feat(program): a sum through a relation is a sum over the axes its join opens by FBumann · Pull Request #662 · energy-models/mathspec

FBumann · 2026-09-23T21:11:04Z

Prompt: Try to do a new branch from #641 and rebase it onto 661. Any issues? […] Dont change anything on 641 […] PR […] Conflicting […] Now, let's discuss a better api for sum, at and potential future join or groupby operations. What's the best syntax imaginable in the yaml? It should be intuitive, safe and close to the internals […] Fold this into 662

Note

The following content was generated by AI.

#641's change, on top of #661, and the syntax decision it implies. The call shape sum(x, by=R, over=a, into=b) and at(...) stays. The docs now say by= names the relation, and a separate join operator is refused.

What this changes

feat(program): a sum through a relation is a sum over the axes its join opens #641's change (1e30b80). Join and JoinColumns replace Direction, GroupSum and Pullback. A join opens one axis named relation.column for each column it drops. sum(by=) becomes Sum(Join(x, cols), over=axes), and at() becomes the bare Join. feat(program): a sum through a relation is a sum over the axes its join opens #641's body has the design, the guard table and the rationale.
The relation-call syntax, in the docs (e88efcf):
- relations.md says by= names the relation and not the columns grouped by. It says over= names what leaves and into= names what the result lands on, as over= does in sum(p, over=d).
- expressions.md drops "the consumed dimension goes and the produced one arrives" from the predicate at() paragraph. That wording is from before the join design.
- limits.md records a join operator apart from sum and at as refused, and names what to write instead.

This PR stacks on #661, so its diff is the above on the #661 stack and nothing else.

Why

The user asked for #641 on top of #661, and then for the syntax discussion to be folded in.

The call shape is kept because it already matches the nodes. Every relation call lowers to Reduce(Join(x, JoinColumns(joined, grouped)), over=axes), and each keyword maps to one part of that node:

by= is the relation.
over= is joined and not grouped: the axes the reduction closes.
into= is grouped and not joined.
The operator name is the reduction, and at is no reduction.

The operator name is a claim the loader checks. sum claims several rows per group and at claims one. The key decides which is true, and a mismatch is refused at load. A named into= means that a new value column changes no call.

A separate join is refused because the axis must close in the call that opens it. An example is sum(join(p, R), over=R.a). This form would let an open relation.column axis into an expression's frame, where no declaration can carry it. It would also need a new "every axis closes" rule, and it would make the author name the relation twice. In the one-call form, no file can leave an axis open.

by= keeps its name. A pandas reader can take by= for the group-by keys, which are into= here. The fix is to say so in the prose. sum p by gen_bus reads well, so the keyword is not renamed.

Method, conflicts, gates, what was not done

Why one commit and not a rebase

#641 still carries #638 as its 13 original commits. main holds #638 as the single squashed commit b3cee88. A git rebase would apply #638 a second time. So I took #641's net diff against its merge base with main (1e010ca), which is 42 files, +742/−466, the same as the #641 PR. I applied that diff to #661's head f0dc3e3 with git apply --3way (1e30b80).

#661 then gained b535973, which removes UnexpandedCurveError. 5c0814c merges that head in. The branch is not rebased and not force-pushed.

Conflicts and how they were resolved

file	resolution
`resolution.py`	The stack moved the expression walk and the where walk out (`a2db22c`). #641's hunks go to `_expression_resolver.py` (the `sum`/`at` node build, `_relation_ref`, `_direction` → `_join` with its refusal wording) and to `_where_resolver.py` (`join_dims`, the predicate `at()` check). `resolution.py` keeps the stack's side.
`lowering.py`	The stack deleted `inline()`, because a program carries `Named`. #641's `Join` case in it goes with it.
`advice.py`	#661's current imports (`LanguageError`, `to_spec`, after `b535973`), with `Join` in place of `GroupSum` and `Pullback`.
`boundedness.py`	`Sum \| Join \| Translate \| WindowSum \| Cases \| Named`: #641's `Join` and the stack's `Named`.
`program.py`	`__all__` holds `Join`, `JoinColumns` and the stack's `Link`.
`typesetting/walk.py`	The stack's `self.program.relations[...].values`, with #641's `join`.
`tests/test_lowering.py`, `tests/test_separability.py`	#641's names and messages, with the stack's `to_spec(...).program`.

