Sequent Prover Developer Guide: Difference between revisions

From Event-B
Jump to navigationJump to search
imported>Pascal
imported>Pascal
 
(8 intermediate revisions by 2 users not shown)
Line 1: Line 1:
{{TOCright}}
{{TOCright}}


= Programmer's Guide =
= Welcome to Event-B Sequent Prover =
== Welcome to Event-B Sequent Prover ==
The Event-B sequent prover is a plug-in for the Eclipse platform which provides different proof rules to be used within the Rodin Tool. The following sections discuss the issues when extending the Event-B sequent prover.  
The Event-B sequent prover is a plug-in for the Eclipse platform which provides different proof rules to be used within the Rodin Tool. The following sections discuss the issues when extending the Event-B sequent prover.  


== Sequent Prover Core ==
= Sequent Prover Core =
=== Sequents ===
== Sequents ==
{{TODO}}
{{TODO}}


=== Proof Rules ===
== Proof Rules ==
{{TODO}}
{{TODO}}


=== Proof Trees ===
== Proof Trees ==
{{TODO}}
{{TODO}}


=== Reasoners ===
== Reasoners ==
==== Concepts ====
=== Concepts ===
Reasoners are used in the Event-B sequent prover to generate [[#Proof_Rules | proof rules]]. A reasoner is provided with the sequent to prove as well as some extra optional input. The reasoner is implemented by a computer program and it is the job of the developers to ensure that the reasoners either fail or generate logically valid proof rules.
Reasoners are used in the Event-B sequent prover to generate [[#Proof_Rules | proof rules]]. A reasoner is provided with the sequent to prove as well as some extra optional input. The reasoner is implemented by a computer program and it is the job of the developers to ensure that the reasoners either fail or generate logically valid proof rules.


An example of a reasoner is the "conjI" reasoner that generates the conjI proof rules, which splits a conjunctive goal into two sub-goals. Here, assume that the input sequent is <math>A, B, C \vdash D \land E</math>, the conjI reasoner produces the following proof rules.
An example of a reasoner is the "conjI" reasoner that generates the conjI proof rules, which splits a conjunctive goal into two sub-goals. Here, assuming that the input sequent is <math>A, B, C \vdash D \land E</math>, the conjI reasoner produces the following proof rules.


{{InfRule||<math>\frac{\quad A, B, C \vdash D ~~~ A, B, C \vdash E\quad}{A, B, C \vdash D \land E}</math>}}
{{InfRule||<math>\frac{\quad A, B, C \vdash D ~~~ A, B, C \vdash E\quad}{A, B, C \vdash D \land E}</math>}}


==== Adding a reasoner ====
=== Adding a reasoner ===
A reasoner is added into the Event-B sequent prover using the extension point <tt>org.eventb.core.seqprover.reasoners</tt>. Below is an example of how to contribute to the extension point.
A reasoner is added into the Event-B sequent prover using the extension point <tt>org.eventb.core.seqprover.reasoners</tt>. Below is an example of how to contribute to the extension point.


Line 33: Line 32:
  </extension>
  </extension>


The above declaration defines a reasoner with a specific id (which will be automatically prefixed by the project name, e.g. org.eventb.contributors.seqprover). The name is an externalised string which is used for display purpose.
The above declaration defines a reasoner with a specific id (which will be automatically prefixed by the project name, e.g. <tt>org.eventb.contributors.seqprover</tt>). The name is an externalised string which is used for display purpose.


The class attribute must be a valid Java class name which will be used to create an instance of the reasoner. This class must implement the <tt>org.eventb.core.seqprover.IReasoner</tt> interface.
The class attribute must be a valid Java class name which will be used to create an instance of the reasoner. This class must implement the <tt>org.eventb.core.seqprover.IReasoner</tt> interface.


However, most of the time, developers only need to extend one of the abstract implementation (depending on the type of the reasoner). For this implementation purpose, the type of the reasoners are specified below:  
However, most of the time, developers only need to extend one of the abstract implementations (depending on the type of the reasoner). For this implementation purpose, the type of the reasoners are specified below:  
# Automatic rewriting reasoners: Reasoners which apply some rewriting rules automatically to simplify the sequent.  
# Automatic rewriting reasoners: Reasoners which apply some rewriting rules automatically to simplify the sequent.  
# Automatic inference reasoners: Reasoners which apply an inference rule automatically.  
# Automatic inference reasoners: Reasoners which apply an inference rule automatically.  
Line 45: Line 44:
Implementing a reasoner is only a first step of contributing to the sequent prover. In order to use a reasoner, the reasoner needs to be wrapped around by a tactic.  
Implementing a reasoner is only a first step of contributing to the sequent prover. In order to use a reasoner, the reasoner needs to be wrapped around by a tactic.  


==== Automatic Rewriting Reasoners ====
See also [[Versioned_Reasoners|Versioned Reasoners]].
===== Concepts =====
 
The reasoners of this type does not have any extra input (except the sequent) and automatically rewrite all the occurrences of certain formulas into other ones. An example of this is the rewriter which rewrites according to the following rules:  
=== Automatic Rewriting Reasoners ===
==== Concepts ====
The reasoners of this type do not have any extra input (except the sequent) and automatically rewrite all occurrences of certain formulas into other ones. An example of this is the rewriter which rewrites according to the following rules:  


* <math>\btrue \limp P \defi P</math>
* <math>\btrue \limp P \defi P</math>
Line 54: Line 55:
* <math>P \limp \bfalse \defi \lnot P</math>
* <math>P \limp \bfalse \defi \lnot P</math>


===== Implementation =====
==== Implementation ====
Step by step:  
Step by step:  
* Create a new plug-in project, e.g. <tt>org.eventb.contributor.seqprover.arith</tt>  
* Create a new plug-in project, e.g. <tt>org.eventb.contributor.seqprover.arith</tt>  
Line 65: Line 66:
      
      
* Implement the class: Click on the class attribute of the extension, a dialog will appear for new class wizard.  
* Implement the class: Click on the class attribute of the extension, a dialog will appear for new class wizard.  
* This class should extend the <tt>import org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AbstractAutoRewrites</tt>. For the moment, the class is internal within the project and subjected to be changed. Clients can make a copy of this class to develop their own implementation.  
* This class should extend the <tt>import org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AbstractAutoRewrites</tt>. For the moment, the class is internal within the plug-in and subject to change. Clients can make a copy of this class to develop their own implementation.  


:An example is as follows:  
:An example is as follows:  
Line 91: Line 92:
  }
  }
      
      
:The constructor shall call the super constructor <tt>super(IFormulaRewriter, boolean)</tt> to specify the formula rewriter to be used, and the reasoner shall hide the source of the rewritten hypothesis or not. Clients need to implement two methods:
:The constructor shall call the super constructor <tt>super(IFormulaRewriter, boolean)</tt> to specify the formula rewriter to be used, and the reasoner may hide the source of the rewritten hypothesis or not. Clients need to implement two methods:
:# The first one, namely <tt>getReasonerID()</tt>, returns the string ID of this reasoner. It must be the same ID as the one declared in the XML file (do not forget the project name prefix).  
:# The first one, namely <tt>getReasonerID()</tt>, returns the string ID of this reasoner. It must be the same ID as the one declared in the XML file (do not forget the project name prefix).  
:# The second method is <tt>getDisplayName()</tt>. It returns the display name of the reasoner, which will be used for display purpose in the UI.  
:# The second method is <tt>getDisplayName()</tt>. It returns the display name of the reasoner, which will be used for display purpose in the UI.  
Line 156: Line 157:
: See for example <tt>org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AutoRewriterImpl.t</tt>, which implements most of the automatic rewriting rules supplied by the Rodin platform.  
: See for example <tt>org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AutoRewriterImpl.t</tt>, which implements most of the automatic rewriting rules supplied by the Rodin platform.  


===== Testing =====
==== Testing ====
The abstract automatic rewriting reasoners have been tested, so clients only need to test if the reasoner has been declared correctly (testing the reasoner ID), and if the formula rewriter used for this reasoner has been implemented correctly.  
The abstract automatic rewriting reasoners have been tested, so clients only need to test if the reasoner has been declared correctly (testing the reasoner ID), and if the formula rewriter used for this reasoner has been implemented correctly.  


