Skip to content

Remove stray files from #729, fix changelog, update changelog policy #735

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
CarloLucibello merged 3 commits into from
Nov 11, 2026
Merged

Remove stray files from #729, fix changelog, update changelog policy #735

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
aa6e064
Remove local setup files and notes committed in #729
CarloLucibello Nov 11, 2026
0012d26
Fix changelog inconsistencies
CarloLucibello Nov 11, 2026
c2ed7d6
Keep changelog edits out of PRs in AGENTS.md
CarloLucibello Nov 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 0 additions & 2 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,8 +6,6 @@
*.swo
Manifest.toml
.vscode
.tools/
.julia-depot/
LocalPreferences.toml
.DS_Store
test.jl
Expand Down
2 changes: 1 addition & 1 deletion AGENTS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,5 +69,5 @@ Layers typically implement their forward pass via `propagate(fmsg, g, aggr; xi,
- **GPU support** is provided through package extensions (`ext/`, e.g. `GNNGraphsCUDAExt`, `GNNlibCUDAExt`/`AMDGPUExt`/`MetalExt`) gated on weakdeps — don't add a hard GPU dependency.
- **Mooncake** and **SimpleWeightedGraphs** support are also extensions; keep them optional.
- **Versioning**: each PR should bump the `version` in the `Project.toml` of every affected package per semver, using a `-DEV` suffix for unreleased changes (e.g. `1.18.0-DEV`), and update inter-package `[compat]` bounds when a frontend needs newer GNNGraphs/GNNlib.
- **Changelog**: there is a single repo-wide `CHANGELOG.md` at the root, organized as one timeline (most recent first), with each release headed `## <Package>.jl <version> — <date>`. Any user-facing change (new feature, bug fix, deprecation, or compat relaxation) must add an entry under the relevant package's latest (or `Unreleased`) heading, in [Keep a Changelog](https://keepachangelog.com/) style, linking the PR (`[#NNN]` reference at the bottom of the file). Skip purely internal churn (tests, CI, docs, refactors). When tagging a release, turn the package's `Unreleased`/`-DEV` heading into a dated `<version> — YYYY-MM-DD` heading placed at the top of the timeline.
- **Changelog**: do **not** edit `CHANGELOG.md` in feature or bug-fix PRs. It is maintained separately, because concurrent PRs editing the same section conflict and force constant rebases. Describe user-facing changes clearly in the PR description instead, so they can be carried over. When explicitly asked to update the changelog: it is a single repo-wide timeline (most recent first) with each release headed `## <Package>.jl <version> - <date>`, in [Keep a Changelog](https://keepachangelog.com/) style. Add user-facing changes (new feature, bug fix, deprecation, or compat relaxation) under the package's latest `-DEV` heading, linking each PR (`[#NNN]` reference at the bottom of the file), and skip purely internal churn (tests, CI, docs, refactors). When tagging a release, turn the package's `-DEV` heading into a dated `<version> - YYYY-MM-DD` heading placed at the top of the timeline.
- Follow the `[workspace]` dev model; do not rewire packages to local paths outside the repo.
61 changes: 31 additions & 30 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,28 +13,30 @@ Entries link to the pull request that introduced them.
## GNNLux.jl 0.3.1-DEV

**Fixed**
- `MEGNetConv` now returns the updated state from its edge and node update layers, so stateful sublayers retain their state across calls ([#734]).
- `NNConv` now returns its state as `(; nn = ...)`, the same structure it receives. It previously returned the inner network's state without the `nn` wrapper, so passing the returned state back into the layer (e.g. on the second step of `Lux.Training.single_train_step!`) failed with a `FieldError` ([#728]).
- `MEGNetConv` now returns the updated states of its edge and node update networks `ϕe` and `ϕv`. It previously returned the input state unchanged, so stateful sublayers (e.g. `BatchNorm` running statistics) lost their updates between calls ([#729]).

