Conversation
…ss-platform CI/CD
- 01-RESEARCH.md: mark Open Questions RESOLVED (Q1 -> Plan 01-05 spike; Q2 -> Plan 01-01 accessors; Q3 -> cmaketemplate auto-download, no explicit zkLLVM CI step per D-09 interpretation) - 01-VALIDATION.md: fill per-task verification map from plan automated blocks; set status: ready; nyquist_compliant: true - Wave/depends_on rebucket: 01-02 wave 2 depends_on [01-01]; 01-03 wave 3 depends_on [01-01, 01-02]; 01-04 wave 3; 01-05 wave 4; 01-06 wave 5 - 01-01 Task 3: pin branch stacking — feature/gcs-globaldb-passthrough created FROM feature/app-restructure HEAD (locked at planning time) - 01-04 Task 3: commit to option C — gcs_init returns nullptr gracefully without GeniusSDK node; remove all gcs_on_message tests (moved to 01-05); drop init-success EXPECT_NE assertions - 01-05: owns all on_message tests; rephrase success criteria — no partial pass, Dart smoke suite passing is the pass condition with skip: only for live-node round-trip - 01-06: pin iOS toolchain — no -DCMAKE_TOOLCHAIN_FILE flag; build/iOS CMakeLists auto-sets to build/apple.toolchain.cmake - ROADMAP: annotate plan list with wave numbers and dependency notes
…doring Vendors SGProcessing test fixtures instead of depending on a sibling GeniusNetwork checkout, removes stale Vulkan-deadlock test skips now that the upstream fix is verified, and fixes SGProcessingBridge to return real output bytes (was returning a manifest hash) with support for windowed/overlapping input schemas.
…ARM FP4 evaluation
- Synthesizes RESEARCH.md into a decision-oriented verdict - MNN sgfp4-pivot: HIGH quality, complete for v2-only scope - GNUS-NEO-SWARM: NOT compatible; fp4_codec is orphaned NF4 dead code - Conditional workstream recommendation with neo config.json siting pattern
…ntegration with GNUS-NEO-SWARM neo workspace FP4 codec and SGProcessingManager verification layer; assess MNN-side implementation quality; scope potential new GNUS-NEO-SWARM workstream for required updates
… (converter integration) MNN's sgfp4-pivot grew from 4 to 12 phases since the original evaluation and now ships mnnconvert --sgfp4, closing the core "nothing produces a real SGFP4 .mnn" gap this report identified. Appends an addendum: the converter path is real and verified against the classic Interpreter/Session API, but corrects the original phases 1-4 verdict (CPU decoder was broken for real multi-tile weights, now fixed), and flags two new open risks (externalized- weight loading, unproven accuracy).
Conflict resolutions:
- App workstream planning + dart/flutter files + gcs_core.cpp: take develop
(live PR#11 line with D-29 ABI replan; our side was the stale pre-PR#11
app-workstream attempt).
- src/app/scaffold submodule: take develop (6016e3f4).
- GNUS-NEO-SWARM submodule: unified tip bc2d52d (app-restructure + phase-04
merged in submodule develop).
- cmake/CommonBuildParameters.cmake: union — keep our vk-bootstrap,
MoltenVK block, and vendored Vulkan-Headers set_target_properties; take
develop's ENV{VULKAN_SDK} syntax fix, top-level enable_testing(), and
gcs_test binary-dir rename.
…++17 Port the corrected implementation from the GNUS-NEO-SWARM twin (src/storage/gcs_global_db.cpp): - GeniusSDK.hpp/GeniusSDKGetNode() no longer exist in GeniusSDK dev_cognitive; use the installed C header and the opaque GeniusSDKGetPubSub() handle with a non-owning shared_ptr alias. - Add the missing <libp2p/basic/scheduler/asio_scheduler_backend.hpp> include for the AsioSchedulerBackend construction. - Replace the designated initializer in gcs_core.cpp (C7555 under /std:c++17) with ordered aggregate init (single-member Config). Bump GNUS-NEO-SWARM submodule to d3f941c (Vulkan header include fix).
