Add tuple-output functions by oflatt-claude · Pull Request #938 · egraphs-good/egglog

oflatt-claude · 2026-06-26T18:15:23Z

Adds tuple-output functions as a first-class alternative to the implicit pair-unboxing in #933.

A function may declare more than one output sort:

(function interval (Math) (i64 i64)
  :merge (values (max old0 new0) (min old1 new1)))   ; col 0 = lo, col 1 = hi

Outputs are stored as separate value columns (no boxing); the functional dependency maps the keys to a tuple of values. They are:

destructured in queries: (= (values lo hi) (interval x))
written in actions: (set (interval x) (values 0 100))
merged with a (values e0 e1 …) clause where ei merges column i, referencing the bound variables old0, new0, old1, new1, …

This collapses analyses that previously needed one function per output (e.g. the two-function interval analysis in tests/interval.egg) into a single function.

Implementation

egglog-bridge: SchemaMath now tracks key vs. value columns independently; new per-column MergeFn::Columns with OldCol/NewCol leaves. core-relations already supported multi-value tables.
front-end: Schema/FuncType/ResolvedSchema carry extra output sorts; new ResolvedCall::Values and values lowering in facts and set actions; per-column merge type-checking.
Tuple outputs are restricted to plain functions (not constructors/relations/view tables) and are not supported by the term/proof encoding (reported cleanly via Error::UnsupportedProofCommand).

Tests

tests/tuple_outputs.rs covers the interval analysis, column-independent merge, rule destructuring, eq-sort outputs with rebuild, and the error cases. Full suite green (files snapshot suite: 744 passed), fmt/clippy -D warnings clean.

🤖 Generated with Claude Code

Functions may now declare more than one output sort, e.g. `(function interval (Math) (i64 i64) :merge (values (max old0 new0) (min old1 new1)))`. Outputs are stored as separate value columns (no boxing); the functional dependency maps the keys to a tuple of values. They are destructured in queries with `(= (values lo hi) (f x))`, written with `(set (f x) (values lo hi))`, and merged with a `(values ...)` clause whose i-th element merges column i using the bound variables `old0`, `new0`, `old1`, `new1`, .... - egglog-bridge: generalize `SchemaMath` to independent key/value column counts and add a per-column `MergeFn::Columns` (with `OldCol`/`NewCol` leaves). The underlying core-relations table already supports multiple value columns. - front-end: `Schema`/`FuncType`/`ResolvedSchema` carry extra output sorts; new `ResolvedCall::Values` plus `values` lowering in facts and `set` actions. - tuple outputs are restricted to plain functions (not constructors, relations, or view tables) and are unsupported by the term/proof encoding. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

codecov-commenter · 2026-06-26T18:26:50Z

Codecov Report

❌ Patch coverage is 90.21739% with 54 lines in your changes missing coverage. Please review.
✅ Project coverage is 86.72%. Comparing base (e22580d) to head (d7a1ba2).
⚠️ Report is 3 commits behind head on main.

Files with missing lines	Patch %	Lines
src/lib.rs	75.47%	13 Missing ⚠️
src/core.rs	81.81%	12 Missing ⚠️
src/ast/mod.rs	91.89%	9 Missing ⚠️
egglog-bridge/src/lib.rs	92.77%	6 Missing ⚠️
src/typechecking.rs	95.53%	5 Missing ⚠️
src/ast/parse.rs	94.91%	3 Missing ⚠️
src/proofs/proof_checker.rs	0.00%	2 Missing ⚠️
src/exec_state.rs	0.00%	1 Missing ⚠️
src/extract.rs	90.90%	1 Missing ⚠️
src/proofs/proof_encoding_helpers.rs	75.00%	1 Missing ⚠️
... and 1 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #938      +/-   ##
==========================================
+ Coverage   86.65%   86.72%   +0.07%     
==========================================
  Files          90       90              
  Lines       27057    27483     +426     
==========================================
+ Hits        23445    23836     +391     
- Misses       3612     3647      +35

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

codspeed-hq · 2026-06-26T18:55:56Z

Merging this PR will improve performance by 5.18%

⚠️

Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
✅ 35 untouched benchmarks
⏩ 226 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
⚡	Simulation	`tests[repro-665-set-union]`	629.1 ms	598.1 ms	+5.18%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