## GNNGraphs.jl 1.6.1-DEV

**Fixed**
- Fixed Zygote differentiation through `add_edges` with non-empty edge metadata by iterating over a collected vector of metadata keys. `sample_neighbors(...; replace=true)` now skips zero-degree nodes instead of trying to sample from an empty neighbor list ([#729]).
- `remove_nodes` now updates the `graph_indicator` of batched graphs. It was previously left unchanged, so calling `unbatch` or `getgraph` on the result threw a `BoundsError` ([#717]).
- Fixed Zygote crashing with a segmentation fault when differentiating through `add_edges`. Gradients now flow through the added edge weights and edge features ([#729]).
- `sample_neighbors(g, nodes, K; replace = true)` now skips nodes without neighbors instead of failing to sample from an empty neighbor list ([#729]).

## GNNGraphs.jl 1.6.0 - 2026-11-04

**Added**
- Added a `Mooncake` package extension, so that Mooncake can differentiate on CUDA through `add_self_loops` and through adjacency-matrix graphs. Float edge weights of adjacency-matrix graphs stay differentiable ([#704]).

**Fixed**
- `remove_nodes` now updates the `graph_indicator` of batched graphs. It was previously left unchanged, so calling `unbatch` or `getgraph` on the result threw a `BoundsError` ([#717]).
- Fixed `scaled_laplacian` (and so `ChebConv`) crashing with a segmentation fault on AMDGPU graphs: the eigensolver's start vector was created on the CPU, so CPU BLAS was handed GPU memory. It is now created on the same device as the graph ([#722]).
- Added an `EnzymeCore` package extension marking `scaled_laplacian` inactive for Enzyme, matching its existing `@non_differentiable` ChainRules declaration. Enzyme previously differentiated the Krylov eigensolve inside it and failed; this unblocks `ChebConv`, `GConvGRUCell` and `GConvLSTMCell` ([#706]).
- Enzyme can now differentiate through the adjacency-matrix → COO graph conversion: a new internal keyword-free helper `_to_coo_graph` avoids the union-typed keyword handling of the `GNNGraph(g; graph_type)` constructor and of `to_coo`, which Enzyme's type analysis cannot compile. Together with upstream fixes in Enzyme ≥ 0.13.197 this removes the `EnzymeInternalError` crash on `:dense`/`:sparse` graphs ([#703]).

## GNNLux.jl 0.3.0 - 2026-11-04

**Fixed**
- `GNNChain` now returns its state with the same structure it receives, `(; layers = (; layer_1, ...))`, as the Lux layer interface requires. It previously returned the per-layer states without the `layers` wrapper, so feeding the returned state back into the chain (as `Lux.Training.single_train_step!` does) failed with a `FieldError`. This is breaking for code that worked around the bug by re-wrapping the state as `(; layers = st)`, as the docs examples did: drop the wrapper.
- `NNConv` now returns its state as `(; nn = ...)`, the same structure it receives. It previously returned the inner network's state without the `nn` wrapper, so passing the returned state back into the layer (e.g. on the second step of `Lux.Training.single_train_step!`) failed with a `FieldError` ([#728]).
- `GNNChain` now returns its state with the same structure it receives, `(; layers = (; layer_1, ...))`, as the Lux layer interface requires. It previously returned the per-layer states without the `layers` wrapper, so feeding the returned state back into the chain (as `Lux.Training.single_train_step!` does) failed with a `FieldError`. This is breaking for code that worked around the bug by re-wrapping the state as `(; layers = st)`, as the docs examples did: drop the wrapper ([#715]).

## GNNlib.jl 1.4.1 - 2026-11-04

Expand All @@ -48,7 +50,7 @@ Entries link to the pull request that introduced them.
- `GNNRecurrence` now hands each cell an indexed time slice instead of an `eachslice` view, matching the Lux frontend. Enzyme's type analysis fails on `SubArray` cell inputs, so this unblocks `DCGRU`, `EvolveGCNO`, `GConvGRU` and `GConvLSTM` under Enzyme ([#707]).
- The recurrent temporal cells now delegate their forward-pass math to shared `GNNlib` functions (requires GNNlib ≥ 1.4); the layer behaviour is unchanged ([#696]).

## GNNLux.jl 0.2.0 — 2026-07-22
## GNNLux.jl 0.2.0 - 2026-07-22

**Added**
- Added pooling layers (`GlobalPool`, `GlobalAttentionPool`, `TopKPool`) ([#576]).
Expand All @@ -62,15 +64,15 @@ Entries link to the pull request that introduced them.
**Removed**
- Removed the `A3TGCN` layer, mirroring its removal from the Flux frontend ([#696]).