The previous commit recorded the phase-13 branch tip (6af5fd1) because the submodule working tree was checked out there. For the dev_phase04 -> develop PR the pointer must be the submodule develop tip so phase-13 content lands via its own submodule PR, not here.
# Conflicts: # GNUS-NEO-SWARM # cmake/CommonBuildParameters.cmake # src/lib/gcs_core.cpp
…op's 3d9ebab pinned 7760ee2 which was never pushed to the scaffold remote
…alize The develop merge (b63b61e) textually combined our GeniusSDKGetPubSub() C-API acquisition with develop's node->GetGraphsyncNetwork() body, leaving 'node' undeclared and calling a nonexistent 1-arg Initialize overload. Restore develop's 0c68fb2 implementation verbatim: GeniusSDKGetNode() -> node->GetPubSub() / node->GetGraphsyncNetwork(), and the GeniusSDK.hpp include it requires (verified present in the installed local GeniusSDK package the build resolves).
… fix
- workflow_dispatch: add per-platform build toggles (build_<target>-<abi>
<build-type> inputs + exit 78 skip step) and thirdparty/supergenius/
geniussdk dep-tag overrides that default to the release tag, matching
the GeniusSDK/SuperGenius workflows
- Migrate Linux container jobs off EOL Debian Bullseye to
ghcr.io/geniusventures/almalinux-8:latest; split Configure Linux host
into dnf/apt branches and pin clang via
-DCMAKE_{C,CXX}_COMPILER=clang{,++} (AlmaLinux /usr/bin/c++ is a real
gcc-owned file, so the update-alternatives takeover fails silently)
- Create release tag now also runs on manual tag dispatches and treats
an existing release as reusable for the shared tag, with a
create-race retry for sibling matrix jobs
…iscovery Run 35652579381 (Linux x86_64 Release) died at configure: supergeniusTargets.cmake:363 set_target_properties on sgns::secure_storage references PkgConfig::LIBSECRET, which the prebuilt SuperGenius package does not define — every consumer must run its own pkg_check_modules before find_package(SuperGenius), exactly as GeniusSDK's cmake/CommonBuildParameters.cmake does. Add that discovery (Linux only) to GCS's cmake/CommonBuildParameters.cmake, guarded the same way. Also replace D-09 (cmaketemplate's configure-time zkLLVM auto-download, which always fetches develop) with an explicit Download zkLLVM release step modeled on GeniusSDK/SuperGenius: zkllvm_tag dispatch input (falls back to the tag input), Release-type artifacts for non-Windows (cmakemplate's own fallback mapping), assigner binaries excluded, and -DZKLLVM_BUILD_DIR passed to the Linux configure. Non-Linux configures keep the auto-download path for now (Linux is the current focus).
…egistration Run 35654341772 (Linux builds now green, tests red): 1. test_gcs_ffi (Release+Debug): gnome-keyring-daemon dies with 'Operation not permitted' in the AlmaLinux container, so org.freedesktop.secrets never comes up and node boot fails with 'Account creation failed'. Add options: --cap-add=IPC_LOCK to the job container (SuperGenius precedent) so the daemon can mlock. 2. test_gcs_ffi_sdk / test_gcs_ffi_coldboot (Debug only): MNN aborts at Backend.cpp:135 MNNInsertExtraRuntimeCreator 'duplicate type'. These binaries link libgcs_ffi.so (which statically embeds MNN via gcs_core -> neoswarm_core) AND gcs_storage, whose SuperGenius chain (SGProcessors) whole-archives MNN into the executable - two MNN registries in one process. MNN's registry map is a function-local static (one per copy) but its registration functions are global, so on ELF the exe's symbols interpose into the .so and the second registration lands in the exe's already-populated map. Debug-only because MNN_ASSERT is plain assert (NDEBUG compiles it out); OSX historically passed because dyld's two-level namespace binds each image to its own symbols. Fix: -Wl,-Bsymbolic-functions on gcs_ffi so the dylib binds its internal MNN calls locally (functions only - vtables/typeinfo keep default binding).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Requires updates to thirdparty, supergenius, sgprocessingmanager.