Bump the all-julia-packages group across 3 directories with 4 updates

[dependabot]

Updates the requirements on DiffEqBase, CUDA, SciMLBase and DiffEqDevTools to permit the latest version.
Updates DiffEqBase to 7.0.0

Changelog

Sourced from DiffEqBase's changelog.

OrdinaryDiffEq.jl v7 Breaking Changes

This release bumps to SciMLBase v3, RecursiveArrayTools v4, and includes breaking changes across DiffEqBase, OrdinaryDiffEqCore, and all solver sublibraries.

Themes of the v7 release

Most of the breaking changes fall into a small set of recurring themes. Keep these in mind while reading the migration table — they explain why an individual change exists and often suggest the right migration direction:

• Time to first solve (TTFS) reduction. Direct deps on Static.jl, StaticArrayInterface.jl, Polyester.jl, and StaticArrays.jl were dropped; using OrdinaryDiffEq now loads only the default solver set; ODEFunction switched to AutoSpecialize. All of this means less code loaded and more precompilation caching on first solve.
• Type stability everywhere. All Bool solver/solve keyword arguments (autodiff, verbose, alias, lazy, …) were replaced by typed objects. Passing a Bool no longer changes dispatch in ways the compiler cannot specialize on, and the reverse is no longer allowed to silently fall back through slow generic paths.
• Generality beyond ForwardDiff. chunk_size, diff_type, standardtag, etc. encoded ForwardDiff-specific or FiniteDiff-specific knobs on every solver. They are replaced by the ADTypes interface (AutoForwardDiff, AutoFiniteDiff, AutoEnzyme, AutoZygote, …) so every solver automatically generalizes to any AD backend.
• Controller is now an object, not a pile of solve kwargs. gamma, beta1, beta2, qmin, qmax, qsteady_min, qsteady_max, qoldinit were moved onto concrete PIController / PIDController / IController / PredictiveController structs, and EEst moved to the controller cache. This is prep work for pluggable controllers and removes a large amount of dead state from the integrator struct.
• Cleanup of old re-exports / deprecations. Functions like has_destats (now has_stats), sol.destats (now sol.stats), DEAlgorithm/DEProblem/DESolution abstract types, tuples()/intervals(), QuadratureProblem, fastpow, concrete_solve, etc. were on a deprecation path for one or more minor releases. v7 removes them.

Recommended upgrade path

The cleanest path is not to jump straight from an old environment onto v7. Most renamed APIs (e.g. DEAlgorithm → AbstractDEAlgorithm, u_modified! → derivative_discontinuity!, has_destats → has_stats, sol.destats → sol.stats, the construct* tableau functions, alias_u0/alias_du0, beta1/beta2, PID kwargs) already exist under their new names in SciMLBase v2 / OrdinaryDiffEq v6 with deprecation warnings. The recommended sequence is:

1. Stay on SciMLBase v2 / OrdinaryDiffEq v6. Update your code to the new names (has_stats, sol.stats, AbstractDEAlgorithm, derivative_discontinuity!, ODEAliasSpecifier, ODEVerbosity, ADTypes-based autodiff, explicit controller = … objects, new tableau names) while the deprecation shims still exist.
2. Verify your tests pass on v6 with no deprecation warnings.
3. Then bump to v7. At this point your code should compile and run against v7 without further changes aside from the genuinely new breakage (RAT v4 array semantics, ensemble prob_func/output_func signature, struct type parameter removals, kwargs that truly no longer exist, default changes like CheckInit and williamson_condition=false).

Doing it in two steps keeps the diff small per step and lets the deprecation warnings on v6 point you at the exact call sites that will break on v7.

Fallback for RAT v4 indexing

If you cannot update sol[i] / length(sol) / eachindex(sol) call sites yet (see the RAT v4 table below), you can opt back into v3 semantics on a per-solution basis by converting the container type to the ragged variant:

using RecursiveArrayToolsRaggedArrays
sol_old = RaggedVectorOfArray(sol)   # indexes like v3: sol_old[i] is the i-th timestep

RecursiveArrayToolsRaggedArrays.jl preserves the previous AbstractVectorOfArray indexing behavior (timestep-first, not element-first). This is the escape hatch for code that assumes sol[i] returns the i-th timestep. It is, however, recommended that you update to the sol.u[i] / sol[:, i] style — the ragged wrapper is a compatibility layer, not the canonical API going forward.

