Installing Stack

:: yakshaving, dev environment, haskell

By: Pookleblinky

I’m running Crouton, which makes some things difficult and others outright cartoonish. Here is what was involved in getting Stack to work.

Round 1: I installed stack from gethaskellstack. My ppa ghc was 7.4, which is cartoonishly outdated and which stack will yell at you about if you try to use it as global ghc. So: stack setup and it’ll install ghc, right?

No. First, I found that stack kept repeatedly freezing while downloading ghc 8.2, then starting over again. Sometimes the resource would vanish. When it did manage to download entirely, it’d attempt to unpack all 1.4gb into about half that amount of tmpfs. I knew that the tmpfs thing was a known issue, so I cleverly set $TMPDIR=~/tmp. Then I discovered that stack was not obeying this environment variable, and persisted in attempting to jam 10 gallons of shit into a 5 gallon bucket.

My tmpfs was half of total ram, and I my free space was about 1.5gb anyway. I manually plopped the ghc–8.2 tarball into where stack expected it, bypassing stack’s dodgy download process, and then tried to see if I could bypass stack in building it, taking advantage of an actual tmp directory. No dice: just not enough actual space for that.

So I moved my chroot out of internal storage and into an sdcard with 50gb of free space. This move was one command, worked instantly, and I haven’t seen any performance hit whatsoever. My filesystem now had plenty of space for updating from trusty to xenial (and ghc 7.4 to 7.10, which stack won’t yell about).

So I upgraded to xenial. Everything went absolutely smoothly. The only thing I had to tweak was that tmux’s updated version collapsed several mouse options into one, so I just had to replace them with that one setting.

I had plenty of space, a modern enough ghc, and the will to keep fighting stack.

Round 2. Xenial has haskell-stack in its repo. I installed it, and began working on the tmpfs problem. I discovered that it wasn’t listening to fish; I had to switch to zsh in order to set $TMPDIR such that it’d actually listen. This turned out to be a known issue since 2015. Stack will yell at you if you attempt to use a version below 1.4, so I had to upgrade it before I could install ghc 8.2.

I opened zsh in a pane, exported TMPDIR=~/tmp, and ran stack upgrade. Switched to another pane, cd’d to tmp, and stuff was going on! Things were happening! Compilation failed at 27/34 modules, something about ZLib compression having set invalid code lengths. Hey, at least it wasn’t my fault. I ran it again, this time it said only 29 modules were needed. It compiled all of them, exited successfully, and the version was not changed. The successful exit was a lie, it had not actually upgraded to 1.4. In a rage I uninstalled it and wiped ~/.stack.

I knew now that manually plopping things where stack expects them works, and that stack needs a non-fish shell interaction.

Round 3. I installed stack from gethaskellstack, so I would have 1.4 and wouldn’t have to be distracted by upgrading it. I opened a zsh pane, set TMPDIR=~/tmp, and then ran stack setup —git. It yelled that there was no package index, tried to download it, and then gave up saying the resource had vanished. I tried stack update —git. This caused stack to scream that there was no package index, and then try to download it, and then exit after the aws resource “vanished.” I ran it again with -v and discovered that the mirror it’s pointing at, simply doesn’t respond. It turns out this is a known issue, as well.

So: I needed to get that package index without relying on the mirror. I google-fu’d and found the json it was trying to get, and downloaded it into my frog static site. I redirected the url stack update was trying to get to into a localhost frog server, and voila. It installed the package index, and exited successfully. I then ran stack setup —git -v, and it miraculously began downloading ghc–8.2 without any interruptions. It downloaded the entire thing, then froze on installing ghc for 48 minutes. Then, exited successfully.

Right now it appears that stack is up and running, with a local ghc that is sufficiently up to date that it should stop yelling at me.