- Generated by
1.9.8
|
A11 (C++ runtime)
Native C++ implementation of the A11 streaming action runtime
|
One stream in the plan. More...
#include <cpp/a11/flow/graph.h>
Public Attributes | |
| RefKind | kind = RefKind::kExpr |
| std::string | label |
How a reader would say it: search.hits | truncate 200. | |
| BodyId | owner = kNone |
| The body this belongs to, which is where it is materialised. | |
| bool | writable = false |
| Whether this flow could ever write it, which is what makes it a destination rather than something finished by being read to its end. | |
| bool | unary = false |
| Whether this stream provably carries at most one value. | |
| long long | skip = 0 |
How many of this stream's first values skip n has spoken for. | |
| std::string | name |
kFlowPort, kCallPort, kNode: the port or node name. | |
| syntax::PortDirection | direction = syntax::PortDirection::kInput |
kFlowPort, kCallPort: which side it is on. | |
| StepId | call = kNone |
kCallPort: the call it belongs to. | |
| ExprId | id_expr = kNone |
kNode: the id expression it attaches to, and the map it lands in. | |
| std::string | node_map |
| RefId | subject = kNone |
kNodeId: the node. kStatus: the node or port it is the outcome of. | |
| StepId | subject_step = kNone |
kStatus: the call or barrier it is the outcome of. | |
| std::string | header |
kHeader: the header name, and the default when it was not sent. | |
| syntax::Constant | fallback |
| bool | has_fallback = false |
| ExprId | expr = kNone |
kExpr: the expression evaluated once. | |
| RefId | source = kNone |
kDerived: what it reads, and the stage applied to it. | |
| Stage | stage |
| std::string | role |
kBound: which of a loop's streams this is – item, index, carry. | |
| StepId | bound_by = kNone |
| std::vector< RefId > | sources |
kZip: the streams read in step, in the order they were written, which is the order their values appear in each tuple. | |
One stream in the plan.
A tagged struct rather than a class hierarchy: which fields mean anything depends on kind, and the runtime switches on it anyway. What every kind shares is the two questions the analysis asks – what it reads as a stream, and what it reads for one value – and those are answered by [Graph::Upstreams] and [Graph::ValueRefs] rather than by each kind separately.
| syntax::PortDirection a11::flow::graph::Ref::direction = syntax::PortDirection::kInput |
kFlowPort, kCallPort: which side it is on.
| syntax::Constant a11::flow::graph::Ref::fallback |
| bool a11::flow::graph::Ref::has_fallback = false |
| std::string a11::flow::graph::Ref::header |
kHeader: the header name, and the default when it was not sent.
kNode: the id expression it attaches to, and the map it lands in.
| RefKind a11::flow::graph::Ref::kind = RefKind::kExpr |
| std::string a11::flow::graph::Ref::label |
How a reader would say it: search.hits | truncate 200.
| std::string a11::flow::graph::Ref::name |
kFlowPort, kCallPort, kNode: the port or node name.
| std::string a11::flow::graph::Ref::node_map |
The body this belongs to, which is where it is materialised.
| std::string a11::flow::graph::Ref::role |
kBound: which of a loop's streams this is – item, index, carry.
| long long a11::flow::graph::Ref::skip = 0 |
How many of this stream's first values skip n has spoken for.
Applied where the stream is produced, upstream of the fan-out, so it is the same values every reader does not see. Several skip n statements naming one node add up: the ref is one object however many times the flow mentions it.
| std::vector<RefId> a11::flow::graph::Ref::sources |
kZip: the streams read in step, in the order they were written, which is the order their values appear in each tuple.
kMerge: the streams read at once, where the order they were written in says nothing about the order their values arrive in.
| Stage a11::flow::graph::Ref::stage |
kNodeId: the node. kStatus: the node or port it is the outcome of.
kStatus: the call or barrier it is the outcome of.
| bool a11::flow::graph::Ref::unary = false |
Whether this stream provably carries at most one value.
A claim, so the default is the absence of one: something that forgets to say is treated as a stream, which costs a check rather than being wrong. A declared port says so (in q: string against in q: string stream), an action's schema says so for its own ports, and everything built out of those is derived from them by [Builder::AddRef] – a reducing stage makes one value out of many, a per-value stage keeps the count it was given, and a node the flow writes from anywhere is never one.
What it is for: reading a stream where a value is expected. A unary stream can be consumed – take the value, and a second one is an error the language can name – while a stream of many has to say which value it means.
| bool a11::flow::graph::Ref::writable = false |
Whether this flow could ever write it, which is what makes it a destination rather than something finished by being read to its end.
1.9.8