====== Testing the reasoner ID ======
===== Testing the reasoner ID =====
This test based on the abstract class <tt>org.eventb.core.seqprover.rewriterTests.AbstractAutomaticReasonerTests</tt>. An example is as follows.  
This test is based on the abstract class <tt>org.eventb.core.seqprover.rewriterTests.AbstractAutomaticReasonerTests</tt>. An example is as follows.  


  import org.eventb.core.seqprover.SequentProver;
  import org.eventb.core.seqprover.SequentProver;
Line 190: Line 191:
In this test, clients only need to provide the reasoner ID under test.  
In this test, clients only need to provide the reasoner ID under test.  


====== Testing the formula rewriter ======
===== Testing the formula rewriter =====
This test based on the abstract class <tt>org.eventb.core.seqprover.rewriterTests.AbstractFormulaRewriterTests</tt>. An example is as follows.  
This test is based on the abstract class <tt>org.eventb.core.seqprover.rewriterTests.AbstractFormulaRewriterTests</tt>. An example is as follows.  


  import org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AutoRewriterImpl;
  import org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AutoRewriterImpl;
Line 224: Line 225:
More instructions about how to use the method <tt>predicateTest</tt> and <tt>expressionTest</tt> (along with some other utility methods) can be found in the <tt>org.eventb.core.seqprover.rewriterTests.AbstractFormulaRewriterTests</tt> class.
More instructions about how to use the method <tt>predicateTest</tt> and <tt>expressionTest</tt> (along with some other utility methods) can be found in the <tt>org.eventb.core.seqprover.rewriterTests.AbstractFormulaRewriterTests</tt> class.


==== Automatic Inference Reasoners ====
=== Automatic Inference Reasoners ===
===== Concepts =====
==== Concepts ====
The reasoners of this type does not have any extra input (except the sequent) and implements an inference rule or a collection of inference rules. For example, a reasoner can implement the following inference rules which dealing with conjunction in the hypothesis and in goal.  
The reasoners of this type do not have any extra input (except the sequent) and implements an inference rule or a collection of inference rules. For example, a reasoner can implement the following inference rules which deal with conjunction in the hypothesis and in goal.  


{{InfRule||<math>\frac{\quad H, P \vdash Q \quad}{H \vdash P \limp Q}</math>}}
{{InfRule||<math>\frac{\quad H, P \vdash Q \quad}{H \vdash P \limp Q}</math>}}


===== Implementation =====
==== Implementation ====
In order to implement a reasoner of this type, clients extends the abstract class <tt>org.eventb.core.seqprover.reasonerInputs.EmptyInputReasoner</tt>. Example include <tt>org.eventb.internal.core.seqprover.eventbExtensions.ImpI</tt> which implements the above proof rule.  
In order to implement a reasoner of this type, clients extend the abstract class <tt>org.eventb.core.seqprover.reasonerInputs.EmptyInputReasoner</tt>. Examples include <tt>org.eventb.internal.core.seqprover.eventbExtensions.ImpI</tt> which implements the above proof rule.  


===== Testing =====
==== Testing ====
{{TODO}}
{{TODO}}


==== Manual Rewriting Reasoners ====
=== Manual Rewriting Reasoners ===
===== Concepts =====
==== Concepts ====
The reasoners of this type does not have a single string input (except the sequent) representing the position in the formula (either a hypothesis or the goal) where the rewriting occurs. An example of this is the rewriter which rewrites according to the following rules:  
The reasoners of this type have an input that refers to a predicate of the sequent (either a hypothesis or null if the goal is rewritten) and the position in that predicate where the rewriting occurs. An example of this is the rewriter which rewrites according to the following rules:  


{{TODO}}
{{TODO}}


===== Implementation =====
==== Implementation ====
{{TODO}}
{{TODO}}


===== Testing =====
==== Testing ====
{{TODO}}
{{TODO}}


==== Manual Inference Reasoners ====
=== Manual Inference Reasoners ===
===== Concepts =====
==== Concepts ====
The reasoners of this type does not have a single string input (except the sequent) representing the position in the formula (either a hypothesis or the goal) where most appropriate to indicate which inference rule to apply. For example, a reasoner can implement the following inference rules which dealing with conjunction in the hypothesis and in goal.  
The reasoners of this type have an input that refers to a predicate of the sequent (either a hypothesis or null if the goal is rewritten) and the position in that predicate where the rule is applied. For example, a reasoner can implement the following inference rules which deal with conjunction in the hypothesis and in goal.  