One line merged without a conflict but needed a change: #641's new test_a_sum_over_a_lookup_is_a_sum_and_a_lookup_not_a_grouping called ms.to_program, which the stack removed. It now calls to_spec(...).program.

Generated files

python -m tools.schema and python -m tests.typesetting.golden were run again after each step. The output is the same as the merged text.

Docs sentences

After (before): relations.md n 45, avg 16.3, median 15, over 25 words 8 (43, 16.2, 14, 8). expressions.md n 92, avg 15.3, median 14, over 25 words 11 (91, 15.4, 14, 11). limits.md is unchanged at n 55, avg 18.3, median 19, over 25 words 14, because the new text is a table row.

Gates

Pixi is not reachable here (the proxy refuses pixi.sh), so the gates ran from a uv venv on Python 3.12, with the pinned ruff and pyrefly. The results below are for e88efcf unless a row says otherwise.

gate	result
`pytest -q -n auto`	1624 passed, 1 skipped
`ruff check`, `ruff format --check`	clean on `5c0814c`; `e88efcf` changes only `docs/`
`pyrefly check`	0 errors on `5c0814c`
`mkdocs build --strict`	clean, with the `docs.python.org` inventory left out for the run (the proxy refuses it)
`typos`	clean on `docs/`
`reuse lint`	clean on `1e30b80`; not run again
`prettier --check`	clean on the three changed pages; warns on `mkdocs.yml` and the schema, as #661's head does
`compile-tex`, `zizmor`, the full `lefthook` lint	not run

Not done

feat(program): a sum through a relation is a sum over the axes its join opens #641 is untouched. It is still open against main, and it still has its full history.
feat(program): a sum through a relation is a sum over the axes its join opens #641's history is not kept. This branch has one commit for feat(program): a sum through a relation is a sum over the axes its join opens #641's change.
This PR depends on the whole feat(language): nothing writes a formulation out unasked #661 stack, not only on feat(language): nothing writes a formulation out unasked #661.
No new reduction. A mean(x, by=R, over=a, into=b) would take the same call shape and lower to a Mean over the join's axes. It also needs a language decision: whether an empty group is absent. It is not in this PR.
No keyword is renamed. The refusal messages and JoinColumns.grouped keep "groups by", which is the internal term.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PHNF149r8B4tGmGUKmAdUd

… and every curve is written out before a model becomes one Resolved is gone. Lowering builds the Program as a model loads and holds it on the Spec; to_program(spec) returns that object, and returns the expansion's for a model with a curve, so a program holds the rows a curve states and never the curve. Named joins program.Expression, so a use of an expressions: entry stands where it is read and the typesetter prints from the program beside the file. ExpressionDeclaration carries its dims, and piecewise.curve reads a block's typed links and frame off the model for the expansion and the walk. PiecewiseDeclaration, Program.piecewise, Spec.resolved, lowering.inline and the refusal of a model still carrying a curve are removed. The typesetter's output is unchanged: the golden files and the generated pages regenerate byte for byte. Docs sentences, after (before): reading.md n 75 avg 16.1 median 15 over25 10 (72, 16.4, 14, 13); piecewise.md n 70 avg 17.6 median 15 over25 13 (69, 17.8, 15, 14). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…s included, and the typesetter reads it alone to_program lowers the model as it arrived: a piecewise: block still in it is a curve on the program, typed, with its links, signs, method, gate, mask and frame, and the program of spec.expand('piecewise') carries the rows instead. Every declaration carries its description, and the program the file's. The typesetter takes a Program and reads nothing else; the walk no longer re-resolves a curve's links at print time, and piecewise.curve is gone. relations_of moves from Spec to Program. advice() writes curves out itself and refuses a program still carrying one. The typeset output is unchanged: the golden files and the generated pages regenerate byte for byte. Docs sentences, after (before): reading.md n 75 avg 16.8 median 15 over25 12 (75, 16.1, 15, 10); piecewise.md n 70 avg 17.6 median 15 over25 14 (70, 17.6, 15, 13). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

… lowering, before any expression is read The nine after-validators on Spec that read one declaration against the others — name collisions, frames over declared dimensions, relation targets, bound names, set shapes and bounds, curve references and the names an expansion would collide with — are functions in validation.py, collected by reference_errors and run first by lowering.lower. Spec keeps the shape rules pydantic decides per block, and no longer imports sos or the operator table; side_columns is the public name of the relation helper the rules share. Every message is the same string, and every refusal the same class. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

… is advised as its rows are The guard landed in the previous commit without the test that fails without it. With the guard deleted, the new test fails: a program with a block is advised on the file's own rows as if the curve stated none. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…t-faraday-coc7mj-3

… the program owns the vocabulary the file declares in Loading no longer expands every curve to validate the expansion: the rows a curve states are held to the language when expand() writes them out, since an expansion is a model like any other. A model with a curve is validated once at load rather than twice. The dtype, domain, absence, sense, set-order and method vocabulary moves from model.py into program.py, which model.py imports. program.py no longer imports model.py, so the type-only import of Program on Spec and its noqa go, and sos.py imports Spec plainly instead of inside a function. Load cost, before (after): examples/piecewise.yaml validated 2 (1) times and parsed 9 (4) expressions; examples/sos.yaml the same. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…he one door to both states is to_spec to_program is gone. Spec.program is the model typed, section for section, built once as the model loads; spec.expand(...).program is its rows. A consumer building rows reads the sections it takes and refuses a curve or a set it finds, the way it already had to for a set; advice() does so for a Program handed to it, and writes curves out itself for a file or a Spec. typeset and typeset_declaration take a Program as before. Docs sentences, after (before): reading.md n 75 avg 17.1 median 15 over25 14 (76, 16.8, 15, 12); piecewise.md n 70 avg 17.7 median 15 over25 13 (70, 17.6, 15, 14); what-counts-as-public-api.md n 18 avg 16.0 median 14 over25 3 (unchanged count); limits.md n 55 avg 18.3 median 19 over25 14 (unchanged count). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…the model loads The private attribute lowering filled and the property asserting over it are one cached_property computing lower(self); the after-validator forces it, so a Spec in hand has still passed the whole language. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…ules beside the namespace _Resolver, one class over both grammars, is ExpressionResolver in _expression_resolver.py and WhereResolver in _where_resolver.py, the where walk building a side that is an expression through the expression walk. resolution.py keeps the Namespace and the doors lowering calls. The three methods the where walk reads from the expression walk are public on it; names_in and the literal-number helper move to the parser module beside the other helpers over parsed nodes. Every message is the same string. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…he program carries, and nothing else The Curve alias, a tacit protocol between the pydantic block and the program declaration, is gone: assumptions_of, Emitted.of and the curvature rule take a PiecewiseDeclaration. The expansion keeps the block for the link text its rows repeat and takes the declaration for the frame and the names it writes. The two emitted-name collision rules read the program rather than the file, so they run once the declarations exist; every message is the same string. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…s use off the program, and the walk keeps no record The glossary and the three kinds of note move from walk.py to legend.py. What they explain is read off the program before anything prints, by notice(), so the walk no longer fills a Noticed record as it prints and a subscript no longer mutates the walk through its context. Walk.line refuses a name declared as none of the five kinds or as two, so typeset_declaration is one call. Symbols is a frozen record built by symbols_for. The typeset output is unchanged: the golden files match byte for byte. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…day-coc7mj

…faraday-coc7mj-2

…t-faraday-coc7mj-3

…t-faraday-coc7mj-4

…t-faraday-coc7mj-6

…t-faraday-coc7mj-7

…t-faraday-coc7mj-8

…t-faraday-coc7mj-9

…, the errors, the advice and the typesetter Thirteen names leave math_spec's top level: the message builders call_shape_error, edge_error, unknown_operator_message and schema_error, the tables BUILTIN_NAMES and EDGE_WRAP, the vocabulary sets ADVICE_KINDS, DIMENSION_DTYPES, PARAMETER_DTYPES, VARIABLE_DOMAINS, VARIABLE_ABSENCE and CURVATURES, and SosBlock. The message builders and tables stay in their modules for the resolver; the vocabulary sets are deleted, since the Literals they were the set form of are what a consumer pins against, and nothing in the package read them. did_you_mean stays: it is the one wording a consumer's own refusals share with the language's. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…vocabulary with no Literal form A consumer pins its operator table against BUILTIN_NAMES, and unlike the dtype vocabularies it has no Literal on math_spec.program to pin against instead; the module it lives in is package-private. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…d advice writes nothing out on the caller's behalf UnexpandedCurveError carries the one sentence every consumer building rows says of a program still carrying a piecewise: block, naming the blocks and the expansion to pass. The language does not raise it at load, since a model with a curve is printed and edited as written; advice() raises it for a file, a Spec or a Program alike, and no longer expands a file or a Spec itself. The check verb and the check how-to expand first, as a front end may. Docs sentences, after (before): reading.md n 76 avg 17.1 median 15 over25 14 (75, 17.1, 15, 14); check.md n 20 avg 12.5 median 14 over25 1 (19 sentences before, one added). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…ck verb takes --expand like every other The check verb read a file with its curves written out, the one verb that read a file differently from the rest. Every verb now reads the file as written and takes --expand; check refuses a curve model in the one wording until asked. The premise is on the public-API page, beside the other things every function keeps, and the reading page and the check how-to say it where a consumer meets it. Docs sentences, after (before): check.md n 22 avg 13.0 median 14 over25 2 (20, 12.5, 14, 1); what-counts-as-public-api.md n 22 avg 16.4 median 14 over25 4 (18, 16.0, 14, 3); reading.md n 77 avg 17.1 median 15 over25 14 (76, 17.1, 15, 14). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

UnexpandedCurveError is gone. advice words the refusal of a curve left as written itself, as a LanguageError, and a consumer building rows words its own; the reading page shows the idiom with the consumer's own error. A sentence is not a thing one tool should export for another to reuse, where a computation such as did_you_mean is. Docs sentences, after (before): reading.md n 76 avg 17.4 median 15 over25 15 (77, 17.2, 15, 14); piecewise.md n 70 avg 17.7 median 15 over25 13 (70, 17.7, 15, 14). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

…in opens The change of #641, carried onto #661 as one commit. #641 carries #638 as its separate commits, and main holds #638 squashed, so its net diff against its merge base with main (1e010ca) is applied here, not its history. The resolver edits move into _expression_resolver and _where_resolver, where #661's stack split resolution.py. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PHNF149r8B4tGmGUKmAdUd

read-the-docs-community · 2026-09-23T21:13:41Z

Documentation build overview

📚 math-spec | 🛠️ Build #34754236 | 📁 Comparing e92e826 against latest (b56e28e)

🔍 Preview build

57 files changed · + 5 added · ± 30 modified · - 22 deleted

+ Added

± Modified

- Deleted

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PHNF149r8B4tGmGUKmAdUd

…, and a join is never written apart from its reduction The relations page says by= is not the grouping, so a reader does not read it as a group-by key. The predicate at() paragraph drops the consumed and produced wording the join design retired. The limits page records a join operator apart from sum and at as refused, with the reason: the axis a join opens must close in the call that opens it. Docs sentences, after (before): relations.md n 45 avg 16.3 median 15 over25 8 (43, 16.2, 14, 8); expressions.md n 92 avg 15.3 median 14 over25 11 (91, 15.4, 14, 11); limits.md n 55 avg 18.3 median 19 over25 14 (unchanged; the new text is a table row). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PHNF149r8B4tGmGUKmAdUd

…s expand "states rows rather than being one" was hard to parse on first read, and the sentence named the fix with a different word from the method the user calls. The refusal now says the block is still a curve, that advice reads the rows a curve is expanded into, and to expand first. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

The unboundedness pass read a set as it is, every variable it restricts being named by a row, and refused a curve. A curve states its rows the same way: each link names the variables a link row would, so the pass reads the links and the answer is the expansion's with nothing expanded. The refusal goes, and check loses --expand, a flag that would change no answer. Guard: with the line reading the links deleted, three tests fail: the advice test over the file, the Spec and the Program; the boundedness case carried-by-a-curve; and the existing test that a curve holds its variables, which now runs on the block. Docs sentences, after (before): check.md n 20 avg 12.8 median 14 over25 1 (21, 13.3, 14, 2); what-counts-as-public-api.md n 20 avg 18.7 median 15 over25 4 (20, 18.1, 15, 4). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TrvhjFQCQJ6ATBQkfhcoMi

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PHNF149r8B4tGmGUKmAdUd

…rg-m8xwru main landed the stack this branch was built on as one squash commit per PR. The squash of #661 has the same tree as 279997a, the stack head this branch contains, so this merge records it and changes no file. The merge of main that follows then brings only what landed after it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GpnFcQDuvt2bExGngb6MYk

After the squash of the stack was recorded, main brought #666, #668 and the release. One conflict: the golden census imports this branch's Join in place of GroupSum and Pullback, and main's legend module. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GpnFcQDuvt2bExGngb6MYk

…662 No textual conflict. The legend test main added in #666 wrote a grouped shift as by=season_of, within=season; this branch spells it within=season_of[season], so the test does too. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GpnFcQDuvt2bExGngb6MYk

#633 retired fan_in, FanIn, quotients and divisor_parameters. The branch's Join-aware fan_in and its tests go with them, and the divisor test under a join asserts through parameters_of. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012JFQDAjuZ3sCndQ4cD77Gc

…u-2, through the updated #662 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012JFQDAjuZ3sCndQ4cD77Gc

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Q8Y8AoPsJCyMsURHvgzm7h

Ports #662 onto main's rename to mathspec, its autoref docstrings and the model-to-spec wording, and adds its changelog line. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YUK4sePJqJfxJJZLhdvZnS

read-the-docs-community · 2026-09-29T10:07:15Z

Documentation build overview

📚 mathspec | 🛠️ Build #34852000 | 📁 Comparing 53406dc against latest (ae4a44c)

🔍 Preview build

15 files changed · + 1 added · ± 14 modified

+ Added

about/relations-as-linear-maps/index.html

± Modified

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019TMqXLvfDGYm8hRND63cs8

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MKjqJocyxdRYvx7xin6NCM

…wru-3, with #662 and #664 as one changelog line Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MKjqJocyxdRYvx7xin6NCM

FBumann · 2026-09-30T12:59:55Z

Prompt: Review the stack of #665 and below. Should we merge it into a single PR for better reviewability? […] Merge them and update the PR description. And close 641

Note

The following content was generated by AI.

Closed in favour of #664, which holds this PR's commits and now targets main. #664 replaced this PR's by=R, over=, into= syntax, so the two PRs are reviewed as one. The branch is kept.

Generated by Claude Code

claude added 25 commits September 23, 2026 14:29

Merge branch 'claude/confident-faraday-coc7mj-2' into claude/confiden…

336b407

…t-faraday-coc7mj-3

Merge remote-tracking branch 'origin/main' into claude/confident-fara…

6edc16f

…day-coc7mj

Merge branch 'claude/confident-faraday-coc7mj' into claude/confident-…

3eaaa7e

…faraday-coc7mj-2

Merge branch 'claude/confident-faraday-coc7mj-2' into claude/confiden…

8aeac2a

…t-faraday-coc7mj-3

Merge branch 'claude/confident-faraday-coc7mj-3' into claude/confiden…

4a4b969

…t-faraday-coc7mj-4

Merge branch 'claude/confident-faraday-coc7mj-4' into claude/confiden…

51c02bf

…t-faraday-coc7mj-6

Merge branch 'claude/confident-faraday-coc7mj-6' into claude/confiden…

c4da095

…t-faraday-coc7mj-7

Merge branch 'claude/confident-faraday-coc7mj-7' into claude/confiden…

e252384

…t-faraday-coc7mj-8

Merge branch 'claude/confident-faraday-coc7mj-8' into claude/confiden…

59d845b

…t-faraday-coc7mj-9

FBumann requested a review from brynpickering as a code owner September 23, 2026 21:11

claude added 3 commits September 23, 2026 21:38

Merge #661's head into the port of #641

5c0814c

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PHNF149r8B4tGmGUKmAdUd

claude added 2 commits September 23, 2026 22:07

Merge #661's head into the port of #641

b1b8417

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PHNF149r8B4tGmGUKmAdUd

This was referenced Sep 23, 2026

feat(language)!: a call names the columns of a relation as relation[column], and a sum through a relation is a sum over the axes its join opens #664

Open

feat(program): the axis a join opens names the relation column it stands for #665

Closed

FBumann added this pull request to stack #652 September 23, 2026 22:47

FBumann removed this pull request from stack #652 September 24, 2026 08:35

FBumann changed the base branch from claude/confident-faraday-coc7mj-11 to main September 24, 2026 09:44

FBumann added this pull request to stack #670 September 24, 2026 09:44

claude added 2 commits September 24, 2026 09:53

FBumann added the area: relations relations and dimensions: the relation design label Sep 24, 2026

claude added 2 commits September 25, 2026 08:05

Merge remote-tracking branch 'origin/main' into sync/662

e92e826

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Q8Y8AoPsJCyMsURHvgzm7h

Merge origin/main into claude/serene-goldberg-m8xwru

53406dc

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019TMqXLvfDGYm8hRND63cs8

FBumann pushed a commit that referenced this pull request Sep 30, 2026

chore: one changelog line for #662 and #664, which now land as one PR

770b8c8

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MKjqJocyxdRYvx7xin6NCM

FBumann closed this Sep 30, 2026

FBumann removed this pull request from stack #670 September 30, 2026 13:28

Navigation Menu

feat(program): a sum through a relation is a sum over the axes its join opens - #662

feat(program): a sum through a relation is a sum over the axes its join opens#662
FBumann wants to merge 36 commits into
mainfrom
claude/serene-goldberg-m8xwru

FBumann commented Sep 23, 2026 •

edited

Loading

Uh oh!

read-the-docs-community Bot commented Sep 23, 2026 •

edited

Loading

Uh oh!

read-the-docs-community Bot commented Sep 29, 2026 •

edited

Loading

Uh oh!

FBumann commented Sep 30, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

FBumann commented Sep 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this changes

Why

Why one commit and not a rebase

Conflicts and how they were resolved

Generated files

Docs sentences

Gates

Not done

Uh oh!

read-the-docs-community Bot commented Sep 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Documentation build overview

Uh oh!

read-the-docs-community Bot commented Sep 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Documentation build overview

Uh oh!

FBumann commented Sep 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

FBumann commented Sep 23, 2026 •

edited

Loading

read-the-docs-community Bot commented Sep 23, 2026 •

edited

Loading

read-the-docs-community Bot commented Sep 29, 2026 •

edited

Loading

FBumann commented Sep 30, 2026 •

edited

Loading