ai/github-copilot-cli 1.0.75 MidnightBSD

The program header relocation needs python, but python:build registers a BUILD_DEPENDS, and those are only installed at the start of the configure phase -- well after post-extract runs: make: exec(/usr/local/bin/python3.12) failed (No such file or directory) Use python:patch and do the work in post-patch, which runs after patch-depends. The relocation now reports "moved 14 program headers to offset 64" and the port packages. AI-Assisted-by: Claude Opus 5 (1M context) <[email protected]> Signed-off-by: Lucas Holt <[email protected]>

Replace the npm-built port with the upstream copilot-linux-x64 release running under Linux emulation, and keep the old npm-based port as ai/github-copilot-cli-legacy. The upstream binary is a single-file Node SEA: postject injects a large LOAD/NOTE segment and relocates the program header table ~45MB into the file. The kernel image activator only accepts program headers within the first page (sys/kern/imgact_elf.c), so exec(2) failed with ENOEXEC and the installed copilot binary could not run at all. files/relocate-phdrs.py rewrites the table over the stale pre-injection copy at offset 64, dropping PT_PHDR so the 14 remaining entries fit in front of .interp. The kernel then derives AT_PHDR from the PT_LOAD that maps file offset 0, so ld.so and the dl_iterate_phdr lookup the SEA uses …