Repository navigation
Conversation
nicolas-abril
added a commit
to oxura/bend
that referenced
this pull request
Oct 9, 2026
…ld order compile_book puts each constructor's spine in CID_T: the word of its last field whose type holds its own datatype (255 when none). term_drop swaps the spine into the last slot when it opens the node, so the last child's early free of bendlang#1438 fires on the spine whatever the field order: a list with its tail first, before a word or behind a List drops in constant frames. The inner loop has one pop for the last and the exhausted child, and skips the dead counter and cursor writes there. tests/rfc/drop_spine covers the three shapes, owned and shared. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ld order compile_book puts each constructor's spine in CID_T: the word of its last field whose type holds its own datatype (255 when none). term_drop swaps the spine into the last slot when it opens the node, so the last child's early free of bendlang#1438 fires on the spine whatever the field order: a list with its tail first, before a word or behind a List drops in constant frames. The inner loop has one pop for the last and the exhausted child, and skips the dead counter and cursor writes there. tests/rfc/drop_spine covers the three shapes, owned and shared. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
nicolas-abril
force-pushed
the
fix/1307-stable-wrapped-erasure
branch
from
October 11, 2026 19:44
d3bc283 to
d757f53
Compare
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
Fixes #1307. Repeatedly building and automatically erasing the reported 524,283-element affine list progressively slows down within one process; the structural-disposal control stays steady.
Change
When the iterative native drop reaches a node’s last nontrivial field, capture that child and the parent-frame link before returning the node to its existing size-class free chain. The last child no longer needs that frame. Earlier children, shared-reference accounting and trivial-field traversal remain unchanged. No sorting, new allocation, recursion, runtime flag, ABI, checker, gate or cap change. The drop remains O(n).
Actual verification
Linux x86_64, current main 60fa05d. The issue’s exact 24 × 524283 workload is retained, with only its old Chan.send Bool annotation adapted to today’s Result contract.
Fresh processes, before/after/after/before, then structural/after/before:
Observed automatic-workload ratio: 4.57×. This is not a universal speedup claim. Across three 24-cycle patched runs, build times stay at 4–5 ms after each first allocation and disposal at 20–22 ms; main reaches 89–90 ms build and 130–134 ms disposal. The generic drop still costs more than the explicit traversal.
No new timing assertion: this performance regression is demonstrated by the actual issue workload and interleaved runs rather than a machine-sensitive permanent test. Existing behavior regressions are intentionally unchanged. No mini-cluster, Metal, CUDA, macOS or Windows execution claimed.
AI-assisted: omp / GPT-6.1 Sol, working alone without subagents.
Release-note policy
Removed my unreleased changelog entry following maintainer review. CHANGELOG.md is now unchanged from main; implementation and previously exercised verification are unchanged.