Focus value consistency by luketpeterson · Pull Request #163 · Adam-Vandervorst/PathMap

luketpeterson · 2026-10-10T13:30:22Z

This should be ready to go. Details about the changes in #104 issue.

Builds on #162

@imlvts Make sure I didn't screw the lean model. Changes seem like improvements / simplifications to me.

…e correct behavior has been settled for over a year and it was just cruft

Changing PathMap::restrict and WriteZipper::restrict to return None instead of Identity when the destination started empty

…k_path does. #104

…nsistent behavior around zippers positioned at dangling paths #104

…ng out tests, and updating the API to match "blind zipper" patterns. Improving documentation and adding a diagram to illustrate behavior #104

github-actions · 2026-10-10T13:33:14Z

Differential fuzz vs Lean model: success

job log · 2026-10-28 14:18:18 UTC

WARNING: 147 new divergence(s) relative to base

crate: 20000 inputs, seed 7

agree = model and crate match; known = the divergence matches a classified bug in differential.py; new = it matches none. Only the new set is compared input by input below.

side	agree	known	new divergences
head	19154/20000	548	298
base	6439/20000	4940	8621

130 input(s) diverge on head but not on base; 8453 diverge on base but not on head.

Known-class hits that changed

known class	base	head
subtract_into() drops a value under a path present in the source [subtract_drops_value]	2084	0
join_into() drops the source subtrie when the destination map is empty [join_into_empty_dst]	1486	0
AlgebraicStatus::Identity is not returned reliably when nothing changed (finding 8); reproducer in l	1159	233
graft_masked_branches() creates the focus when grafting absent branches (finding 15) [graft_child_ma	88	216
which value survives a collision follows node layout rather than the left-biased spec (join/meet and	99	66
an algebraic op keeps a dangling child the spec drops: an empty child node meets/subtracts to itself	3	12

Newly diverging inputs (head only): 130 input(s), 1 kind(s)

first differs at	inputs	first example
`meet_k_path_into`	130	`random#00119`: line 10

Reproducers: first 1 distinct kind(s), shrunk

random#00119: first differs at meet_k_path_into, 268 -> 37 bytes

line 10
lean: 10 meet_k_path_into ret=0 W=_ o01 e1 v92 c0 n1 f? R=0200 o0200 e1 v164 c0 n1 f00
crate: 10 meet_k_path_into ret=0 W=_ o01 e1 v- c0 n0 f? R=0200 o0200 e1 v164 c0 n1 f00

//! Generated by `pathmap_trace --repro`.  Reproduces a fuzzer input as
//! plain `pathmap` API calls.
//!
//! `wz` writes into map0 and `rz` reads map1; they are separate maps so
//! both can be held at once.  Values are `u64`.

use pathmap::PathMap;
use pathmap::utils::ByteMask;
use pathmap::zipper::*;

fn main() {
    let mut map0 = PathMap::<u64>::new();
    map0.set_val_at(&[], 0);
    map0.set_val_at(&[], 0);
    map0.set_val_at(&[0u8, 0, 0, 0, 0], 0);
    map0.set_val_at(&[0u8, 0, 0], 0);
    map0.set_val_at(&[1u8, 0], 0);
    let mut map1 = PathMap::<u64>::new();

    let mut wz = map0.write_zipper_at_path(&[]);
    let mut rz = map1.read_zipper_at_path(&[]);

    /*   0 */ if 1 != 0 && wz.child_count() != 0 { wz.meet_k_path_into(1, false); }

    drop(wz);

    // What the maps hold at the end -- compare against the trace's MAP0/MAP1.

    println!("MAP0 {:?}", map0.iter().collect::<Vec<_>>());

    println!("MAP1 {:?}", map1.iter().collect::<Vec<_>>());
}

8453 input(s) fixed on head

random#00000
random#00013
random#00015
random#00016
random#00017
random#00019
random#00020
random#00022
random#00027
random#00028
random#00030
random#00032
random#00034
random#00036
random#00038
random#00041
random#00043
random#00044
random#00045
random#00046
random#00047
random#00048
random#00051
random#00057
random#00060
random#00061
random#00064
random#00066
random#00068
random#00069
random#00071
random#00072
random#00076
random#00077
random#00083
random#00086
random#00087
random#00089
random#00091
random#00093
random#00094
random#00096
random#00098
random#00100
random#00103
random#00106
random#00108
random#00110
random#00111
random#00113

act: 5000 inputs, seed 7

agree = model and crate match; known = the divergence matches a classified bug in differential.py; new = it matches none. Only the new set is compared input by input below.

