ESL0002 - Relation syntax modification

LEP

ESL0002

Author(s)

  1. Wilschut

Reviewer(s)

A.T. Hofkamp, T.J.L. Schuijbroek

Status

Accepted

Type

Standard (S)

Created

24-10-2019

Finalized

t.b.d.

Abstract

In this LEP the relation parameter syntax is replaced with an extended form stating the type of each parameter and direction of the data with respect to the relation environment. This introduces specifying input-output relations. In addition, the interface of a relation is explicitly defined, which is useful for instantiating a relation as well as improving type checking.

Motivation

In engineering design detailed (mathematical) analysis is usually required to ensure that a system will function according to the set requirements. An analysis often consists of defining a mathematical function that requires a set of input arguments and returns a set of output arguments. Based on the values of the input arguments, the values of the output arguments are computed. The values of the output arguments are subsequently (manually) verified against the set requirements.

The current version of ESL contains relations, which can be used to denote the existence of a (mathematical) relation between variables. For example, in the listing below the relation definition newtons-second-law (line 2) is defined. Relation r1 (line 9) is an instance of newtons-second-law with arguments f, m, and a.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
define relation
  newtons-second-law

world
  variables
    f, m, a is a real

  relation
    r1: newtons-second-law with arguments
      f, m, a

Relation r1 denotes the existence a (mathematical) dependency between variables f, m, and a, which is described by relation definition newtons-second-law. The relation definition name newtons-second-law is considered to be an implicit reference to an external file in which the mathematical relation is implemented.

The current relation syntax, however, has a few shortcomings. First of all, the expected type of arguments and the number of arguments of newtons-second-law is not specified. As such, the compiler cannot verify correct use of a relation.

Secondly, a relation definition could be instantiated multiple times with different numbers of arguments of different types, which could lead to erroneous specifications.

Third of all, relations are considered to be undirected. That is, neither input nor output is specified. In practice, however, many analyses have distinct input and output variables. In the literature, input-output relations are often visualized in a parameter dependency structure matrix (DSM). By sequencing a parameter DSM one typically aims to obtain an optimal sequence in which the values of the parameters should be determined to minimize design iterations. Supporting such analysis fits well within the scope of ESL. However, due to the current lack of directionality it cannot be supported.

Hence, in this LEP, we replace the relation syntax with an extended version to enable type checking, to ensure one correctly instantiates relations, and enable one to define directed and undirected relations between variables. The number of parameters of a relations can be defined as a being fixed or arbitrary.

Rationale

The amount of information one has one a relation between variables may change during the course of a design project. At the start of a design project, one may only be aware of the existence of a relation. Later, one may obtain additional information that denotes the input-output relations. Therefore, the proposed relation syntax allows for the specification of directed and undirected relations. This enables users to convert undirected relations into directed relations during the course of the design project.

For example, the listing below shows an example specification in which two relations are defined and instantiated following the proposed syntax. At line 2, relation ohms-law is defined which requires parameters current and resistance as input and returns parameter voltage as output. At line 9, relation newtons-second-law is defined which relates parameters force, mass, and acceleration.

At line 21, ohms-law is instantiated as relation r1 which requires arguments I and R and returns argument U. At line 28, newtons-second-law is instantiated as relation r2 which relates arguments F, m, and a.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
define relation
  ohms-law
    requiring parameters
      current is a real
      resistance is a real
    returning parameter
      voltage is a real

 newtons-second-law
   relating parameters
     force is a real
     mass is a real
     acceleration is a real

 world
   variables
     I, R, U is a real
     F, m, a is a real

 relation
   r1: ohms-law
     requiring arguments
       I
       R
     returning argument
       U

   r2: newtons-second-law
     relating arguments
       F
       m
       a

The relations between the parameters of ohms-law are directed and follow a clear input-output pattern. The relation between the parameters of newtons-second-law are undirected. That is, the requiring keyword indicates that all parameters that follow are inputs to the relation. The returning keyword indicates that all parameters that follow are outputs of the relation. The relating keyword indicates that one has not decided whether the parameters that follow are input or output of the relation.

All parameters within relation definitions are assigned a type. The type of each parameter must match the type of the respective positional argument upon instantiation. For example, the type of argument a must be equal to the type of parameter acceleration. This mechanism allows for automated type checking. That is, the compiler can automatically check if all relations are instantiated with the correct number and correct type of arguments.

So far, the proposed syntax requires the user to define all relation parameters a priori. However, many mathematical formulas can take an arbitrary number of arguments. For example, a summation or balance equation. Therefore, the one or more keywords are introduced which indicate that a relation requires one or more parameters of a certain type. For example, in the following listing the relation cost-summation requires one or more cost arguments as input and returns the variable total-cost as argument.

1
2
3
4
5
6
define relation
  cost-summation
    requiring parameters
      one or more cost is a real
    returning parameter
      total-cost is a real

Specification

In this section the formal syntax and dependency derivation rules are given and discussed.