  H, P |- Q
{{InfRule||<math>\frac{\quad H, P \vdash Q \quad}{H \vdash P \limp Q}</math>}}
----------------- 
 
H |- P => Q
==== Implementation ====
   
{{TODO}}
===== Implementation =====
 
==== Testing ====
{{TODO}}
 
=== General Reasoners ===
==== Concepts ====
If the reasoner does not fall into the above categories, clients need to implement the interface <tt>IReasoner</tt> directly. It includes how to serialize/deserialize the reasoner input. More information to follow ...
 
{{TODO}}
 
==== Implementation ====
{{TODO}}
{{TODO}}


===== Testing =====
==== Testing ====
{{TODO}}
{{TODO}}


==== General Reasoner ====
== Tactics ==
=== Concepts ===
Tactics provide a way to structure strategic or heuristic knowledge about proof search. They provide control structures to call [[#Reasoners |reasoners]] or other tactics to make modifications to the [[#Proof_Trees | proof trees]]. In the Rodin platform, the users are dealing with tactics, not reasoners.


=== Tactics ===
A tactic can just be the wrapper around a reasoner, but it can be more complicated by combining different tactics.


== Event-B Tactics ==
==== Adding a tactic ====
=== POM Tactics ===
Tactics are divided into two classes: automatic and manual tactics.


=== Post-tactics ===
===== Automatic Tactics =====
A tactic is added into the Event-B sequent prover using the extension point <tt>org.eventb.core.seqprover.autoTactics</tt>.


= Reference =
  <extension point="org.eventb.core.seqprover.autoTactics">
    <autoTactic
            class="org.eventb.contributors.seqprover.TrueGoalTac"
            description="%trueGoalTacDesc"
            id="trueGoalTac"
            name="%trueGoalTacName">
    </autoTactic>
  </extension>


= Examples =
The above declaration defines an automatic tactic with a specific id (which will automatically be prefixed by the project name, e.g. org.eventb.contributors.seqprover). The name attribute is an externalised string used for display purposes. The description attribute is an externalised string that gives the detailed description of the tactic (what it does).


= Questions Index =
The <tt>class</tt> attribute must be a valid Java class name which will be used to create an instance of the tactic. This class must implement the <tt>org.eventb.core.seqprover.ITactic</tt> interface.
 
For automatic tactics, they must have no input, and most of the time they are just a wrapper around an automatic reasoner. Below is an example of such a tactic.
 
public static class AutoTac extends AbsractLazilyConstrTactic {
  @Override
  protected ITactic getSingInstance() {
    return BasicTactics.reasonerTac(new AutoReasoner(), EMPTY_INPUT);
  }
}
 
Here, the implementation extends the abstract class <tt>org.eventb.core.seqprover.eventbExtensions.AutoTactics#AbstractLazilyConstrTactic</tt>. <tt>AutoReasoner</tt> is the name of the class which implements an automatic reasoner expecting an empty input (See [[#Reasoners | reasoners]]).
 
===== Manual Tactics =====
{{TODO}}
 
= Event-B Tactics =
== Adding a POM Tactic ==
POM tactics are tactics that are run by the auto POM. Only automatic tactics (See [[#Tactics | tactics]]) can be declared as POM tactics.
 
POM tactics are added using the extension point <tt>org.eventb.core.pomTactic</tt>. An example is given below:
 
<extension point="org.eventb.core.pomTactics">
  <tactic id="org.eventb.core.seqprover.trueGoalTac"/>
</extension>
 
Each tactic element only contains an attribute id, which must be a valid tactic ID (including the project prefix) as declared using the <tt>org.eventb.core.seqprover.autoTactics</tt> extension point.
 
