antigravity-acp fails on aarch64 with non-4K page size

[lelahx]

I'm guessing this issue needs to be floated up to the Antigravity team at Google, but I found no official channel for bug reports so I'm hoping the ACP Registry maintainers know who to talk to.

In short, I tried to use the Antigravity integration with Zed on my M2 Macbook running Asahi Linux, which is famously one of the few desktop platforms that uses a 16K page size CPU, whereas most historically used 4K. This frequently causes software with this (erroneous) assumption to fail when run on CPUs where it doesn't hold, such as mine. Usually such bugs quickly get fixed in open source software, since anyone can submit patches to fix them, and it usually is trivial to do so. But this ACP adapter is not open source, so here I am, opening this issue.

When opening a new thread with Antigravity ACP for the first time, it fails at the install step and spits out the following error message:

Server exited with status signal: 6 (SIGABRT) (core dumped)
F0824 00:44:27.079241   12030 page-size.cc:19] Check failed: sysconf(_SC_PAGE_SIZE) == kPageSize (16384 vs. 4096) Compile-time page size does not match runtime page size
*** Check failure stack trace: ***
    @     0xaaab3d8c53f8  absl::log_internal::LogMessage::Flush()
    @     0xaaab48ca0748  bamm2::internal::google_init_module_bamm2_init_verify_page_size()
    @     0xaaab3deb9c58  GoogleInitializer::TypeData::RunIfNecessary()
    @     0xaaab3deb996c  GoogleInitializer::RunInitializers()
    @     0xaaab483f6d84  RealInitGoogle()
    @     0xaaab47e8786c  Run()
    @     0xaaab47e87214  google3_base_python_clif_googleinit_clifwrap::wrapRun()
    @     0xaaab3de8b9d8  PyObject_Vectorcall
    @     0xaaab3e9c3750  _PyEval_EvalFrameDefault
    @     0xaaab3e9b6c88  _PyEval_Vector
    @     0xaaab405254a4  PyEval_EvalCode
    @     0xaaab48a9a918  builtin_exec
    @     0xaaab3de8ca7c  PyObject_Vectorcall
    @     0xaaab3e9bbfcc  _PyEval_EvalFrameDefault
    @     0xaaab3e9b6c88  _PyEval_Vector
    @     0xaaab3e1778a0  PyObject_CallMethod
    @     0xaaab481889c8  devtools::python_launcher::Launcher_Main()
    @     0xffff07496f1c  __libc_start_call_main
    @     0xffff0749705c  __libc_start_main@GLIBC_2.17
*** SIGABRT received by PID 12030 (TID 12030) on cpu 5, si_code=-6, from PID 12030; ***
E0824 00:44:27.123908   12030 coredump_hook.cc:309] RAW: Remote crash data gathering hook invoked.
E0824 00:44:27.123913   12030 client.cc:261] RAW: Coroner client retries enabled, will retry for up to 30 sec.
E0824 00:44:27.123914   12030 coredump_hook.cc:404] RAW: Sending fingerprint to remote end.
E0824 00:44:27.123926   12030 coredump_hook.cc:414] RAW: Cannot send fingerprint to Coroner: [NOT_FOUND] stat failed on crash reporting socket /var/google/services/logmanagerd/remote_coredump.socket (Is the listener running?): No such file or directory
E0824 00:44:27.123927   12030 coredump_hook.cc:466] RAW: Dumping core locally.
base/elfcore.c:2307 Failed to write mapping 13 at 0xffff06dc0000 of size 131072: Bad address(14)
F0824 00:44:27.079241   12030 page-size.cc:19] Check failed: sysconf(_SC_PAGE_SIZE) == kPageSize (16384 vs. 4096) Compile-time page size does not match runtime page size
E0824 00:44:27.178812   12030 process_state.cc:792] RAW: Raising signal 6 with default behavior

Also fails with the same error when manually running the agy_acp_server.par binary linked to in antigravity-acp/agent.json from this repo.

Thanks for hearing me out!