When I'm idling, I like to just… look at things. I might scroll HN, or trace random transactions on a blockchain explorer, or poke about in trending git repositories surfaced by Github's Explore.

I'm not looking for anything in particular. I'm hardly going to come back to anything I see today. But over time, just simply looking at things has helped me form a decent feel of what's out there.

This is handy in odd, unpredictable ways - Simply being aware of what else is out there is a problem-solving super power.

Sometimes, this extends to fixing something. Typically, this takes the form of trying to make nixpkgs just ever so slightly greener. For its various faults, nixpkgs does have good toolling. PRs are well labelled, there's a small army of people contributing, and OfBorg¹ and Hydra² do an amazing job of CI/CD.

I do maintain some packages for nixpkgs, and generally help out with things when I can, but fixing a random package is a window into another world. Where the window leads is decided by whichever failing Hydra job³ or GitHub issue I chance upon, and it can be a journey.

Earlier this week, this led me to the bicpl package - it was failing to build for darwin platforms⁴, and was a dependency for 4 other packages, causing them to fail as well.

Stepping through this window revealed that bicpl is a library built by the McConnel Brain Imaging Centre at the Montreal Neurological Institute. It's a C library used to process brain imaging data, and is used by a number of other programs. Neuroimaging wasn't a field I expected to be looking at today, but nevertheless that's where I found myself.

Looking at their code reveals that while the master branch gave the impression the project was abandoned, occasional fixes did crop up on other branches. In particular, the develop-apple branch did contain a fix for the darwin build issues. Bringing that into nixpkgs was simply a matter of bumping the version we were building.

Another issue surfaced after the version bump - the builder insisted that time-related functions weren't properly declared:

/tmp/nix-build-bicpl-unstable-2023-01-19.drv-0/source/Prog_utils/random.c:80:16: error: call to undeclared function 'time'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
        seed = time(NULL);

Now, I haven't touched C in a long, long time, but fortunately this does give us enough information to go on. A quick search for -Wimplicit-function-declaration later, and we find that clang16 has made this warning into an error. Nixpkgs, in an effort to keep things clean, safe, and stable, started defaulting to clang16 for darwin on 2023-06-11⁵, which is also when the bicpl package started to fail.

diff --git a/Prog_utils/random.c b/Prog_utils/random.c
index f61dbb6..f5dcb9c 100644
--- a/Prog_utils/random.c
+++ b/Prog_utils/random.c
@@ -23,6 +23,7 @@
 #endif

 #include <stdlib.h>
+#include <time.h>

 static  VIO_BOOL  initialized = FALSE;

Armed with this, the fix was simple - we just needed to include the time.h header file. Now, nixpkgs does have convenient ways to patch source files - I could add the patch to the repository and include it, but I elected to send a PR to the upstream repository instead. This way, the fix would be available to everyone, and not just nixpkgs users. It's an issue that would inevitably bite any users outside of nixpkgs too as clang16 adoption grows. At the same time, I opened a nixpkgs PR, pulling in the patch during the build for now. At the very least, this would fix any issues within nixpkgs and anyone downstream of it.

diff --git a/pkgs/development/libraries/science/biology/bicpl/default.nix b/pkgs/development/libraries/science/biology/bicpl/default.nix
index 5ad45b1f8c0431..45d1b159d07471 100644
--- a/pkgs/development/libraries/science/biology/bicpl/default.nix
+++ b/pkgs/development/libraries/science/biology/bicpl/default.nix
@@ -2,6 +2,7 @@
   lib,
   stdenv,
   fetchFromGitHub,
+  fetchpatch,
   cmake,
   libminc,
   netpbm,
@@ -9,16 +10,24 @@

 stdenv.mkDerivation rec {
   pname = "bicpl";
-  version = "unstable-2020-10-15";
+  version = "unstable-2023-01-19";

-  # current master is significantly ahead of most recent release, so use Git version:
+  # master is not actively maintained, using develop and develop-apple branches
   src = fetchFromGitHub {
     owner = "BIC-MNI";
     repo = pname;
-    rev = "a58af912a71a4c62014975b89ef37a8e72de3c9d";
-    sha256 = "0iw0pmr8xrifbx5l8a0xidfqbm1v8hwzqrw0lcmimxlzdihyri0g";
+    rev = "884b3ac8db945a17df51a325d29f49b825a61c3e";
+    hash = "sha256-zAA+hPwjMawQ1rJuv8W30EqKO+AI0aq9ybquBnKlzC0=";
   };

+  patches = [
+    # fixes build by including missing time.h header
+    (fetchpatch {
+      url = "https://github.com/RaghavSood/bicpl/commit/3def4acd6bae61ff7a930ef8422ad920690382a6.patch";
+      hash = "sha256-VdAKuLWTZY7JriK1rexIiuj8y5ToaSEJ5Y+BbnfdYnI=";
+    })
+  ];
+
   nativeBuildInputs = [ cmake ];
   buildInputs = [
     libminc

With that done and merged, bicpl and its dependents were building successfully again.

As a pleasant surprise, despite bicpl's repository being inactive, the upstream PR was merged within 24 hours, allowing me to remove the patch from the nixpkgs build.

But why?

By almost any definition, this was a pointless exercise - I don't use bicpl, or any of its derivates. Heck, I don't even use C enough to have to care about breaking changes in clang16. But random-walk discoveries like this are how I've learned about countless things over the years.

Now I know bicpl exists. Just based on the brief look through the window of open-source brain imaging, I can hazard a guess that their data files contain some kind of volumetric data, expressed as geomtric shapes and coordinates. Maybe there's some type or tag associated with these values to indicate density or mass⁶. Maybe it's computed on the fly. I know that they're active enough to notice uncalled for PRs on inactive repositories, and that they understand enough about software engineering to evaluate it quickly⁷.

I know there's a small sliver of open source centered around dealing with brain imaging data and how to manipulate it. A sliver that's been at it since at least 2001⁸. I don't know if anyone's going to use this package from nixpkgs, probably not. But having it functional means that if someone does, it's one less barrier for them to cross. If every package maintainer only mainted things they actively used, nixpkgs would be a much poorer place.

That's good enough for me.

And who knows, maybe one day I will need to know that clang16 broke implicit function declarations. I've used more arbitrary knowledge in the past⁹. Might even scan a brain someday.

Notes/References