Once declared as a POM tactic, the information about the tactic will appear in the preferences for the POM tactic and the users of the platform. However, this tactic will not be a default POM tactic and the users must change the preferences to enable this tactic.
 
== Adding a Post Tactic ==
Post tactics are tactics that are run after every manual proof steps in interactive mode. Only automatic tactics (See [[#Tactics | tactics]]) can be declared as post tactics.
 
Post tactics are added using the extension point <tt>org.eventb.core.postTactic</tt>. An example is as follows:
 
<extension point="org.eventb.core.postTactics">
  <tactic id="org.eventb.core.seqprover.trueGoalTac"/>
</extension>
 
Each tactic element only contains an attribute id which must be a valid tactic ID (including the project prefix) as declared using the <tt>org.eventb.core.seqprover.autoTactics</tt> extension point.
 
Once declared as a post tactic, the information about the tactic will appear in the preferences for the post tactic and the users of the platform. However, this tactic will not be a default post tactic and the users must change the preferences to enable this tactic.
 
= UI Tactic Providers =
 
Tactic providers allow to define how manual tactics contribute to the User Interface.
 
{{TODO}}


= Legal =
See also [[New_Tactic_Providers | New Tactic Providers]].




[[Category:Developer documentation]]
[[Category:Developer documentation]]
[[Category:Work in progress]]
[[Category:Work in progress]]

Latest revision as of 09:48, 17 June 2010

Welcome to Event-B Sequent Prover

The Event-B sequent prover is a plug-in for the Eclipse platform which provides different proof rules to be used within the Rodin Tool. The following sections discuss the issues when extending the Event-B sequent prover.

Sequent Prover Core

Sequents

TODO

Proof Rules

TODO

Proof Trees

TODO

Reasoners

Concepts

Reasoners are used in the Event-B sequent prover to generate proof rules. A reasoner is provided with the sequent to prove as well as some extra optional input. The reasoner is implemented by a computer program and it is the job of the developers to ensure that the reasoners either fail or generate logically valid proof rules.

An example of a reasoner is the "conjI" reasoner that generates the conjI proof rules, which splits a conjunctive goal into two sub-goals. Here, assuming that the input sequent is A, B, C \vdash D \land E, the conjI reasoner produces the following proof rules.

\frac{\quad A, B, C \vdash D ~~~ A, B, C \vdash E\quad}{A, B, C \vdash D \land E}

Adding a reasoner

A reasoner is added into the Event-B sequent prover using the extension point org.eventb.core.seqprover.reasoners. Below is an example of how to contribute to the extension point.

<extension point="org.eventb.core.seqprover.reasoners">
  <reasoner
     class="org.eventb.contributors.seqprover.reasoners.Hyp"
     id="hyp"
     name="%hypName"/>
</extension>

The above declaration defines a reasoner with a specific id (which will be automatically prefixed by the project name, e.g. org.eventb.contributors.seqprover). The name is an externalised string which is used for display purpose.

The class attribute must be a valid Java class name which will be used to create an instance of the reasoner. This class must implement the org.eventb.core.seqprover.IReasoner interface.

However, most of the time, developers only need to extend one of the abstract implementations (depending on the type of the reasoner). For this implementation purpose, the type of the reasoners are specified below:

  1. Automatic rewriting reasoners: Reasoners which apply some rewriting rules automatically to simplify the sequent.
  2. Automatic inference reasoners: Reasoners which apply an inference rule automatically.
  3. Manual rewriters: Reasoners which apply a rewriting rule at a given position.
  4. Manual inference: Reasoner which apply an inference rule manually.
  5. General reasoners: If the reasoner does not fall into the above categories, clients need to implement the interface IReasoner directly.

Implementing a reasoner is only a first step of contributing to the sequent prover. In order to use a reasoner, the reasoner needs to be wrapped around by a tactic.

See also Versioned Reasoners.