Syntax ====== The EBNF listing below shows the new syntax for relation definitions and a relation section. A relation-definition starts with the keywords define relation followed by a new line. The new line starts with a RELATION-DEF-NAME, which is the name of the relation definition. On the following lines the input, output and undirected parameters lists are specified. A parameter list starts with one of the keywords requiring, returning or relating followed by the keyword parameter followed by the respective input, output, or undirected parameter list.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
relation-definition ::=
    "define" "relation" \n
        { RELATION-DEF-NAME \n
            { ( "requiring" | "returning" | "relating" ) "parameter"  \n
                { ["one" "or" "more"] PARAMETER-NAME "is" ("a" | "an") TYPE-NAME \n }+
            }+ (*At most one requiring, returning and relating.*)
        }+

relation-section ::=
    "relation" \n
        { RELATION-NAME ":" RELATION-DEF-NAME \n
            { ( "requiring" | "returning" | "relating" ) "argument"  \n
                { ARGUMENT-NAME \n }+
            }+ (*At most one requiring, returning and relating.*)
        }+

A parameter list consists of one or more lines that start with the optional keywords one or more, a PARAMETER-NAME followed by the keywords is a or is an and a TYPE-NAME. A relation definition can have at most one input, one output, and one undirected parameter list. The order of the parameter lists is unconstrained.

A relation-section starts with the keyword relation and a new line. Next one can instantiate one or more relations. A relation instantiation starts with a RELATION-NAME followed by a :, a RELATION-DEF-NAME and a new line. Subsequently, one must specify the input, output, and undirected arguments lists.

Dependency derivations

The new syntax imposes a partitioning of the variables used by a relation. Therefore, a relation \(r \in \mathcal{R}\) is given:

\[r = (V, V', V'') \in \mathcal{R}\]

where \(V\) is the set of input variables, \(V'\) is the set of output variables and \(V''\) is the set of undirected variables. The partioning requires a reformulation of the set of relation dependencies as well. The set of dependencies between relations \(E_\mathrm{r}\) is given by:

\[\begin{split}\begin{array}{rll} E_\mathrm{r} =\{(r_i, r_j) &\mid& r_i = (V_i, V_i', V_i'') \in \mathcal{R}, r_j=(V_j, V_j', V_j'') \in \mathcal{R}, \\ & & ( V_i' \cup V_i'' ) \cap ( V_j \cup V_j'' ) \neq \emptyset, \\ & & r_i \neq r_j \} \end{array}\end{split}\]

That is, two relations \(r_i\) and \(r_i\) have a dependency if the intersection of the union of the output variables \(V_i'\) and undirected variables \(V_i''\) of relation \(r_i\) with the union of the input variables \(V_j\) and undirected variables \(V_j''\) of relation \(r_j\) is not empty. In other words, if an output or undirected variable of \(r_i\) is an input or undirected variable of \(r_j\), then \(r_j\) depends on \(r_i\). Note that if \(V''\) is empty for all \(r \in \mathcal{R}\) that all relation dependencies become directed.

The set of design dependencies between variables \(E_\mathrm{v_d}\) is affected by the partioning as well. Let \(\mathcal{Q}\) be the set of all design rules. Reformulation of \(E_\mathrm{v_d}\) then yields:

\[\begin{split}\begin{array}{rlll} E_\mathrm{v_d} = & \{(v_i, v_j) &\mid& (v_i, v_j) \in \mathcal{Q} \} \\ \cup & \{(v_i, v_j) &\mid& (V, V', V'') \in \mathcal{R}, \\ & & & v_i \in (V' \cup V''), v_j \in (V \cup V'') \} \\ \end{array}\end{split}\]

That is, two variables \(v_i\) and \(v_j\) have a dependency if there exists a design rule \((v_i, v_j) \in \mathcal{Q}\) or there exists a relation \(r = (V, V', V'') \in \mathcal{R}\) such that \(v_i\) is a member of the union of the output variables \(V'\) and undirected variables \(V''\) and \(v_j\) is a member of the union of the input variables \(V\) and undirected variables \(V''\) of \(r\). Note that if \(V''\) is empty for all \(r \in \mathcal{R}\) that all dependencies between variables become directed.

To determine the mapping relations between relations and all other ESL elements one can simply use the existing formulas and use \(V_r = V \cup V' \cup V''\) as the set of variables used by a relation \(r\).

Backwards Compatibility

This modification will break all existing ESL specifications that contain relation definitions and instantiations. As such, backwards compatibility issues are expected.

The simplest way to resolve these issues is to add relating parameters and parameter type information to all relation definitions and to add relating arguments to all relation instantiations.

Proof of concept

The section will be added once the syntax modifications have been implemented.

Rejected Ideas

The following ideas where rejected.

Open Issues

The following open issues are identified.

Format of mapping file

In the Rejected Ideas Section, it has been proposed to to create an external mapping file in which relation definition names are mapped to names of external definitions. The format of such files has yet to be designed.