---

RecursiveArrayTools v4

ODESolution is now an AbstractArray

AbstractVectorOfArray (the parent type of ODESolution, RODESolution, DAESolution, etc.) now subtypes AbstractArray. This changes the semantics of several common operations:

Operation	v3 (old)	v4 (new)	Migration
sol[i]	Returns i-th timestep (Vector)	Returns i-th scalar element (column-major)	Use sol.u[i] or sol[:, i]
length(sol)	Number of timesteps	prod(size(sol)) (total elements)	Use length(sol.t) or length(sol.u)
eachindex(sol)	1:nsteps	CartesianIndices(size(sol))	Use eachindex(sol.u)
iterate(sol)	Iterates over timesteps	Iterates over scalar elements	Use for u in sol.u
first(sol) / last(sol)	First/last timestep	First/last scalar element	Use first(sol.u) / last(sol.u)

... (truncated)

Commits

• See full diff in compare view

Updates CUDA to 6.0.0

Release notes

Sourced from CUDA's releases.

v6.0.0

CUDA v6.0.0

Diff since v5.11.0

Breaking changes: Ideally none. CUDA.jl has been split into subpackages, which is a major change, so the major version bump is out of caution. Deprecations have been introduced though, e.g., submodules like CUBLAS are now proper packages, cuBLAS.jl.

This release has been identified as a backport. Automated changelogs for backports tend to be wildly incorrect. Therefore, the list of issues and pull requests is hidden.

Commits

• a9a687c Add licenses and READMEs to subpackages.
• a79b516 Profiler: relax parsing of color flags in marker data. (#3075)
• 2e540fd Remove stale NEWS.md
• 0b05451 Put all GPU-only functions in the GPU method table (#2998)
• cbd36b0 Bump docker/setup-buildx-action from 3 to 4 (#3071)
• 74e1705 Bump docker/metadata-action from 5 to 6 (#3070)
• 9877922 Bump docker/build-push-action from 6 to 7 (#3069)
• 19c3638 Bump docker/login-action from 3 to 4 (#3068)
• 87da928 Mark @device_override and friends as public (#3066)
• 441fec5 Make CUDA & friends loadable on systems without NVPTX LLVM backend (#3067)
• Additional commits viewable in compare view

Updates DiffEqBase to 7.0.0

Changelog

Sourced from DiffEqBase's changelog.

OrdinaryDiffEq.jl v7 Breaking Changes

This release bumps to SciMLBase v3, RecursiveArrayTools v4, and includes breaking changes across DiffEqBase, OrdinaryDiffEqCore, and all solver sublibraries.

Themes of the v7 release

Most of the breaking changes fall into a small set of recurring themes. Keep these in mind while reading the migration table — they explain why an individual change exists and often suggest the right migration direction:

• Time to first solve (TTFS) reduction. Direct deps on Static.jl, StaticArrayInterface.jl, Polyester.jl, and StaticArrays.jl were dropped; using OrdinaryDiffEq now loads only the default solver set; ODEFunction switched to AutoSpecialize. All of this means less code loaded and more precompilation caching on first solve.
• Type stability everywhere. All Bool solver/solve keyword arguments (autodiff, verbose, alias, lazy, …) were replaced by typed objects. Passing a Bool no longer changes dispatch in ways the compiler cannot specialize on, and the reverse is no longer allowed to silently fall back through slow generic paths.
• Generality beyond ForwardDiff. chunk_size, diff_type, standardtag, etc. encoded ForwardDiff-specific or FiniteDiff-specific knobs on every solver. They are replaced by the ADTypes interface (AutoForwardDiff, AutoFiniteDiff, AutoEnzyme, AutoZygote, …) so every solver automatically generalizes to any AD backend.
• Controller is now an object, not a pile of solve kwargs. gamma, beta1, beta2, qmin, qmax, qsteady_min, qsteady_max, qoldinit were moved onto concrete PIController / PIDController / IController / PredictiveController structs, and EEst moved to the controller cache. This is prep work for pluggable controllers and removes a large amount of dead state from the integrator struct.
• Cleanup of old re-exports / deprecations. Functions like has_destats (now has_stats), sol.destats (now sol.stats), DEAlgorithm/DEProblem/DESolution abstract types, tuples()/intervals(), QuadratureProblem, fastpow, concrete_solve, etc. were on a deprecation path for one or more minor releases. v7 removes them.