side	agree	known	new divergences
head	3502/5000	1432	66
base	1559/5000	1528	1913

17 input(s) diverge on head but not on base; 1864 diverge on base but not on head.

Known-class hits that changed

known class	base	head
subtract_into() drops a value under a path present in the source [subtract_drops_value]	400	0
ACTZipper::val_count() counts from the zipper root, not the focus [act: val_count_ignores_focus]	1093	1420
which value survives a collision follows node layout rather than the left-biased spec (join/meet and	35	12

Newly diverging inputs (head only): 17 input(s), 1 kind(s)

first differs at	inputs	first example
`meet_k_path_into`	17	`random#00410`: line 25

Reproducers: first 1 distinct kind(s), shrunk

random#00410: first differs at meet_k_path_into, 252 -> 22 bytes (act mode: the harness reads map1 through an ArenaCompactTree; the reproducer uses a PathMap read zipper)

line 25
lean: 25 meet_k_path_into ret=0 W=_ o0000 e1 v126 c0 n1 f? R=_ o_ e1 v- c1 n1 f?
crate: 25 meet_k_path_into ret=0 W=_ o0000 e1 v- c0 n0 f? R=_ o_ e1 v- c1 n1 f?

//! Generated by `pathmap_trace --repro`.  Reproduces a fuzzer input as
//! plain `pathmap` API calls.
//!
//! `wz` writes into map0 and `rz` reads map1; they are separate maps so
//! both can be held at once.  Values are `u64`.

use pathmap::PathMap;
use pathmap::utils::ByteMask;
use pathmap::zipper::*;

fn main() {
    let mut map0 = PathMap::<u64>::new();
    let mut map1 = PathMap::<u64>::new();

    let mut wz = map0.write_zipper_at_path(&[]);
    let mut rz = map1.read_zipper_at_path(&[]);

    /*   0 */ wz.insert_prefix(&[0u8]);
    /*   1 */ wz.get_val_or_set_mut(0);
    /*   2 */ if 1 != 0 && wz.child_count() != 0 { wz.meet_k_path_into(1, false); }

    drop(wz);

    // What the maps hold at the end -- compare against the trace's MAP0/MAP1.

    println!("MAP0 {:?}", map0.iter().collect::<Vec<_>>());

    println!("MAP1 {:?}", map1.iter().collect::<Vec<_>>());
}

1864 input(s) fixed on head

random#00000
random#00008
random#00013
random#00015
random#00016
random#00017
random#00020
random#00027
random#00030
random#00035
random#00038
random#00041
random#00043
random#00044
random#00048
random#00051
random#00052
random#00054
random#00060
random#00064
random#00068
random#00070
random#00072
random#00076
random#00077
random#00086
random#00087
random#00088
random#00092
random#00093
random#00094
random#00096
random#00098
random#00099
random#00100
random#00103
random#00106
random#00107
random#00108
random#00110
random#00111
random#00113
random#00114
random#00121
random#00122
random#00124
random#00127
random#00130
random#00134
random#00135

github-actions · 2026-10-10T13:33:28Z

Bench A/B vs base: done

job log · 2026-10-28 14:26:12 UTC

base 91abaa8 → head 75cc482, 3 round(s), median of each run averaged; negative is faster