Automatic Rewriting Reasoners

Concepts

The reasoners of this type do not have any extra input (except the sequent) and automatically rewrite all occurrences of certain formulas into other ones. An example of this is the rewriter which rewrites according to the following rules:

  • \btrue \limp P \defi P
  • P \limp \btrue \defi \btrue
  • \bfalse \limp P \defi \btrue
  • P \limp \bfalse \defi \lnot P

Implementation

Step by step:

  • Create a new plug-in project, e.g. org.eventb.contributor.seqprover.arith
  • Added dependency: org.eventb.core.seqprover.
  • Add an extension to the point org.eventb.core.seqprover.reasoners, e.g.,
class="org.eventb.core.seqprover.arith.ArithRewrites"
id="org.eventb.core.seqprover.arithmetic"
name="%arithRewritesName"
   
  • Implement the class: Click on the class attribute of the extension, a dialog will appear for new class wizard.
  • This class should extend the import org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AbstractAutoRewrites. For the moment, the class is internal within the plug-in and subject to change. Clients can make a copy of this class to develop their own implementation.
An example is as follows:
import org.eventb.core.ast.IFormulaRewriter;
import org.eventb.core.seqprover.SequentProver;
 
public class ArithRewrites extends AbstractAutoRewrites {
  private static final IFormulaRewriter rewriter = new ArithRewriterImpl();
   
  public AutoRewrites() {
    super(rewriter, true);
  }
   
  public static String REASONER_ID = SequentProver.PLUGIN_ID + ".autoRewrites";
   
  public String getReasonerID() {
    return REASONER_ID;
  }
   
  @Override
  protected String getDisplayName() {
    return "simplification rewrites";
  }
}
   
The constructor shall call the super constructor super(IFormulaRewriter, boolean) to specify the formula rewriter to be used, and the reasoner may hide the source of the rewritten hypothesis or not. Clients need to implement two methods:
  1. The first one, namely getReasonerID(), returns the string ID of this reasoner. It must be the same ID as the one declared in the XML file (do not forget the project name prefix).
  2. The second method is getDisplayName(). It returns the display name of the reasoner, which will be used for display purpose in the UI.
In order to implement the formula rewriter, clients are suggested to use Tom. The instructions are as follows:
  1. Create a file named build-tom.xml within your project. The best way to do this is to copy from org.eventb.core.seqprover, and modify this accordingly.
  2. Right click on the name of your project and choose "Properties".
  3. Choose "Builders", then click on "New" and choose "Ant Build" and click "OK". A dialog will open for the new builder wizard.
  4. At the "Main" tab:
    1. Enter the name of the builder (e.g., Tom).
    2. Buildfile: "Browse Workspace" to your build-tom.xml.
    3. Base Directory: "Variables" and choose "build_project".
  5. At the "Targets" tab:
    1. Auto Build: Click "Set Targets" and choose "tom" from the list.
    2. During a "clean": Click "Set Targets" and choose "clean" (unchecked "tom").
You will come back to this builder later on.
  • Implementing the rules:
    • Create a file called ArithRewriterImpl.t (the name should be consistent with the step above).
    • The content of the file should look like this:
package org.eventb.core.seqprover.arith;
 
import java.util.ArrayList;
import java.util.Collection;
import java.math.BigInteger;
 
import org.eventb.core.ast.AssociativeExpression;
import org.eventb.core.ast.AssociativePredicate;
import org.eventb.core.ast.AtomicExpression;
import org.eventb.core.ast.BinaryExpression;
import org.eventb.core.ast.BinaryPredicate;
import org.eventb.core.ast.BoolExpression;
import org.eventb.core.ast.BoundIdentDecl;
import org.eventb.core.ast.BoundIdentifier;
import org.eventb.core.ast.DefaultRewriter;
import org.eventb.core.ast.Expression;
import org.eventb.core.ast.Formula;
import org.eventb.core.ast.FormulaFactory;
import org.eventb.core.ast.FreeIdentifier;
import org.eventb.core.ast.Identifier;
import org.eventb.core.ast.IntegerLiteral;
import org.eventb.core.ast.LiteralPredicate;
import org.eventb.core.ast.Predicate;
import org.eventb.core.ast.QuantifiedExpression;
import org.eventb.core.ast.QuantifiedPredicate;
import org.eventb.core.ast.RelationalPredicate;
import org.eventb.core.ast.SetExtension;
import org.eventb.core.ast.SimplePredicate;
import org.eventb.core.ast.UnaryExpression;
import org.eventb.core.ast.UnaryPredicate;
import org.eventb.core.seqprover.eventbExtensions.Lib;
 