_{Comparing oflatt-claude:tuple-outputs (d7a1ba2) with main (42a4fb6)}

226 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

`values` is the tuple constructor for tuple-output functions and appears in expression-head position, so it must not be usable as a user identifier. While here, also reserve the other built-in keywords — most command, action, and schedule heads (`function`, `set`, `union`, `rule`, `run`, ...) — so a function named e.g. `set` can no longer be confused with the `set` action. Names starting with `:` are reserved too, since the parser uses that prefix to mark option keywords (`:merge`, `:cost`, ...). `input` and `output` are common words, so they are only *partially* reserved: usable as variables, but not as definition (table/sort/etc.) names nor as the head of a call expression (so `(input ...)` always means the command). egglog previously only reserved an internal symbol *prefix*; this adds an exact-name keyword list (and the `:` prefix), checked when parsing declared names and variable references, and skipped while re-parsing egglog's own generated programs. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…aphs-good#938 port) Port PR egraphs-good#938's multi-value merge machinery onto the shared backend interface, with no behavior change. Adds MergeFn::{OldCol, NewCol, Columns} to egglog-backend-trait; the bridge SchemaMath gains a key/value column split (n_keys/n_vals derived from the merge) + a per-column merge callback with a single-output fast path. Unifies the merge-target-UF mechanism: MergeFn::UnionIntoUf is now the single source of truth on all four backends (read at add_table; register_native_merge_view demoted to a consistency assertion), replacing flowlog/feldera's schedule-time native_merge_uf population. Dataflow/SQL backends collapse Columns([single]) to single-output; multi-value Columns is a documented unsupported error there pending tuple-output table support. New variants are wired but not yet emitted by the encoder. Bit-exact across all backends; single-output hot path perf-flat (ratio 0.994/0.997). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…oof-half machinery) Add the merge machinery for the bridge proof-half native :merge (not yet emitted by the encoder). MergeFn::UnionIntoUfWithProof composes a congruence edge proof (Trans(larger_pf, Sym(smaller_pf)), matching @congruence_rule's orientation bit-exactly) and stages the 4-column [lhs, rhs, proof, ts] insert into the proof-mode @UF_Sf; EclassMinProof returns the term proof of the surviving (min) eclass. Bridge implements them; duckdb/feldera/flowlog get inert arms (proof-mode native-merge is bridge-only). Additive/unreachable until the encoder emits the tuple-output proof view (next: port PR egraphs-good#938 frontend multi-output). No behavior change. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…ut merge) Port the front-end half of egglog PR egraphs-good#938: a function may declare multiple output sorts, stored as separate value columns, with per-column :merge (values e0 e1 ...) referencing old0/new0/old1/new1. Adds Schema.extra_outputs, FuncType/ResolvedSchema multi-output, ResolvedCall::Values, (values ...) parse/lowering in facts and set actions, per-column merge typechecking, and multi-output extraction. The (values ...) merge lowers to the existing MergeFn::Columns. Tuple outputs are restricted to plain functions (rejected for constructors, view tables, and the term/proof encoding -- the encoder-minted proof view exception comes next). Ports tests/tuple_outputs.rs (11/11). Single-output path unchanged; bit-exact sweep across normal/term/native-merge/proofs shows zero regression. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…arse regression) The PR egraphs-good#938 frontend port (0166a2d) added a RESERVED_KEYWORDS list reserving every command/action/schedule head (rewrite, birewrite, rule, set, ...) as a name, plus a ':'-prefix check on the expression-var path. This broke 56/106 Herbie sample100 files: Herbie names a ruleset 'rewrite' and the experimental run-with ... :until scheduler needs ':until' as a var. Command heads are recognized by position, so they never needed name-reservation; only 'values' (the tuple-output keyword) genuinely needs it. Reduce RESERVED_KEYWORDS to ["values"]; move the ':'-prefix rejection to definition-name + let-binder sites only. Restores 106/106 sample100 parse (was 50/106); tuple_outputs 12/12; no other regression. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Ports egraphs-good/egglog#938 ("Add tuple-output functions") into the vendored egglog/ subtree. A function may declare more than one output sort, stored as separate value columns (no boxing); outputs are destructured in queries, written in actions, and merged per column with a (values ...) clause. Restricted to plain functions and unsupported by the term/proof encoding. The upstream diff applied cleanly except for one context conflict in src/proofs/proof_encoding_helpers.rs (this fork carries an extra NaiveEqSortPrimitiveFact reason); the TupleOutputFunction variant and guard were reapplied by hand. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

oflatt-claude requested a review from a team as a code owner June 26, 2026 18:15

oflatt-claude requested review from ezrosent and removed request for a team June 26, 2026 18:15

oflatt marked this pull request as draft June 26, 2026 18:19

oflatt-claude force-pushed the tuple-outputs branch 2 times, most recently from 0ad1048 to 3197ad2 Compare June 26, 2026 21:49

oflatt-claude force-pushed the tuple-outputs branch from 3197ad2 to d7a1ba2 Compare June 26, 2026 22:21

oflatt-claude mentioned this pull request Jul 2, 2026

Port tuple-output functions from egglog#938 saulshanabrook/egglog-encoding#3

Closed

saulshanabrook mentioned this pull request Jul 13, 2026

Reduce proof overhead with native tuple UFs and rule-name hoisting saulshanabrook/egglog-encoding#13

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add tuple-output functions#938

Add tuple-output functions#938
oflatt-claude wants to merge 2 commits into
egraphs-good:mainfrom
oflatt-claude:tuple-outputs

oflatt-claude commented Jun 26, 2026

Uh oh!

codecov-commenter commented Jun 26, 2026 •

edited

Loading

Uh oh!

codspeed-hq Bot commented Jun 26, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

oflatt-claude commented Jun 26, 2026

Implementation

Tests

Uh oh!

codecov-commenter commented Jun 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

codspeed-hq Bot commented Jun 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will improve performance by 5.18%

Performance Changes

Footnotes

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov-commenter commented Jun 26, 2026 •

edited

Loading

codspeed-hq Bot commented Jun 26, 2026 •

edited

Loading