while, Qt-based applications seem to be twice as large (text scale, and widgets) than win32 based ones (sure it's a Qt thing, no Windows), Steam can't figure out where the pointer is after a while (seems to track the pointer as 1/4 of where it is in x/y coordinates), and win32 applications often appear blurred and require some compatibility setting via the .exe file's properties thing which I don't understand because it seems to make no difference.

Honestly, Pop_OS! on the same setup is much smoother. Probably something to do with the HiDPI Daemon it has running.

The only flawless operating system on this 4k external setup has been macOS. Can't think of any problems and that's my daily system.

I've never experienced the issues you describe with status icons or steam having pointer issues except perhaps when changing the DPI settings without logging out and back in. When was the last time you tried it?

With regard to Win32 apps appearing blurry, I think that's a fair compromise considering Windows heavy focus on backwards compatibility.

Same here. I honestly thought they were talking about Linux, not Windows. Windows 4K/HiDPI support is just great with many different compatibility options.

Are you on Windows 10's 2020 release? That doesn't resonate with my experience at all – macOS has been vastly more frustrating to deal with on 4K displays in my experience, constant reflowing and forgetting window states, takes forever to recognize the displays when plugging in or resuming from sleep, etc... I even have all of the pointer and icon problems you describe with the top bar. Sometimes, with a 4K display plugged in, it takes 10+ seconds to figure out how to use it. Meanwhile, none of that on my Windows machine!

> 2020 release?

Probably. It's Windows 10 build 2004 or something.

The mac I have is mid-2014 w/ Mojave (also Catalina), and I'm using the hdmi out. It seems to remember the display, moves the applications I had on it last time to it, with all the other details like wallpaper scaling settings. It takes maybe 1/2 second to display.

To be fair that Mac was one of the greatest ever made, haha – I blame most of my issues with all of this on stability issues with the 2019 MBP I'm forced to use for work.

The primary display at login time determines the windows system dpi, so will have apps scaled correctly and if they are dpi aware they will appear sharp. If apps appear wrongly scaled on the primary display at login, they implemented dpi awareness in the wrong way. So, if you make sure the external 4k is primary before you login to windows, you should have a better experience.

On other screens with different dpi basically the only apps that look right are the ones using the latest dpi aware api’s.

FWIW, but I've had pretty good results from GNOME fractional scaling since like Fedora 26, across a wide range of hardware.

It's not perfect, but it's fine. WSL2 is also great for some use cases, but for anything JVM-based I expect first-class support for JetBrains tooling and that really doesn't work well. Keeps me solidly on a Linux desktop for now.

It was really really really slow when I tried it. My ass end mac mini with 1.5x scaling on a 4k display on macOS is noticeably faster than Fedora on a Ryzen 7 3700X with GTX1660 which lags like hell. Not impressed.

I'd buy performance issues, but on that hardware that surprises me. My desktop was a 2700X with a 980Ti and I didn't notice issues, but it's good to know that that mileage still varies.

(Now it's a 3900X with a 2070S and surprise, still none.)

It doesn't surprise me. It is Nvidia. Nvidia doesn't run Wayland. Of course your HiDPI experience will be bad.

Here, 4k with Vega and Mutter/Wayland is flawless.

That makes sense then.

I tried Kubuntu with KDE (on Macbook 2015) and am amazed at how good and polished it is, even compared to later versions of macOS. Fractional scaling works perfectly, even on NVidia cards (my desktop).

Even the Windows issues with HiDPI are non existent on Kubuntu. Really a good OS.

A MacBook Pro doesn't have a HighDPI 4K display, which is where it falls apart. Indeed, it works fine on a MacBook because Retina is not the same thing! :)

I also have a computer with NVidia gpu (which turns fractional scaling into an unresolvable problem on Gnome currently, that's why I tried Kubuntu in the first place) and 28" 4K display running Kubuntu and use the display with the Macbook too. Works flawlessly.

Running Manjaro Gnome on a 4K display; it works perfectly without any of the problems you describe. You might want to give Linux a chance again.

Last time I did it was February, and it was just as problematic back then – I'm not sure what advantage it offers over WSL2 (which is containerized) when combined with all of the benefits of Windows these days, for the hassle.

Windows containers aren't the same as Linux containers, as Windows containers run via Hyper-V, and Linux runs container processes natively.

After you start understanding how the system works, you can do things that are simply not possible with Windows. WSL2 is nice, but it's always still a VM. You don't have the flexibility of running a full Linux distro on real hardware. For instance, I made a portable work environment that boots straight from an SSD in a USB 3.1 case; it is fully encrypted, and it boots the same on every single machine under the sun. I can even quickly boot it in a VM using QEMU.

I personally use pretty much all the major OSes, and I think each one has its ups and down. I still think that Linux can potentially offer a much leaner and snappier experience than the other OS, because you can actually control what goes into your system and what runs and when. I also think that "easy" operating systems (and with them I also group distributions such as Ubuntu and similar) tend to give more advanced users an illusion of reliability that then sorely bites you in the back as soon as something goes awry. A few weeks ago I lost hours trying to understand why Windows was refusing to update to 2004, because Microsoft has hidden every single detail of how the system works away from the user.

I think that many people don't really know how to properly administer Windows systems, and that when things go south you still need to know what to do just like with every other OS.

Counterpoint, I've been running 4k screens with various scaling (27", 28" and now 43") with kubuntu and now fedora for over 3 years without any issues, using both Intel or AMD cards.

I also use Windows for music production. I built a Ryzen PC for around $1500 which is more powerful than the $6000 base Mac Pro. Thank god I left Logic years ago.

Oh absolutely, but to be fair the specs of the base level Mac Pro aren't really relevant to anything. If you're spending less than 5 digits on your Mac Pro, you're not doing it right.

I wouldn't need to spend that much.

It seems the highest MT score on Geekbench is the Mac Pro W-3275M with 19,189 points.

An AMD Ryzen Threadripper 3990X gets 24,049 points and costs less than $4000.

Reply on news.ycombinator.com