I think in this case it's better to create a patch rather than using sed. It's quite confusing like that.

And why are you using prePatch now?

Earlier you opened #34847 which we merged. It was still not properly fixed?

The timeout was not fixed there, the deadline was but due to my inexperience (my first "big" PR to nixpkgs) I thought preCheck was the right stage to go with, @dotlambda was right of course: I want prePatch.

Thanks for your patience and help, guys! Creating a patch file now 😊

After this has been resolved, I'll try building the package on a veeeeery slow ARMv7 machine. So please don't merge before my confirmation.

Actually, I think this derivation needs to be changed a little more.
According to https://github.com/pyca/pynacl/blob/master/INSTALL.rst#linux-source-build, pynacl ships a copy of the libsodium source. During installation, libsodium is built. However, it would be much nicer to use the libsodium provided by nixpkgs, which is possible using SODIUM_INSTALL=system.
However, I can do that in a different PR if you don't want to do it, @va1entin.

Thanks for pointing that out, @dotlambda
If you can spare some time, feel free to do that.
I tried to do it in installPhase but wasn't successful so far.

I can confirm that tests do not fail anymore on my Banana Pi with this PR applied.
So from my point of view this is ready to go and then we can further investigate how to use system libsodium.

I propose the following changes:

diff --git a/pkgs/development/python-modules/pynacl/default.nix b/pkgs/development/python-modules/pynacl/default.nix
index 981427e46f3..865b8c02fef 100644
--- a/pkgs/development/python-modules/pynacl/default.nix
+++ b/pkgs/development/python-modules/pynacl/default.nix
@@ -1,4 +1,4 @@
-{ stdenv, buildPythonPackage, fetchFromGitHub, pytest, coverage, libsodium, cffi, six, hypothesis}:
+{ lib, buildPythonPackage, fetchFromGitHub, isPyPy, pytest, libsodium, cffi, six, hypothesis }:
 
 buildPythonPackage rec {
   pname = "pynacl";
@@ -11,17 +11,21 @@ buildPythonPackage rec {
     sha256 = "0z9i1z4hjzmp23igyhvg131gikbrr947506lwfb3fayf0agwfv8f";
   };
 
-  #set timeout to unlimited, remove deadline from tests, see https://github.com/pyca/pynacl/issues/370
+  # set timeout to unlimited, remove deadline from tests, see https://github.com/pyca/pynacl/issues/370
   patches = [ ./pynacl-no-timeout-and-deadline.patch ];
 
-  checkInputs = [ pytest coverage hypothesis ];
-  propagatedBuildInputs = [ libsodium cffi six ];
+  checkInputs = [ pytest hypothesis ];
+  propagatedBuildInputs = [ libsodium six ] ++ lib.optional (!isPyPy) cffi;
+
+  preBuild = ''
+    export SODIUM_INSTALL=system
+  '';
 
   checkPhase = ''
-    coverage run --source nacl --branch -m pytest
+    py.test
   '';
   
-  meta = with stdenv.lib; {
+  meta = with lib; {
     maintainers = with maintainers; [ va1entin ];
     description = "Python binding to the Networking and Cryptography (NaCl) library";
     homepage = https://github.com/pyca/pynacl/;

I can't actually test with PyPy because hypothesis depends on configparser, which isn't supported for PyPy. That must still be fixed somehow :)
But pynacl's setup.py clearly states it is compatible with PyPy 2.7.

Either you incorporate the changes into your PR @va1entin or @FRidh merges this one and I open a new one.

Thanks for the proposal @dotlambda 👍
If it's okay with you, I'd say you create a new one because it addresses things that are out of scope for this particular PR and I can't work on it until evening.

@GrahamcOfBorg build python3Packages.pynacl python2Packages.pynacl

@FRidh Is there any reason why this isn't merged yet?