Build NIF 2.18 precompiled artifacts on OTP 29, release v0.8.22

.github/workflows/precompile.yml

@@ -18,10 +18,12 @@ jobs:
             otp: '27.0'
             elixir: '1.18'
             backend: metal
+          # OTP 29 is the first release to report NIF 2.18; OTP 27/28 both
+          # report 2.17, so building on 28 would just overwrite the 27 artifact.
           - os: macos-14
             target: aarch64-apple-darwin
-            otp: '28.0'
-            elixir: '1.18'
+            otp: '29.0'
+            elixir: '1.20'
             backend: metal
           - os: ubuntu-22.04
             target: x86_64-linux-gnu
@@ -30,8 +32,8 @@ jobs:
             backend: cpu
           - os: ubuntu-22.04
             target: x86_64-linux-gnu
-            otp: '28.0'
-            elixir: '1.18'
+            otp: '29.0'
+            elixir: '1.20'
             backend: cpu
 
     runs-on: ${{ matrix.os }}

CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## v0.8.22
+
+### Fixed
+
+- **Precompiled NIF 2.18 artifacts** — `mix.exs` advertises precompiled NIFs for NIF versions 2.17 and 2.18, but the precompile workflow built each target on OTP 27 *and* OTP 28, and both of those report NIF 2.17 (2.18 only arrived with OTP 29). The two jobs therefore produced identically named `nif-2.17` tarballs that overwrote each other on upload, and no 2.18 artifact was ever published — so installing on OTP 29 (NIF 2.18) failed with a 404 when fetching the precompiled binary. The precompile matrix now builds on OTP 27 (NIF 2.17) and OTP 29 (NIF 2.18), publishing both NIF versions for each target. The llama.cpp submodule is unchanged (4c6595503, tag b9601).
+
 ## v0.8.21
 
 ### Changed

docs/release-guide.md

@@ -96,7 +96,7 @@ git push origin vX.Y.Z
 ```
 
 The tag push triggers the **precompile workflow** (`.github/workflows/precompile.yml`) which:
-1. Builds precompiled NIFs for macOS (Metal) and Linux (CPU) across OTP 27 and 28
+1. Builds precompiled NIFs for macOS (Metal) and Linux (CPU) across OTP 27 (NIF 2.17) and OTP 29 (NIF 2.18)
 2. Uploads `.tar.gz` artifacts to the GitHub release
 3. Runs `mix elixir_make.checksum --all --ignore-unavailable` and auto-commits `checksum.exs` to master
 

mix.exs

@@ -37,7 +37,7 @@ end
 defmodule LlamaCppEx.MixProject do
   use Mix.Project
 
-  @version "0.8.21"
+  @version "0.8.22"
   @source_url "https://github.com/nyo16/llama_cpp_ex"
 
   def project do