Add new post

2019-01-08 19:07:48 -05:00 · 2019-01-08 19:07:48 -05:00 · b1ce238846
commit b1ce238846
parent 2b7012b930
1 changed files with 66 additions and 0 deletions
--- a/blog/2019-01-08-recurse-center-day-#-2.markdown
+++ b/blog/2019-01-08-recurse-center-day-#-2.markdown
@ -0,0 +1,66 @@
+---
+layout: post
+title: "Misplaced inodes - Recurse Center Day # 2"
+date: 2019-01-08
+comments: true
+categories:
+  - rc
+  - freebsd
+  - haskell
+---
+
+I am part of the Winter 2 batch of [RC](https://recurse.com) and today is day
+#2!
+
+One of the (many!!) things I wanted to do at RC was to write more here, but
+turns out neglect had broken the code a bit. This blog is generated from bunch
+of markdown files using [hakyll](https://jaspervdj.be/hakyll/) and I was no longer able to build it on my laptop. At this point I should mention that my laptop is running `FreeBSD 13-CURRENT r342798`.
+
+## Errors
+As most haskell projects, build is managed by [stack](https://docs.haskellstack.org/en/stable/README/). On a good day `stack build` is all one need to build the project. stack takes care of setting up the compiler, downloading imported libraries and so on.
+
+Not today..
+
+```
+    Cabal-simple_mPHDZzAJ_1.24.2.0_ghc-8.0.2: No cabal file found.
+    Please create a package description file <pkgname>.cabal
+```
+
+That's weird, also not very helpful - because it has nothing do with the code I wrote.
+
+## Knee-Jerk solutions
+
+My first thought after the usual fast solutions (updating stack and moving to a newer lts etc) was to ditch stack altogether and use cabal to build the
+package, this proved to be a bit risky because I have haskell packages (like
+xmonad) installed using `pkg(7)` from freebsd package repo, and cabal was not
+happy about them. I could have maybe made it work with sandboxes but I am not
+familiar with them and decided to keep that as a backup plan. 
+
+## Reproducing errors
+Next step was to reproduce the error on a smaller project, turns out `stack new
+test && cd test && stack build` raises the same error. So its not the code base
+but the entire toolchain that's broken 😰
+
+Armed with this information and web-search-foo lead me to this [github issue](https://github.com/commercialhaskell/stack/issues/3515), and one key work stood out `ino64` change.
+
+[`inodes`](http://www.grymoire.com/Unix/Inodes.html) are datastructures that hold metadata information about an object in the filesystem. FreeBSD used to use 32bit value to hold this limiting number of objects you could have in the file system to 2^32. [ino64 change expands this to 64bit](https://lists.freebsd.org/pipermail/freebsd-fs/2017-April/024684.html). However this made all pre-compiled ghc versions stack downloads incompatible with FreeBSD versions higher than 12.
+
+Since inode structures are different ghc was somehow not able find files! 🛑
+
+## The solution
+Even though there is a patch landed for this problem in stack, it didn't make it
+to the latest release `v1.9.3`.
+
+So I opted to use the system-ghc for my build by adding following in `stack.yaml`
+
+```
+system-ghc: true
+```
+However this line doesn't have any effect unless you are on a `resolver` that uses the system ghc version, so I downgraded the resolver to `lts-12.26` that uses `ghc version 8.4.4` installed from ports.
+
+And voila builds are back! 🥳
+
+## Reflections
+1. The `stack` fix is landed on master, so next release should fix it (I did not test it). 
+2. A fellow freebsd haskeller suggested looking at [ghcup](https://github.com/haskell/ghcup), they don't support FreeBSD atm, but I like the idea of having a lightweight tool to manage different ghc versions.
+3. `stack` is ports is outdated (`v1.7.1`) and seems to be unmaintained, I should probably try to fix it. 🙂