You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
strided-einsum2 and strided-opteinsum are not being retired. Per #226, all nine workspace crates remain maintained. Any retirement/Phase 2 removal premise below is obsolete; re-audit and implement the acceptance criteria across the maintained workspace.
Part of #199 (Phase 3, API). All file:line references are pinned to origin/main at 9da9b9f. Decision issue; record the outcome in REPOSITORY_RULES.md, then rename without compatibility shims (pre-1.0 policy).
Because einsum2 remains maintained, its beta-carrying _into entry points must be included in the vocabulary decision. The workspace surface is inconsistent:
add / mul / axpy / fma (ops_view.rs:254, :355, :456, :562) are unsuffixed read-modify-write ops (dest[i] += ..., dest[i] *= ...), and axpy_raw / axpy_conj_raw likewise; all are facade-root exports.
mul (dest *= src) vs mul_into (dest = a * b, map_view.rs:1713) are different operations under a name pair suggesting the same op.
copy_scale / copy_conj (ops_view.rs:857, :872) are pure overwrites missing _into while copy_into / copy_transpose_scale_into carry it.
_view means metadata-only exactly once (diagonal_view) while permute / broadcast / transpose_2d / conj are metadata-only without it, and *_view.rs module names mean "operates on views".
_uninit coverage is asymmetric: ErasedCopyPlan lacks execute_uninit (typed CopyPlan has it), gather-family typed execute_uninit is pub(crate) while the erased ones are public, and map_into / zip_map* / copy_into / reduce* have no uninit siblings.
Decide: adopt the tenferro vocabulary (_add_to for accumulation, _into strictly overwrite, one _view meaning) or document a strided-specific convention.
Rename and close the _uninit coverage gaps accordingly.
Scope correction (2026-08-10)
strided-einsum2 and strided-opteinsum are not being retired. Per #226, all nine workspace crates remain maintained. Any retirement/Phase 2 removal premise below is obsolete; re-audit and implement the acceptance criteria across the maintained workspace.
Part of #199 (Phase 3, API). All file:line references are pinned to
origin/mainat 9da9b9f. Decision issue; record the outcome inREPOSITORY_RULES.md, then rename without compatibility shims (pre-1.0 policy).Because einsum2 remains maintained, its
beta-carrying_intoentry points must be included in the vocabulary decision. The workspace surface is inconsistent:add/mul/axpy/fma(ops_view.rs:254,:355,:456,:562) are unsuffixed read-modify-write ops (dest[i] += ...,dest[i] *= ...), andaxpy_raw/axpy_conj_rawlikewise; all are facade-root exports.mul(dest *= src) vsmul_into(dest = a * b,map_view.rs:1713) are different operations under a name pair suggesting the same op.copy_scale/copy_conj(ops_view.rs:857,:872) are pure overwrites missing_intowhilecopy_into/copy_transpose_scale_intocarry it._viewmeans metadata-only exactly once (diagonal_view) whilepermute/broadcast/transpose_2d/conjare metadata-only without it, and*_view.rsmodule names mean "operates on views"._uninitcoverage is asymmetric:ErasedCopyPlanlacksexecute_uninit(typedCopyPlanhas it), gather-family typedexecute_uninitispub(crate)while the erased ones are public, andmap_into/zip_map*/copy_into/reduce*have no uninit siblings.Decide: adopt the tenferro vocabulary (
_add_tofor accumulation,_intostrictly overwrite, one_viewmeaning) or document a strided-specific convention.Rename and close the
_uninitcoverage gaps accordingly.