Recommended upgrade path

The cleanest path is not to jump straight from an old environment onto v7. Most renamed APIs (e.g. DEAlgorithm → AbstractDEAlgorithm, u_modified! → derivative_discontinuity!, has_destats → has_stats, sol.destats → sol.stats, the construct* tableau functions, alias_u0/alias_du0, beta1/beta2, PID kwargs) already exist under their new names in SciMLBase v2 / OrdinaryDiffEq v6 with deprecation warnings. The recommended sequence is:

1. Stay on SciMLBase v2 / OrdinaryDiffEq v6. Update your code to the new names (has_stats, sol.stats, AbstractDEAlgorithm, derivative_discontinuity!, ODEAliasSpecifier, ODEVerbosity, ADTypes-based autodiff, explicit controller = … objects, new tableau names) while the deprecation shims still exist.
2. Verify your tests pass on v6 with no deprecation warnings.
3. Then bump to v7. At this point your code should compile and run against v7 without further changes aside from the genuinely new breakage (RAT v4 array semantics, ensemble prob_func/output_func signature, struct type parameter removals, kwargs that truly no longer exist, default changes like CheckInit and williamson_condition=false).

Doing it in two steps keeps the diff small per step and lets the deprecation warnings on v6 point you at the exact call sites that will break on v7.

Fallback for RAT v4 indexing

If you cannot update sol[i] / length(sol) / eachindex(sol) call sites yet (see the RAT v4 table below), you can opt back into v3 semantics on a per-solution basis by converting the container type to the ragged variant:

using RecursiveArrayToolsRaggedArrays
sol_old = RaggedVectorOfArray(sol)   # indexes like v3: sol_old[i] is the i-th timestep

RecursiveArrayToolsRaggedArrays.jl preserves the previous AbstractVectorOfArray indexing behavior (timestep-first, not element-first). This is the escape hatch for code that assumes sol[i] returns the i-th timestep. It is, however, recommended that you update to the sol.u[i] / sol[:, i] style — the ragged wrapper is a compatibility layer, not the canonical API going forward.

---

RecursiveArrayTools v4

ODESolution is now an AbstractArray

AbstractVectorOfArray (the parent type of ODESolution, RODESolution, DAESolution, etc.) now subtypes AbstractArray. This changes the semantics of several common operations:

Operation	v3 (old)	v4 (new)	Migration
sol[i]	Returns i-th timestep (Vector)	Returns i-th scalar element (column-major)	Use sol.u[i] or sol[:, i]
length(sol)	Number of timesteps	prod(size(sol)) (total elements)	Use length(sol.t) or length(sol.u)
eachindex(sol)	1:nsteps	CartesianIndices(size(sol))	Use eachindex(sol.u)
iterate(sol)	Iterates over timesteps	Iterates over scalar elements	Use for u in sol.u
first(sol) / last(sol)	First/last timestep	First/last scalar element	Use first(sol.u) / last(sol.u)

... (truncated)

Commits

• See full diff in compare view

Updates CUDA to 6.0.0

Release notes

Sourced from CUDA's releases.

v6.0.0

CUDA v6.0.0

Diff since v5.11.0

Breaking changes: Ideally none. CUDA.jl has been split into subpackages, which is a major change, so the major version bump is out of caution. Deprecations have been introduced though, e.g., submodules like CUBLAS are now proper packages, cuBLAS.jl.

This release has been identified as a backport. Automated changelogs for backports tend to be wildly incorrect. Therefore, the list of issues and pull requests is hidden.

Commits

• a9a687c Add licenses and READMEs to subpackages.
• a79b516 Profiler: relax parsing of color flags in marker data. (#3075)
• 2e540fd Remove stale NEWS.md
• 0b05451 Put all GPU-only functions in the GPU method table (#2998)
• cbd36b0 Bump docker/setup-buildx-action from 3 to 4 (#3071)
• 74e1705 Bump docker/metadata-action from 5 to 6 (#3070)
• 9877922 Bump docker/build-push-action from 6 to 7 (#3069)
• 19c3638 Bump docker/login-action from 3 to 4 (#3068)
• 87da928 Mark @device_override and friends as public (#3066)
• 441fec5 Make CUDA & friends loadable on systems without NVPTX LLVM backend (#3067)
• Additional commits viewable in compare view

Updates SciMLBase to 3.3.0

Release notes

Sourced from SciMLBase's releases.

v3.3.0

SciMLBase v3.3.0

Diff since v3.2.0

Merged pull requests:

• Add Mooncake rrules for symbolic indexing of solutions (#1314) (@​ChrisRackauckas-Claude)
• Mooncake observable AD: NonlinearSolution, vector, DAE, init param-grad (#1317) (@​ChrisRackauckas-Claude)

Closed issues:

• Remaining Mooncake SymbolicIndexingInterface AD gaps (#1316)

Commits

• a91f0e1 Update Project.toml
• 556fd80 Merge pull request #1317 from ChrisRackauckas-Claude/mooncake-observable-foll...
• 2aace14 Mooncake observable AD: NonlinearSolution, vector, DAE, init param-grad
• dddf175 Merge pull request #1314 from ChrisRackauckas-Claude/mooncake-symbolic-indexi...
• d8e49f5 Propagate parameter gradient through observable indexing rrule
• 2a90d97 Enable Mooncake test for ODE observable autodiff
• 480674f Add Mooncake rrules for symbolic indexing of solutions
• 22ed166 Bump version from 3.1.0 to 3.2.0
• 8569c5c Merge pull request #1214 from ChrisRackauckas-Claude/mooncake-tmap-responsibl...
• 154b9dd Merge pull request #1315 from ChrisRackauckas-Claude/cc/unbreak-master-ci-aft...
• Additional commits viewable in compare view

Updates CUDA to 6.0.0

Release notes

Sourced from CUDA's releases.

v6.0.0

CUDA v6.0.0

Diff since v5.11.0

Breaking changes: Ideally none. CUDA.jl has been split into subpackages, which is a major change, so the major version bump is out of caution. Deprecations have been introduced though, e.g., submodules like CUBLAS are now proper packages, cuBLAS.jl.

This release has been identified as a backport. Automated changelogs for backports tend to be wildly incorrect. Therefore, the list of issues and pull requests is hidden.

Commits

• a9a687c Add licenses and READMEs to subpackages.
• a79b516 Profiler: relax parsing of color flags in marker data. (#3075)
• 2e540fd Remove stale NEWS.md
• 0b05451 Put all GPU-only functions in the GPU method table (#2998)
• cbd36b0 Bump docker/setup-buildx-action from 3 to 4 (#3071)
• 74e1705 Bump docker/metadata-action from 5 to 6 (#3070)
• 9877922 Bump docker/build-push-action from 6 to 7 (#3069)
• 19c3638 Bump docker/login-action from 3 to 4 (#3068)
• 87da928 Mark @device_override and friends as public (#3066)
• 441fec5 Make CUDA & friends loadable on systems without NVPTX LLVM backend (#3067)
• Additional commits viewable in compare view

Updates DiffEqDevTools to 3.0.0

Changelog

Sourced from DiffEqDevTools's changelog.

OrdinaryDiffEq.jl v7 Breaking Changes

This release bumps to SciMLBase v3, RecursiveArrayTools v4, and includes breaking changes across DiffEqBase, OrdinaryDiffEqCore, and all solver sublibraries.

Themes of the v7 release

Most of the breaking changes fall into a small set of recurring themes. Keep these in mind while reading the migration table — they explain why an individual change exists and often suggest the right migration direction:

• Time to first solve (TTFS) reduction. Direct deps on Static.jl, StaticArrayInterface.jl, Polyester.jl, and StaticArrays.jl were dropped; using OrdinaryDiffEq now loads only the default solver set; ODEFunction switched to AutoSpecialize. All of this means less code loaded and more precompilation caching on first solve.
• Type stability everywhere. All Bool solver/solve keyword arguments (autodiff, verbose, alias, lazy, …) were replaced by typed objects. Passing a Bool no longer changes dispatch in ways the compiler cannot specialize on, and the reverse is no longer allowed to silently fall back through slow generic paths.
• Generality beyond ForwardDiff. chunk_size, diff_type, standardtag, etc. encoded ForwardDiff-specific or FiniteDiff-specific knobs on every solver. They are replaced by the ADTypes interface (AutoForwardDiff, AutoFiniteDiff, AutoEnzyme, AutoZygote, …) so every solver automatically generalizes to any AD backend.
• Controller is now an object, not a pile of solve kwargs. gamma, beta1, beta2, qmin, qmax, qsteady_min, qsteady_max, qoldinit were moved onto concrete PIController / PIDController / IController / PredictiveController structs, and EEst moved to the controller cache. This is prep work for pluggable controllers and removes a large amount of dead state from the integrator struct.
• Cleanup of old re-exports / deprecations. Functions like has_destats (now has_stats), sol.destats (now sol.stats), DEAlgorithm/DEProblem/DESolution abstract types, tuples()/intervals(), QuadratureProblem, fastpow, concrete_solve, etc. were on a deprecation path for one or more minor releases. v7 removes them.