@SuppressWarnings("unused")
public class ArithRewriterImpl extends DefaultRewriter {
 
  public ArithRewriterImpl() {
    super(true, FormulaFactory.getDefault());
  }
  ...
}
See for example org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AutoRewriterImpl.t, which implements most of the automatic rewriting rules supplied by the Rodin platform.

Testing

The abstract automatic rewriting reasoners have been tested, so clients only need to test if the reasoner has been declared correctly (testing the reasoner ID), and if the formula rewriter used for this reasoner has been implemented correctly.

Testing the reasoner ID

This test is based on the abstract class org.eventb.core.seqprover.rewriterTests.AbstractAutomaticReasonerTests. An example is as follows.

import org.eventb.core.seqprover.SequentProver;
 
public class ArithRewriterReasonerTests extends AbstractAutomaticReasonerTests {
 
  @Override
  protected SuccessfulTest[] getSuccessfulTests() {
    // No need to test this. This should be guaranteed by testing the
    // abstract automatic rewrite reasoner and the formula rewriter itself.
    return new SuccessfulTest [] {
    };
  }
   
  @Override
  protected String[] getUnsuccessfulTests() {
    // No need to test this. This should be guaranteed by testing the
    // abstract automatic rewrite reasoner and the formula rewriter itself.
    return new String [] {
    };	
  }
   
  @Override
  public String getReasonerID() {
    return "org.eventb.core.seqprover.arithmetic";
  }
}
   

In this test, clients only need to provide the reasoner ID under test.

Testing the formula rewriter

This test is based on the abstract class org.eventb.core.seqprover.rewriterTests.AbstractFormulaRewriterTests. An example is as follows.

import org.eventb.internal.core.seqprover.eventbExtensions.rewriters.AutoRewriterImpl;
import org.junit.Test;
 
public class ArithFormulaRewriterTests extends AbstractFormulaRewriterTests {
  // The automatic rewriter for testing.
  private static final IFormulaRewriter rewriter = new ArithRewriterImpl();
   
  /**
   * Constructor.
   * 
   * Create an abstract formula rewriter test with the input is the automatic
   * rewriter.
   */
  public ArithFormulaRewriterTests() {
    super(rewriter);
  }
   
  /**
   * Tests for rewriting conjunctions.
   */
  @Test
  public void testRewrite() {
    // Test rewriting predicate
    predicateTest("⊥", "x = 1 ∧ ⊥");
    // Test rewriting expression
      expressionTest("1", "card({x + 1})");
  }
}
   

More instructions about how to use the method predicateTest and expressionTest (along with some other utility methods) can be found in the org.eventb.core.seqprover.rewriterTests.AbstractFormulaRewriterTests class.

Automatic Inference Reasoners

Concepts

The reasoners of this type do not have any extra input (except the sequent) and implements an inference rule or a collection of inference rules. For example, a reasoner can implement the following inference rules which deal with conjunction in the hypothesis and in goal.

\frac{\quad H, P \vdash Q \quad}{H \vdash P \limp Q}

Implementation

In order to implement a reasoner of this type, clients extend the abstract class org.eventb.core.seqprover.reasonerInputs.EmptyInputReasoner. Examples include org.eventb.internal.core.seqprover.eventbExtensions.ImpI which implements the above proof rule.

Testing

TODO

Manual Rewriting Reasoners

Concepts

The reasoners of this type have an input that refers to a predicate of the sequent (either a hypothesis or null if the goal is rewritten) and the position in that predicate where the rewriting occurs. An example of this is the rewriter which rewrites according to the following rules:

TODO

Implementation

TODO

Testing

TODO

Manual Inference Reasoners

Concepts

The reasoners of this type have an input that refers to a predicate of the sequent (either a hypothesis or null if the goal is rewritten) and the position in that predicate where the rule is applied. For example, a reasoner can implement the following inference rules which deal with conjunction in the hypothesis and in goal.

