Skip to content

Use public FunctionWrappersWrapper constructor - #1074

Closed
ChrisRackauckas-Claude wants to merge 1 commit into
SciML:masterfrom
ChrisRackauckas-Claude:agent/use-public-fww-constructor
Closed

ChrisRackauckas-Claude wants to merge 1 commit into
SciML:masterfrom
ChrisRackauckas-Claude:agent/use-public-fww-constructor

Conversation

@ChrisRackauckas-Claude

Copy link
Copy Markdown
Member

Summary

  • construct the inferred FunctionWrapper tuple through the public FunctionWrappersWrapper constructor
  • require FunctionWrappersWrappers 1.11, where that constructor is introduced

Depends on SciML/FunctionWrappersWrappers.jl#70 and its 1.11 release.

This PR should be ignored until reviewed by @ChrisRackauckas.

Local verification

  • NonlinearSolveBase QA on Julia 1.12.6 with FunctionWrappersWrappers GPU Compatibility for 1D Arrays  #70 developed locally: 18/18 passed
  • FunctionWrappersWrappers GPU Compatibility for 1D Arrays  #70 Core: 57/57 and 7/7 passed
  • FunctionWrappersWrappers GPU Compatibility for 1D Arrays  #70 QA: 20/20 passed
  • Runic check over the repository: passed
  • Root Core did not complete: it stopped in the Broyden robustness suite at 94 passed, 1 failed, 1 errored, and 19 broken (generalized Rosenbrock residual 4.4 and a Brown unexpected pass). A separate clean-master reproduction also found the homotopy allocation test at 154.4 B/step for sweep and 80.0 B/step for arclength versus the existing <48 budget. Neither unrelated failure is changed or silenced here.

Construct the inferred wrapper tuple through the public API added in FunctionWrappersWrappers 1.11 instead of reaching into its cache-storage internals.

Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com>
@ChrisRackauckas-Claude

Copy link
Copy Markdown
Member Author

Investigation notes:

  • clean master at 247a8bf reproduced the NonlinearSolveBase QA error: 17 passed, 1 errored, with the sole error being qualified access to non-public FunctionWrappersWrappers.SingleCacheStorage
  • the parent 6a7687f passed the same QA command 18/18
  • git bisect identified 0343930 (Make AutoSpecialize infer through functor residuals; zero-alloc homotopy stepping, Make AutoSpecialize infer through functor residuals; zero-alloc homotopy stepping #1046) as the first bad commit
  • FunctionWrappersWrappers GPU Compatibility for 1D Arrays  #70 supplies the intended public constructor, a docstring rendered by the existing autodocs page, tests, and the required 1.10.1 to 1.11.0 minor-version bump
  • this PR only migrates that internal construction to the public constructor and narrows compat accordingly; it does not alter the inference helper or homotopy behavior.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants