`srdatalog.ir.mir.types`¶

Python mirror of src/srdatalog/mir/mir_types.nim.

Pure IR data. Distinct from the existing python/mir_commands.py which emits C++ for mhk’s codegen and carries codegen-side state (cursor slots, program backref). My MirNode types carry only what the Nim MIR carries; they are suitable for S-expr printing and byte-diff against Nim golden.

Mapping to Nim: moColumnSource -> ColumnSource moScan -> Scan moColumnJoin -> ColumnJoin moCartesianJoin -> CartesianJoin moFilter -> Filter moNegation -> Negation moInsertInto -> InsertInto moExecutePipeline -> ExecutePipeline moRebuildIndex -> RebuildIndex moClearRelation -> ClearRelation moCheckSize -> CheckSize moComputeDelta -> ComputeDelta moComputeDeltaIndex -> ComputeDeltaIndex moMergeIndex -> MergeIndex moMergeRelation -> MergeRelation moRebuildIndexFromIndex -> RebuildIndexFromIndex moFixpointPlan -> FixpointPlan moBlock -> Block moProgram -> Program

All registered Nim MIR op kinds are now covered. Advanced ops (Aggregate, CreateFlatView, InnerPipeline, ProbeJoin, GatherColumn) have types + emitters but aren’t yet produced by Python lowering; they’re here so downstream codegen bridges and future lowering extensions have the node shapes to target.

Module Contents¶

Classes¶

`Aggregate`	(aggregate #:var cnt #:func AggCount #:index (Rel 0 1) #:ver FULL #:prefix (x y))
`BalancedScan`	(balanced-scan #:group-var v #:source1 (…) #:source2 (…))
`Block`	(block <instructions…>)
`CartesianJoin`	(cartesian-join #:vars (…) #:var-from-source ((…)) #:sources (…))
`CheckSize`	(check-size #:schema R #:ver V)
`ClearRelation`	(clear-relation #:schema R #:ver V)
`ColumnJoin`	(column-join #:var x #:sources (…))
`ColumnSource`	(column-source #:index (Rel cols…) #:ver V #:prefix (vars))
`ComputeDelta`	(compute-delta #:schema R)
`ComputeDeltaIndex`	(compute-delta-index #:schema R #:canonical-index (cols…))
`ConstantBind`	(constant-bind #:var v #:code “…” #:deps (…))
`CreateFlatView`	(create-flat-view #:schema R #:index (cols…) #:ver V)
`ExecutePipeline`	(execute-pipeline #:rule N #:sources (tuple …) #:dests (tuple …) )
`Filter`	(filter #:vars (…) #:code “…”)
`FixpointPlan`	(fixpoint-plan <instructions…>)
`GatherColumn`	(gather-column …) — binary-join mode node. Dereferences `input_buffer` of row IDs into values of `column` from `rel_name`, binding the result to `output_var`.
`InjectCppHook`	(inject-cpp-hook #:rule R #:code “…”) — raw C++ injection point.
`InnerPipeline`	(inner-pipeline #:rule R #:bound-vars (…) #:handles (…) #:ops (…))
`InsertInto`	(insert-into #:schema R #:ver V #:dedup-index (cols…) #:terms (vars))
`MergeIndex`	(merge-index #:index (R cols…))
`MergeRelation`	(merge-relation #:schema R)
`Negation`	(negation #:schema R #:ver V #:index (R cols…) #:prefix (…))
`ParallelGroup`	(parallel-group <ops…>) — independent ops that can run concurrently.
`PositionedExtract`	(positioned-extract #:var v #:sources ((…) (…)) #:bind (…))
`PostStratumReconstructInternCols`	(post-stratum-reconstruct-intern-cols #:rel R #:canonical-index (cols…))
`ProbeJoin`	(probe-join …) — binary-join mode node. Performs a binary-search probe of `probe_rel` keyed on `join_key`, writing row-id pairs into `output_buffer`. Uses merge-path balancing for unbalanced output.
`Program`	(program (step #:recursive b ) …)
`RebuildIndex`	(rebuild-index #:index (R cols…) #:ver V)
`RebuildIndexFromIndex`	(rebuild-index-from-index #:source (R cols…) #:target (R cols…) #:ver V)
`Scan`	(scan #:vars (…) #:index (Rel cols…) #:ver V #:prefix (…))

Data¶

MirNode

API¶

class srdatalog.ir.mir.types.Aggregate[source]¶

(aggregate #:var cnt #:func AggCount #:index (Rel 0 1) #:ver FULL #:prefix (x y))

Aggregation body clause lowered to MIR. result_var is the variable bound by the aggregate; func is the C++ aggregation type (AggCount / AggSum / … / custom). prefix_vars are join-prefix vars read by the aggregate’s index lookup.

agg_func: str¶: None

handle_start: int¶: None

index: list[int]¶: None

prefix_vars: list[str]¶: ‘field(…)’

rel_name: str¶: None

result_var: str¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.BalancedScan[source]¶

(balanced-scan #:group-var v #:source1 (…) #:source2 (…))

Pre-computes a work-distribution histogram for a skewed join: partition work across (source1 × source2) pairs grouped by group_var (the “balanced root”). Emitted by lowering when a rule’s plan specifies balanced_root + balanced_sources.

group_var: str¶: None

handle_start: int¶: None

source1: srdatalog.ir.mir.types.ColumnSource¶: None

source2: srdatalog.ir.mir.types.ColumnSource¶: None

vars1: list[str]¶: ‘field(…)’

vars2: list[str]¶: ‘field(…)’

class srdatalog.ir.mir.types.Block[source]¶

(block <instructions…>)

instructions: list[MirNode]¶: None

class srdatalog.ir.mir.types.CartesianJoin[source]¶

(cartesian-join #:vars (…) #:var-from-source ((…)) #:sources (…))

handle_start: int¶: None

sources: list[srdatalog.ir.mir.types.ColumnSource]¶: None

var_from_source: list[list[str]]¶: ‘field(…)’

vars: list[str]¶: None

class srdatalog.ir.mir.types.CheckSize[source]¶

(check-size #:schema R #:ver V)

rel_name: str¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.ClearRelation[source]¶

(clear-relation #:schema R #:ver V)

rel_name: str¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.ColumnJoin[source]¶

(column-join #:var x #:sources (…))

handle_start: int¶: None

sources: list[srdatalog.ir.mir.types.ColumnSource]¶: None

var_name: str¶: None

class srdatalog.ir.mir.types.ColumnSource[source]¶

(column-source #:index (Rel cols…) #:ver V #:prefix (vars))

clause_idx: int¶: None

handle_start: int¶: None

index: list[int]¶: None

prefix_vars: list[str]¶: ‘field(…)’

rel_name: str¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.ComputeDelta[source]¶

(compute-delta #:schema R)

index: list[int]¶: ‘field(…)’

rel_name: str¶: None

class srdatalog.ir.mir.types.ComputeDeltaIndex[source]¶

(compute-delta-index #:schema R #:canonical-index (cols…))

canonical_index: list[int]¶: None

rel_name: str¶: None

class srdatalog.ir.mir.types.ConstantBind[source]¶

(constant-bind #:var v #:code “…” #:deps (…))

Generated by head-constant rewriting: a HIR LetClause lowers to this MIR node, which binds var_name to code once deps are bound.

code: str¶: None

deps: list[str]¶: None

var_name: str¶: None

class srdatalog.ir.mir.types.CreateFlatView[source]¶

(create-flat-view #:schema R #:index (cols…) #:ver V)

Emitted by split-rule lowering to expose temp-relation intern columns as an unsorted view (avoids a GPU sort between Pipeline A and B).

index: list[int]¶: None

rel_name: str¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.ExecutePipeline[source]¶

(execute-pipeline #:rule N #:sources (tuple …) #:dests (tuple …) )

block_group: bool¶: False

clause_order: list[int]¶: ‘field(…)’

concurrent_write: bool¶: False

count: bool¶: False

dedup_hash: bool¶: False

dest_specs: list[srdatalog.ir.mir.types.InsertInto]¶: None

pipeline: list[MirNode]¶: None

rule_name: str = <Multiline-String>¶

source_specs: list[Union[srdatalog.ir.mir.types.ColumnSource, srdatalog.ir.mir.types.Scan, srdatalog.ir.mir.types.Negation, srdatalog.ir.mir.types.Aggregate]]¶: None

use_fan_out: bool¶: False

work_stealing: bool¶: False

class srdatalog.ir.mir.types.Filter[source]¶

(filter #:vars (…) #:code “…”)

code: str¶: None

vars: list[str]¶: None

class srdatalog.ir.mir.types.FixpointPlan[source]¶

(fixpoint-plan <instructions…>)

instructions: list[MirNode]¶: None

schema_arities: list[tuple[str, int]]¶: ‘field(…)’

class srdatalog.ir.mir.types.GatherColumn[source]¶

(gather-column …) — binary-join mode node. Dereferences input_buffer of row IDs into values of column from rel_name, binding the result to output_var.

column: int¶: None

input_buffer: str = <Multiline-String>¶

output_var: str¶: None

rel_name: str¶: None

rel_version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.InjectCppHook[source]¶

(inject-cpp-hook #:rule R #:code “…”) — raw C++ injection point.