bench	cases	geomean	largest gain	largest loss	>5% faster	>5% slower
shakespeare	7	-0.3%	-1.6% `shakespeare/shakespeare_sentences_val_count`	+1.1% `shakespeare/shakespeare_words_val_count`	0	0
cities	5	-0.8%	-3.7% `cities/cities_val_count`	+3.3% `cities/cities_get`	0	0
sparse_keys	108	-1.1%	-12.3% `sparse_subtract_after_join/8000`	+7.5% `sparse_zipper_prev_step_iter/800`	9	4
binary_keys	89	-4.5%	-22.5% `binary_zipper_iter/50`	+14.2% `binary_get/2000`	26	1
superdense_keys	116	+0.8%	-18.6% `superdense_insert/800`	+10.4% `superdense_zipper_step_iter/3200`	4	12
act_paths	46	-0.9%	-9.1% `shape_map_to_act_cata/random_long`	+3.5% `size_map_to_act_cata/50000`	2	0
zipper_head_owned	16	-0.0%	-3.6% `zipper_head_owned/borrowed_head_read_missing_path`	+3.4% `zipper_head_owned/borrowed_head_write_creation_cleanup_unchecked`	0	0
product_zipper	4	-1.7%	-5.0% `product_zipper/introspecting_pathmap_pathmap`	+1.0% `product_zipper/generic_act_act`	1	0

59 case(s) moved more than 5%

bench	case	base	head	change
binary_keys	`binary_zipper_iter/50`	1.44 µs	1.12 µs	-22.5%
superdense_keys	`superdense_insert/800`	17.73 µs	14.44 µs	-18.6%
binary_keys	`binary_descend_until_max_bytes/1000`	73.00 µs	60.94 µs	-16.5%
binary_keys	`binary_descend_until/250`	18.41 µs	15.41 µs	-16.3%
binary_keys	`binary_descend_until/1000`	81.51 µs	68.24 µs	-16.3%
binary_keys	`binary_zipper_prev_iter/50`	50.57 µs	43.29 µs	-14.4%
binary_keys	`binary_get/2000`	136.00 µs	155.33 µs	+14.2%
binary_keys	`binary_zipper_iter/100`	2.56 µs	2.21 µs	-13.6%
binary_keys	`binary_zipper_prev_iter/100`	97.46 µs	84.78 µs	-13.0%
binary_keys	`binary_zipper_prev_iter/200`	189.53 µs	165.33 µs	-12.8%
binary_keys	`binary_k_path_iter/1600`	39.29 µs	34.33 µs	-12.6%
binary_keys	`binary_zipper_prev_iter/1600`	1.38 ms	1.21 ms	-12.4%
binary_keys	`binary_k_path_iter/800`	20.15 µs	17.67 µs	-12.3%
sparse_keys	`sparse_subtract_after_join/8000`	122.67 µs	107.60 µs	-12.3%
binary_keys	`binary_zipper_prev_iter/400`	366.97 µs	322.40 µs	-12.1%
binary_keys	`binary_zipper_prev_iter/800`	707.83 µs	624.33 µs	-11.8%
binary_keys	`binary_zipper_prev_step_iter/50`	91.02 µs	80.48 µs	-11.6%
binary_keys	`binary_descend_until/500`	36.23 µs	32.14 µs	-11.3%
sparse_keys	`sparse_descend_until/500`	14.01 µs	12.46 µs	-11.1%
binary_keys	`binary_zipper_prev_step_iter/100`	172.93 µs	154.77 µs	-10.5%
superdense_keys	`superdense_zipper_step_iter/3200`	24.64 µs	27.21 µs	+10.4%
binary_keys	`binary_zipper_iter/200`	4.90 µs	4.44 µs	-9.4%