\frac{\quad H, P \vdash Q \quad}{H \vdash P \limp Q}

Implementation

TODO

Testing

TODO

General Reasoners

Concepts

If the reasoner does not fall into the above categories, clients need to implement the interface IReasoner directly. It includes how to serialize/deserialize the reasoner input. More information to follow ...

TODO

Implementation

TODO

Testing

TODO

Tactics

Concepts

Tactics provide a way to structure strategic or heuristic knowledge about proof search. They provide control structures to call reasoners or other tactics to make modifications to the proof trees. In the Rodin platform, the users are dealing with tactics, not reasoners.

A tactic can just be the wrapper around a reasoner, but it can be more complicated by combining different tactics.

Adding a tactic

Tactics are divided into two classes: automatic and manual tactics.

Automatic Tactics

A tactic is added into the Event-B sequent prover using the extension point org.eventb.core.seqprover.autoTactics.

 <extension point="org.eventb.core.seqprover.autoTactics">
   <autoTactic
           class="org.eventb.contributors.seqprover.TrueGoalTac"
           description="%trueGoalTacDesc"
           id="trueGoalTac"
           name="%trueGoalTacName">
   </autoTactic>
 </extension>

The above declaration defines an automatic tactic with a specific id (which will automatically be prefixed by the project name, e.g. org.eventb.contributors.seqprover). The name attribute is an externalised string used for display purposes. The description attribute is an externalised string that gives the detailed description of the tactic (what it does).

The class attribute must be a valid Java class name which will be used to create an instance of the tactic. This class must implement the org.eventb.core.seqprover.ITactic interface.

For automatic tactics, they must have no input, and most of the time they are just a wrapper around an automatic reasoner. Below is an example of such a tactic.

public static class AutoTac extends AbsractLazilyConstrTactic {
  @Override
  protected ITactic getSingInstance() {
    return BasicTactics.reasonerTac(new AutoReasoner(), EMPTY_INPUT);
  }
}

Here, the implementation extends the abstract class org.eventb.core.seqprover.eventbExtensions.AutoTactics#AbstractLazilyConstrTactic. AutoReasoner is the name of the class which implements an automatic reasoner expecting an empty input (See reasoners).

Manual Tactics

TODO

Event-B Tactics

Adding a POM Tactic

POM tactics are tactics that are run by the auto POM. Only automatic tactics (See tactics) can be declared as POM tactics.

POM tactics are added using the extension point org.eventb.core.pomTactic. An example is given below:

<extension point="org.eventb.core.pomTactics">
  <tactic id="org.eventb.core.seqprover.trueGoalTac"/>
</extension>

Each tactic element only contains an attribute id, which must be a valid tactic ID (including the project prefix) as declared using the org.eventb.core.seqprover.autoTactics extension point.

Once declared as a POM tactic, the information about the tactic will appear in the preferences for the POM tactic and the users of the platform. However, this tactic will not be a default POM tactic and the users must change the preferences to enable this tactic.

Adding a Post Tactic

Post tactics are tactics that are run after every manual proof steps in interactive mode. Only automatic tactics (See tactics) can be declared as post tactics.

Post tactics are added using the extension point org.eventb.core.postTactic. An example is as follows:

<extension point="org.eventb.core.postTactics">
  <tactic id="org.eventb.core.seqprover.trueGoalTac"/>
</extension>

Each tactic element only contains an attribute id which must be a valid tactic ID (including the project prefix) as declared using the org.eventb.core.seqprover.autoTactics extension point.

Once declared as a post tactic, the information about the tactic will appear in the preferences for the post tactic and the users of the platform. However, this tactic will not be a default post tactic and the users must change the preferences to enable this tactic.

UI Tactic Providers

Tactic providers allow to define how manual tactics contribute to the User Interface.

TODO

See also New Tactic Providers.