## GNNlib.jl 1.4.0 — 2026-07-22
## GNNlib.jl 1.4.0 - 2026-07-22

**Added**
- Added framework-agnostic forward passes for the recurrent temporal cells — `tgcn`, `gconv_gru`, `gconv_lstm`, `dcgru` — shared by the Flux and Lux frontends ([#696]).

**Fixed**
- Worked around a Julia 1.12 code-generation segfault that crashed `CGConv` and `GMMConv` gradients: the layers' `@warn` (in the residual branch) is now wrapped in `ignore_derivatives` to keep the logging macro out of the AD-differentiated code path. Root cause reported upstream as [FluxML/Zygote.jl#1662]; re-enables the previously disabled `CGConv`/`GMMConv` gradient tests ([#695]).

## GNNGraphs.jl 1.5.1 — 2026-07-22
## GNNGraphs.jl 1.5.1 - 2026-07-22

**Changed**
- Relaxed the `CUDA` compat bound to `"5, 6"`, adding support for CUDA.jl v6 ([#690]).
Expand All @@ -81,9 +83,9 @@ Entries link to the pull request that introduced them.
- Fixed `getgraph` failing on GPU graphs ([#691]).
- Fixed `remove_self_loops` mutating the input for adjacency-matrix graphs ([#659]).
- Fixed `unbatch` for COO batches containing zero-edge graphs ([#652]).
- Fixed `sample_nbrs` when sampling without replacement ([#648]).
- Fixed `NeighborLoader` sampling neighbors with replacement instead of without ([#648]).

## GNNlib.jl 1.3.0 — 2026-07-22
## GNNlib.jl 1.3.0 - 2026-07-22

**Added**
- Added Mooncake rules for the `propagate` `copy_xj`, `e_mul_xj`, and `w_mul_xj` fast paths ([#677], [#678], [#679]).
Expand All @@ -93,25 +95,25 @@ Entries link to the pull request that introduced them.
- Bumped the NNlib compat bound ([#687]).

**Fixed**
- Fixed the `d_conv` (`DConv`) forward pass, which did not compute the diffusion convolution from the paper: the degree normalization was inverted (multiplying by degree instead of `D_O⁻¹`/`D_I⁻¹`), one weight slice was applied twice, and a spurious Chebyshev `2T − T₀` recurrence was used in place of the paper's bidirectional random-walk power series. Isolated nodes are now handled without producing `NaN` ([#592]).
- Fixed the `d_conv` (`DConv`) forward pass, which did not compute the diffusion convolution from the paper: the degree normalization was inverted (multiplying by degree instead of `D_O⁻¹`/`D_I⁻¹`), one weight slice was applied twice, and a spurious Chebyshev `2T − T₀` recurrence was used in place of the paper's bidirectional random-walk power series. Isolated nodes are now handled without producing `NaN` ([#693]).

## GNNlib.jl 1.2.1 — 2026-04-29
## GNNlib.jl 1.2.1 - 2026-04-29

**Changed**
- Bumped the DataStructures compat bound ([#670]).

**Fixed**
- Fixed the `gmm_conv` (`GMMConv`) implementation ([#645]).

## GNNlib.jl 1.2.0 — 2026-01-11
## GNNlib.jl 1.2.0 - 2026-01-11

**Added**
- Added support for reducing nodes over heterogeneous graphs ([#634]).

**Fixed**
- Fixed empty-edge handling in `softmax_edge_neighbors` ([#636]).

## GNNGraphs.jl 1.5.0 — 2025-12-24
## GNNGraphs.jl 1.5.0 - 2025-12-24

**Added**
- Added `coalesce` and `is_coalesced` to sort and merge duplicate edges of COO graphs, and to query coalesced state ([#613], [#624], [#625]).
Expand All @@ -125,7 +127,7 @@ Entries link to the pull request that introduced them.
**Fixed**
- Fixed the NNlib CUDA extension ([#621]).

## GNNlib.jl 1.1.0 — 2025-12-24
## GNNlib.jl 1.1.0 - 2025-12-24

**Added**
- Added SpMM-based message passing with CUDA support for coalesced COO graphs ([#617]).
Expand All @@ -138,15 +140,15 @@ Entries link to the pull request that introduced them.
**Fixed**
- Fixed the NNlib CUDA extension ([#621]).