superdense_keys	`superdense_zipper_step_iter/1600`	13.24 µs	14.48 µs	+9.3%
act_paths	`shape_map_to_act_cata/random_long`	43.73 ms	39.75 ms	-9.1%
binary_keys	`binary_descend_until_max_bytes/500`	31.21 µs	28.40 µs	-9.0%
superdense_keys	`superdense_zipper_step_iter/400`	4.18 µs	4.55 µs	+9.0%
binary_keys	`binary_zipper_prev_step_iter/200`	329.83 µs	300.53 µs	-8.9%
binary_keys	`binary_zipper_iter/1600`	36.46 µs	33.32 µs	-8.6%
binary_keys	`binary_zipper_iter/400`	9.38 µs	8.59 µs	-8.4%
sparse_keys	`sparse_meet/100`	1.94 µs	1.78 µs	-8.3%
binary_keys	`binary_descend_until_max_bytes/250`	14.16 µs	13.02 µs	-8.1%
act_paths	`shape_map_to_act_cata/dense_narrow`	10.79 ms	9.92 ms	-8.0%
binary_keys	`binary_zipper_prev_step_iter/400`	630.43 µs	579.97 µs	-8.0%
sparse_keys	`sparse_zipper_prev_step_iter/800`	181.07 µs	194.70 µs	+7.5%
superdense_keys	`superdense_zipper_step_iter/100`	754 ns	811 ns	+7.5%
superdense_keys	`superdense_zipper_step_iter/200`	1.46 µs	1.57 µs	+7.5%
superdense_keys	`superdense_val_count_bench_act/3200`	112.23 µs	120.60 µs	+7.5%
sparse_keys	`sparse_meet/50`	941 ns	871 ns	-7.4%
sparse_keys	`sparse_val_count_bench/125`	292 ns	270 ns	-7.4%
binary_keys	`binary_zipper_iter/800`	18.12 µs	16.78 µs	-7.4%
superdense_keys	`superdense_descend_until/1000`	6.36 µs	6.81 µs	+7.1%
superdense_keys	`superdense_val_count_bench/800`	754 ns	704 ns	-6.6%
superdense_keys	`superdense_insert/200`	2.26 µs	2.12 µs	-6.3%
binary_keys	`binary_descend_until/4000`	558.50 µs	523.20 µs	-6.3%
superdense_keys	`superdense_zipper_prev_step_iter/400`	4.12 µs	4.38 µs	+6.2%
superdense_keys	`superdense_zipper_prev_step_iter/3200`	20.59 µs	21.85 µs	+6.2%
superdense_keys	`superdense_zipper_prev_step_iter/100`	607 ns	644 ns	+6.1%
superdense_keys	`superdense_val_count_bench/400`	437 ns	411 ns	-6.0%
sparse_keys	`sparse_zipper_cursor/50`	624 ns	587 ns	-5.9%
sparse_keys	`join_sparse/50`	571 ns	604 ns	+5.8%
superdense_keys	`superdense_val_count_bench_act/800`	25.40 µs	26.86 µs	+5.8%
sparse_keys	`sparse_zipper_prev_step_iter/100`	23.57 µs	22.22 µs	-5.7%
binary_keys	`binary_k_path_iter/50`	1.32 µs	1.25 µs	-5.6%
sparse_keys	`sparse_zipper_prev_cursor/100`	14.06 µs	13.28 µs	-5.5%
superdense_keys	`superdense_descend_until/16000`	115.07 µs	121.43 µs	+5.5%
sparse_keys	`sparse_get/8000`	183.33 µs	193.23 µs	+5.4%
sparse_keys	`sparse_zipper_prev_cursor/50`	6.76 µs	6.41 µs	-5.2%
sparse_keys	`binary_drop_head/800`	173.53 µs	182.40 µs	+5.1%
product_zipper	`product_zipper/introspecting_pathmap_pathmap`	473.00 ms	449.20 ms	-5.0%

