Files
revng-revng/include/revng/Pipeline/ExecutionContext.h
Massimo Fioravanti c276a439b5 Add pipeline invalidation
Replace the stub implementation of invalidation with the proper
implementation. A ReadPathCache is added to each global so that it can
keep tracks of what target are associated to which read paths.
2024-01-02 11:14:56 +01:00

93 lines
3.1 KiB
C++

#pragma once
//
// This file is distributed under the MIT License. See LICENSE.md for details.
//
#include "llvm/ADT/StringRef.h"
namespace pipeline {
class Context;
class Target;
class ContainerBase;
class Step;
struct PipeWrapper;
// A execution context is created and passed to each invocation of a pipe. It
// provides a reference to the pipeline::Context, as well as carrying around
// information about what is the state of the pipeline while the pipe is being
// executed, such as what is the current step.
//
// The execution context is the intended way of letting the pipeline know when a
// target is created, so that it may keep track of what paths inside pipeline
// globals have been read to produce such target and thus it can be invalidated
// when of element of the global indicated by such paths changes.
//
// We will not explain here how exactly the pipeline achieve this, but as a
// mental model what is happening is that every access to any field of the model
// is instrumented. When a ExecutionContext is created, or a when it commits (at
// the end of the call, just after the commit happens), the entire tracking
// state of the global is reset, and from that moment forward each time a
// variable is accessed it is marked as being accessed.
//
// When commit with arguments Target and Container is inovked, the following
// MUST be true:
// * Target must be == to the last target that has been created in Container
// * no field of any global can be read in betweet the last target being created
// and Target being committed.
// * Target cannot be erased by Container after commit has been invoked.
// * No field of a global, directly or indirectly, can be read to produce a
// target that is not the next Target to be committed. (that is, you can't
// read fields to produce target X but then produce Y, commit Y, produce X,
// commit X, unless you use pop and push to save and restore the state of read
// fields).
//
// Less formally, you must produce and commit a target at the time.
//
// Example:
//
// void SomePipeProducingFunctions(const ExecutionContext& Ctx, SomeContainer&
// Container) {
// ...
// for (auto& Function : Container)
// {
// ...
// Ctx.getContext().pushReadFields();
// Ctx.commit(Container, Target(Function.metaadress()));
// Ctx.getContext().popReadFields();
// }
// }
//
class ExecutionContext {
private:
Context *TheContext = nullptr;
Step *CurrentStep = nullptr;
PipeWrapper *Pipe = nullptr;
// false when running on a analysis
bool RunningOnPipe = true;
void commit(const Target &Target, llvm::StringRef ContainerName);
public:
~ExecutionContext();
public:
ExecutionContext(Context &TheContext, Step &Step, PipeWrapper *Pipe);
public:
void commit(const ContainerBase &Container, const Target &Target);
void commitUniqueTarget(const ContainerBase &Container);
void clearAndResumeTracking();
public:
const Context &getContext() const { return *TheContext; }
Context &getContext() { return *TheContext; }
public:
const Step &getStep() const { return *CurrentStep; }
Step &getStep() { return *CurrentStep; }
};
} // namespace pipeline