The Nim emitter prints #:code “…” without the actual code body (just the ellipsis), matching debug/tooling display; real content stays in the node’s code field for codegen.

code: str¶: None

rule_name: str = <Multiline-String>¶

class srdatalog.ir.mir.types.InnerPipeline[source]¶

(inner-pipeline #:rule R #:bound-vars (…) #:handles (…) #:ops (…))

JIT-generated inner device function for nested joins (Level 2+). Nim emits an explicit C++ functor rather than recursive template metaprogramming.

bound_vars: list[str]¶: ‘field(…)’

inner_ops: list[MirNode]¶: ‘field(…)’

input_handles: list[MirNode]¶: ‘field(…)’

rule_name: str¶: None

class srdatalog.ir.mir.types.InsertInto[source]¶

(insert-into #:schema R #:ver V #:dedup-index (cols…) #:terms (vars))

index: list[int]¶: None

rel_name: str¶: None

vars: list[str]¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.MergeIndex[source]¶

(merge-index #:index (R cols…))

index: list[int]¶: None

rel_name: str¶: None

class srdatalog.ir.mir.types.MergeRelation[source]¶

(merge-relation #:schema R)

rel_name: str¶: None

srdatalog.ir.mir.types.MirNode¶: None

class srdatalog.ir.mir.types.Negation[source]¶

(negation #:schema R #:ver V #:index (R cols…) #:prefix (…))

const_args: list[tuple[int, int]]¶: ‘field(…)’

handle_start: int¶: None

index: list[int]¶: None

prefix_vars: list[str]¶: ‘field(…)’

rel_name: str¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.ParallelGroup[source]¶

(parallel-group <ops…>) — independent ops that can run concurrently.

ops: list[MirNode]¶: None

class srdatalog.ir.mir.types.PositionedExtract[source]¶

(positioned-extract #:var v #:sources ((…) (…)) #:bind (…))

After a BalancedScan binds its group variable, any subsequent ColumnJoin for that variable becomes a positioned extract: point-lookup rather than iteration.

bind_vars: list[str]¶: ‘field(…)’

sources: list[srdatalog.ir.mir.types.ColumnSource]¶: None

var_name: str¶: None

class srdatalog.ir.mir.types.PostStratumReconstructInternCols[source]¶

(post-stratum-reconstruct-intern-cols #:rel R #:canonical-index (cols…))

Emitted once per stratum per relation: single cleanup step that replaces a per-index RebuildIndex loop, pushing complexity into the C++ runtime.

canonical_index: list[int]¶: None

rel_name: str¶: None

class srdatalog.ir.mir.types.ProbeJoin[source]¶

(probe-join …) — binary-join mode node. Performs a binary-search probe of probe_rel keyed on join_key, writing row-id pairs into output_buffer. Uses merge-path balancing for unbalanced output.

input_buffer: str = <Multiline-String>¶

join_key: str¶: None

output_buffer: str = <Multiline-String>¶

probe_index: list[int]¶: None

probe_rel: str¶: None

probe_version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.Program[source]¶

(program (step #:recursive b ) …)

steps: list[tuple[MirNode, bool]]¶: None

class srdatalog.ir.mir.types.RebuildIndex[source]¶

(rebuild-index #:index (R cols…) #:ver V)

index: list[int]¶: None

rel_name: str¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.RebuildIndexFromIndex[source]¶

(rebuild-index-from-index #:source (R cols…) #:target (R cols…) #:ver V)

rel_name: str¶: None

source_index: list[int]¶: None

target_index: list[int]¶: None

version: srdatalog.ir.hir.types.Version¶: None

class srdatalog.ir.mir.types.Scan[source]¶

(scan #:vars (…) #:index (Rel cols…) #:ver V #:prefix (…))

handle_start: int¶: None

index: list[int]¶: None

prefix_vars: list[str]¶: ‘field(…)’

rel_name: str¶: None

vars: list[str]¶: None

version: srdatalog.ir.hir.types.Version¶: None

srdatalog.ir.mir.types¶

Module Contents¶

Classes¶

Data¶

API¶

`srdatalog.ir.mir.types`¶