Full tables per bench are in the job log and the bench-out artifact.

…s the purview of #117, so it doesn't appear as a new divergence

luketpeterson · 2026-10-10T23:27:43Z

Full perf data attached. No detectable change in any recursive workloads (real-world cases). Degenerate cases with like joining two single-path tries together show the cost of checking the focus value (about 10%), but I don't think it's worth sacrificing to optimize for these degenerate cases, and the cost will actually be reversed (become a gain) when the new node contract is implemented.
focus_values_benchmark.zip

Adam-Vandervorst · 2026-10-11T00:55:13Z

Can we hold off on merging this? This touches a ton of stuff there's no benchmark for here. I have two sets of benchmarks, but need some time to integrate them.

…header CLI arg

…ency

imlvts

This is a bit more involved than I expected, but this certainly improves consistency.
If I'm not mistaken this also fixes one of the issues I encountered when doing algebraic equivalence, specifically because of the root/focus value.

Looks good.

luketpeterson · 2026-10-13T05:06:25Z

This is a bit more involved than I expected, but this certainly improves consistency. If I'm not mistaken this also fixes one of the issues I encountered when doing algebraic equivalence, specifically because of the root/focus value.

Looks good.

I snuck in an API consistency fix and better documentation (both for remove_prefix). That's some of the extra bulk.

I'm holding until we get the all-clear on real-world perf.

…rune when a lredy empty and a zero-arg was passed, and an infinite loop when already empty

# Conflicts: # Cargo.toml # src/write_zipper.rs

imlvts · 2026-10-28T14:24:05Z

🎉

luketpeterson added 8 commits October 9, 2026 05:43

Removing the graft_root_vals feature, because the question about th…

9ab2607

…e correct behavior has been settled for over a year and it was just cruft

Updating join_into and join_into_take, to include focus values for #104

7953351

Updating WriteZipper::restrict to match PathMap::restrict

010f515

Changing PathMap::restrict and WriteZipper::restrict to return None instead of Identity when the destination started empty

Making drop_head aka join_k_path treat values the same way that meet_…

8df19a7

…k_path does. #104

Updating meet_2 to include focus values. #104

788558f

Fixing prune bug exposed by behavior changes to alg-ops

5d9543d

Making insert_prefix shift the focus value down, and eliminating inco…

9f98238

…nsistent behavior around zippers positioned at dangling paths #104

Updating remove_prefix for consistent handling of focus value, roundi…

b51dde5

…ng out tests, and updating the API to match "blind zipper" patterns. Improving documentation and adding a diagram to illustrate behavior #104

luketpeterson requested a review from imlvts October 10, 2026 13:30

Adding exception to correctly classify meet + dangling issue, which i…

a75296e

…s the purview of #117, so it doesn't appear as a new divergence

luketpeterson added 2 commits October 12, 2026 20:05

Updating lean harness to write newest header and get rid of explicit …

fc9bfc8

…header CLI arg

Merge remote-tracking branch 'origin/master' into focus_value_consist…

88ef78e

…ency

imlvts approved these changes Oct 13, 2026

View reviewed changes

luketpeterson added 2 commits October 28, 2026 08:12

Fixing meet_k_path_into edge cases, including replacing root value, p…

482c248

…rune when a lredy empty and a zero-arg was passed, and an infinite loop when already empty

Merge branch 'master' into focus_value_consistency

864a2b8

# Conflicts: # Cargo.toml # src/write_zipper.rs

Navigation Menu

Repository navigation

Focus value consistency - #163

Focus value consistency#163
luketpeterson wants to merge 13 commits into
masterfrom
focus_value_consistency

luketpeterson commented Oct 10, 2026

Uh oh!

github-actions Bot commented Oct 10, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Oct 10, 2026 •

edited

Loading

Uh oh!

luketpeterson commented Oct 10, 2026

Uh oh!

Adam-Vandervorst commented Oct 11, 2026

Uh oh!

imlvts left a comment

Uh oh!

luketpeterson commented Oct 13, 2026 •

edited

Loading

Uh oh!

imlvts commented Oct 28, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

luketpeterson commented Oct 10, 2026

Uh oh!

github-actions Bot commented Oct 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Differential fuzz vs Lean model: success

crate: 20000 inputs, seed 7

Known-class hits that changed

Newly diverging inputs (head only): 130 input(s), 1 kind(s)

Reproducers: first 1 distinct kind(s), shrunk

act: 5000 inputs, seed 7

Known-class hits that changed

Newly diverging inputs (head only): 17 input(s), 1 kind(s)

Reproducers: first 1 distinct kind(s), shrunk

Uh oh!

github-actions Bot commented Oct 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Bench A/B vs base: done

Uh oh!

luketpeterson commented Oct 10, 2026

Uh oh!

Adam-Vandervorst commented Oct 11, 2026

Uh oh!

imlvts left a comment

Choose a reason for hiding this comment

Uh oh!

luketpeterson commented Oct 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

imlvts commented Oct 28, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

github-actions Bot commented Oct 10, 2026 •

edited

Loading

github-actions Bot commented Oct 10, 2026 •

edited

Loading

luketpeterson commented Oct 13, 2026 •

edited

Loading