Recommended upgrade path

The cleanest path is not to jump straight from an old environment onto v7. Most renamed APIs (e.g. DEAlgorithm → AbstractDEAlgorithm, u_modified! → derivative_discontinuity!, has_destats → has_stats, sol.destats → sol.stats, the construct* tableau functions, alias_u0/alias_du0, beta1/beta2, PID kwargs) already exist under their new names in SciMLBase v2 / OrdinaryDiffEq v6 with deprecation warnings. The recommended sequence is:

1. Stay on SciMLBase v2 / OrdinaryDiffEq v6. Update your code to the new names (has_stats, sol.stats, AbstractDEAlgorithm, derivative_discontinuity!, ODEAliasSpecifier, ODEVerbosity, ADTypes-based autodiff, explicit controller = … objects, new tableau names) while the deprecation shims still exist.
2. Verify your tests pass on v6 with no deprecation warnings.
3. Then bump to v7. At this point your code should compile and run against v7 without further changes aside from the genuinely new breakage (RAT v4 array semantics, ensemble prob_func/output_func signature, struct type parameter removals, kwargs that truly no longer exist, default changes like CheckInit and williamson_condition=false).

Doing it in two steps keeps the diff small per step and lets the deprecation warnings on v6 point you at the exact call sites that will break on v7.

Fallback for RAT v4 indexing

If you cannot update sol[i] / length(sol) / eachindex(sol) call sites yet (see the RAT v4 table below), you can opt back into v3 semantics on a per-solution basis by converting the container type to the ragged variant:

using RecursiveArrayToolsRaggedArrays
sol_old = RaggedVectorOfArray(sol)   # indexes like v3: sol_old[i] is the i-th timestep

RecursiveArrayToolsRaggedArrays.jl preserves the previous AbstractVectorOfArray indexing behavior (timestep-first, not element-first). This is the escape hatch for code that assumes sol[i] returns the i-th timestep. It is, however, recommended that you update to the sol.u[i] / sol[:, i] style — the ragged wrapper is a compatibility layer, not the canonical API going forward.

---

RecursiveArrayTools v4

ODESolution is now an AbstractArray

AbstractVectorOfArray (the parent type of ODESolution, RODESolution, DAESolution, etc.) now subtypes AbstractArray. This changes the semantics of several common operations:

Operation	v3 (old)	v4 (new)	Migration
sol[i]	Returns i-th timestep (Vector)	Returns i-th scalar element (column-major)	Use sol.u[i] or sol[:, i]
length(sol)	Number of timesteps	prod(size(sol)) (total elements)	Use length(sol.t) or length(sol.u)
eachindex(sol)	1:nsteps	CartesianIndices(size(sol))	Use eachindex(sol.u)
iterate(sol)	Iterates over timesteps	Iterates over scalar elements	Use for u in sol.u
first(sol) / last(sol)	First/last timestep	First/last scalar element	Use first(sol.u) / last(sol.u)

... (truncated)

Commits

• 8c2148f tests through data arrays pass
• a810720 fix interpolations
• 8da638b move integrators back
• f050aee fix ssprk
• 09335aa fix symplectic and rkn to new syntax
• 5dc79c4 update test problems to new syntax
• 9c37d3c bug catching in the ESDIRK methods
• 928b563 syntax fixes, passing p through
• c57cdf2 caches upgraded to have p
• 1d1ac57 most perform step changes
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore <dependency name> major version will close this group update PR and stop Dependabot creating any more for the specific dependency's major version (unless you unignore this specific dependency's major version or upgrade to it yourself)
• @dependabot ignore <dependency name> minor version will close this group update PR and stop Dependabot creating any more for the specific dependency's minor version (unless you unignore this specific dependency's minor version or upgrade to it yourself)
• @dependabot ignore <dependency name> will close this group update PR and stop Dependabot creating any more for the specific dependency (unless you unignore this specific dependency or upgrade to it yourself)
• @dependabot unignore <dependency name> will remove all of the ignore conditions of the specified dependency
• @dependabot unignore <dependency name> <ignore condition> will remove the ignore condition of the specified dependency and ignore conditions

[dependabot]

Looks like these dependencies are updatable in another way, so this is no longer needed.