From: Kirill A. Korinsky <kirill@korins.ky>
Subject: Re: Update llama.cpp
To: Percy Piper <piper.percy@googlemail.com>
Cc: ports@openbsd.org
Date: Thu, 11 Sep 2025 19:24:20 +0200

On Thu, 11 Sep 2025 17:52:07 +0200,
Percy Piper <piper.percy@googlemail.com> wrote:
> 
> This update brings some nice improvements (e.g. gpt-oss support[1]).
> 
> However, I am unable to resolve a blocking issue. During the build
> -I/usr/local/include appears before the source's include dirs, so any
> previously installed headers are found before those in the sources,
> causing the build to fail.
> 
> If the old package is uninstalled first, the port will build, package
> and run fine.
> 
> This seems likely to be a common scenario and I'm embarrased I haven't
> figured out how to solve it.
> 
> I had hoped to get this in shape before 7.8 so welcome any suggestions
> or hints on how this might be solved?
> 

Have you tried add BEFORE in target_include_directories and
include_directories? Like this:

Index: src/CMakeLists.txt
--- src/CMakeLists.txt.orig
+++ src/CMakeLists.txt
@@ -37,8 +37,8 @@ add_library(llama
             unicode.h
             )
 
-target_include_directories(llama PRIVATE .)
-target_include_directories(llama PUBLIC ../include)
+target_include_directories(llama BEFORE PRIVATE .)
+target_include_directories(llama BEFORE PUBLIC ../include)
 target_compile_features   (llama PRIVATE cxx_std_17) # don't bump
 
 target_link_libraries(llama PUBLIC ggml)



maybe it needed to add AFTER for other include dirs as well.

-- 
wbr, Kirill