## GraphNeuralNetworks.jl 1.1.0 — 2025-12-24
## GraphNeuralNetworks.jl 1.1.0 - 2025-12-24

**Added**
- `TGCN` now supports non-linear activation functions ([#596]).

**Fixed**
- Fixed a Zygote differentiation error ([#579]).

## GNNGraphs.jl 1.4.2 — 2025-02-07
## GNNGraphs.jl 1.4.2 - 2025-02-07

**Changed**
- Improved type inference for `num_nodes` ([#588]).
Expand All @@ -156,16 +158,16 @@ Entries link to the pull request that introduced them.
- Fixed a corner case in `normalize_graphdata` ([#589]).
- Fixed a Zygote differentiation error ([#579]).

## GNNlib.jl 1.0.1 — 2025-01-12
## GNNlib.jl 1.0.1 - 2025-01-12

Maintenance release: internal fixes and test-infrastructure updates, no user-facing API changes.

## GNNGraphs.jl 1.4.1 — 2024-12-25
## GNNGraphs.jl 1.4.1 - 2024-12-25

**Changed**
- Removed the constraint requiring an equal number of features across node/edge types in the `gdata` of heterographs ([#570]).

## GNNGraphs.jl 1.4.0 — 2024-12-21
## GNNGraphs.jl 1.4.0 - 2024-12-21

**Added**
- Added `broadcast`, `iterate`, and `setindex!` support for `TemporalSnapshotsGNNGraph` ([#563]).
Expand All @@ -176,27 +178,27 @@ Maintenance release: internal fixes and test-infrastructure updates, no user-fac
**Fixed**
- Fixed `show` for graphs whose features are not arrays ([#564]).

## GNNlib.jl 1.0.0 — 2024-12-21
## GNNlib.jl 1.0.0 - 2024-12-21

First stable release of GNNlib.jl.

**Changed**
- Rewrote the recurrent temporal layers for Flux v0.16 ([#560]).
- Updated for compatibility with Flux v0.15 ([#550]).

## GraphNeuralNetworks.jl 1.0.0 — 2024-12-21
## GraphNeuralNetworks.jl 1.0.0 - 2024-12-21

First stable release of the Flux frontend.

**Changed**
- Rewrote the recurrent temporal layers for Flux v0.16 ([#560]).
- Updated for compatibility with Flux v0.15 ([#550]).

## GNNLux.jl 0.1.1 — 2024-12-09
## GNNLux.jl 0.1.1 - 2024-12-09

Documentation release: added the "Hands On" tutorial, a version selector, and general docs improvements ([#549], [#543], [#542], [#539]). No API changes.

## GNNLux.jl 0.1.0 — 2024-12-02
## GNNLux.jl 0.1.0 - 2024-12-02

Initial release of the Lux-based frontend for GraphNeuralNetworks.jl, providing
Lux implementations of the graph convolutional, pooling, and temporal layers
Expand All @@ -218,7 +220,6 @@ Lux implementations of the graph convolutional, pooling, and temporal layers
[#587]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/587
[#588]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/588
[#589]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/589
[#592]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/issues/592
[#596]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/596
[#601]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/601
[#605]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/605
Expand Down Expand Up @@ -249,17 +250,17 @@ Lux implementations of the graph convolutional, pooling, and temporal layers
[#687]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/687
[#690]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/690
[#691]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/691
[#623]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/issues/623
[#693]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/693
[#695]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/695
[#696]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/696
[#703]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/703
[#704]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/704
[#707]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/707
[#706]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/706
[#707]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/707
[#715]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/715
[#717]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/717
[#722]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/722
[#718]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/718
[#729]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/729
[#722]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/722
[#728]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/728
[#734]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/734
[#729]: https://github.com/JuliaGraphs/GraphNeuralNetworks.jl/pull/729
[FluxML/Zygote.jl#1662]: https://github.com/FluxML/Zygote.jl/issues/1662
27 changes: 0 additions & 27 deletions DEVELOPMENT.md
View file
Open in desktop

This file was deleted.

27 changes: 0 additions & 27 deletions GNNLux/reproduce_issue_732.jl
View file
Open in desktop

This file was deleted.

Loading
Loading