Copyright Nancy Leveson, Sept. 1999 Programming Languages As difficult to discuss rationally as religion or politics. Prone to extreme statements devoid of data. Examples: "It is practically impossible to teach good programming to students that have had a prior exposure to BASIC; as potential programmers they are mentally mutilated beyond hope of regeneration." (Dijkstra) "The use of COBOL cripples the mind; its teaching should, therefore, be regarded as a criminal offence." (Dijkstra) Like anything else, decision making should be a rational process based on the priorities and features of the project. a0 c Copyright Nancy Leveson, Sept. 1999 Some Decision Factors c Features of application: Hard real time? Not just efficiency Predictability (need to guarantee deadlines will be met) High assurance? Portability? Maintainability? Others? Features of development environment: Availability of programmers, compilers, development tools? Schedule constraints? Others? a1 Copyright Nancy Leveson, Sept. 1999 Relationship between PL and Correctness Decreasing emphasis as an explicit design goal and research topic. Masterability Not complex: programmers understand it in its entirety The most important decisions in language design concern what is to be left out." (Wirth) Powerful features OK only if easy to use correctly. c Balance against need to keep language simple "Natural" Language should not surprise us in any of its effects. Should correspond to our experience with natural languages, mathematics, and other PLs a2 Copyright Nancy Leveson, Sept. 1999 Relationship between PL and Correctness (2) Error Proneness Language design should prevent errors. Should be difficult or impossible to write an incorrect program. If not possible, then allow their detection (as early as possible) Need for general principles and hypotheses so can predict error-prone features and improve language design. Some hypotheses and data about: Go to Global variables Pointers Selection by position (long parameter lists) Defaults and implicit type conversion Attempts to interpret intentions or fix errors c Meaning of features should be precisely defined (not dependent on compiler. a3 Copyright Nancy Leveson, Sept. 1999 Relationship between PL and Correctness (3) Understandability "The primary goal of a programming language is accurate communication among humans." Readability more important than writeability. Well "punctuated" (easy to directly determine statement types and major subunits without intermediate inferences) Use of distinct structural words (keywords, reserved words) for distinct concepts (no overloading, e.g., = for equal, assignment) Avoidance of multiple use of symbols unless serve completely analogous functions (e.g., commas as separators, parentheses for grouping). Necessary to be able to see what is being accomplished at a higher level of abstraction. Permit programmers to state their "intentions" along with instructions necessary to carry them out. a4 c Copyright Nancy Leveson, Sept. 1999 Relationship between PL and Correctness (4) Maintainability Locality -- possible to isolate changes. Self-documenting Programming decisions should be recorded in program, independent of external documentation. Good comment convention, freedom to choose meaningful variable names, etc. User-defined types and named constants e.g., type direction=(north, south, east, west) Explicit interfaces Should cater to construction of hierarchies of modules a5 c Copyright Nancy Leveson, Sept. 1999 Relationship between PL and Correctness (5) Checkability Every error should transform a correct program into one whose errors are detectable by the system. All error detection based on redundancy (but some forms can cause errors). Examples of userful redundancy: type declarations and type checking declarative redundancy invariance conditions or assertions Run-time assertions, exception handling checking subscripts vs. array bounds case selector vs. case bounds a6 c Copyright Nancy Leveson, Sept. 1999 Relationship between PL and Correctness (6) General High-level languages take many decisions out of programmer’s hands. One reason they are so fiercely resented by experienced programmers. Language should restrict programmer to decisions that really matter. Decisions should be recorded in program independent of external documentation. Simplicity of language less important than ability to write conceptually simple programs. a7 c c Copyright Nancy Leveson, Sept. 1999 Green: Design and Use of PLs "Clarifying the psychological processes of using programming languages will, I believe, clarify the requirements of language design and of environmental support." Some examples of structured programming hypotheses: Control structures should be hierarchical. Thus they should be nested, rather than allowed to have arbitrary branching. In this way, successive layers of detail can be added. The comprehensibility of hierarchically constructed programs will be easier, since they can be understoof by a reverse process -- understand the outer layer, then the inner layers, etc. These programs will be easy to modify because the inter-relations between parts will be simple. Have accepted these hypotheses but have never been validated. a8 Copyright Nancy Leveson, Sept. 1999 Green: Program Comprehension Cites experiments ("atheoretical" ) that evaluate only current programming practice. More interesting question: Can we elucidate underlying psychological principles to allow generalization of results to other classes of information structure in programming? Hypothesis 1: If one language is better than another, it c is always better, whatever the context. Hypothesis 2: Every notation highlights some type of information at the expense of others; the better notation for a given task is theone that highlights the information that given task needs. More generally, the comprehensibility of a notation may depend on the number and complexity of mental operations required to extract needed information. a0a10a9 Copyright Nancy Leveson, Sept. 1999 Green: Program Comprehension (2) Cites results supporting second hypothesis better than first. Not predicted by arguments of structured programming, which are based solely on presence or absence of good structure. Programmers were not simply decoding programming structure top down into some undescribed mental representation. Were reworking one structure into another. Difficulty of answering questions depended not only on source structure but also on relation between source and target structures. Observes that result "appears to raise insuperable difficulties for those simple-minded computer scientists who attempt to measure ‘psychological complexity’ of a program by means of a single number, such as McCabe." a0a11a0 c Statement of Problem: Fry: everything that is juicy but not hard Boil: everything that is hard Chop and roast: everything that is neither hard nor juicy L2: fry stop L1: boil stop chop roast stop if juicy go to L2 if hard go to L1 if hard then begin boil end else begin if juicy then begin fry end else begin chop roast end end if hard: boil not hard: if juicy: fry not juicy: chop roast end juicy end hard Using Dijkstra’s guarded command: if hard: boil if not hard, juicy: fry if not hard, not juicy: chop roast a0a12a1 c Copyright Nancy Leveson, Sept. 1999 Green: Program Creation Programs as plans. Role expressiveness: Outcome of a programmer’s effort is a structure in which each part plays some role vis-a-vis the programmer’s original intention. Easy program comprehension and creation requires role expressiveness. Rapid chunking into components Visible or easily inferred purposes for each part Visible or easily inferred relationships between each part and the larger structure. Important to alleviate mismatches between programmer’s task and program structure. Hypothesis that role expressiveness tends to detract from reusability of program fragments. "When a program fragment makes its role and purpose very clear, it is probably not easy to transport it unchanged to a new environment, because its role may be slightly but significantly different. a0a10a2 c Copyright Nancy Leveson, Sept. 1999 Green: Program Creation (2) Linguistic consistency Metarules vs. BNF (simplified syntax rules) Not the number of rules that counts but the consistency between the form of the rules. Similar roles should be indicated by similar syntax: syntax should map roles consistently. Significant omissions (defaults) Perceptual cues Humans not good at discerning structure of a string of arbitrary symbols but good at differentiating shapes, spatial positions, and other perceptual cues. Dangling else: use of indentation the best solution Green asks: "Why did it take so long to find the solution? a0 a3 c Copyright Nancy Leveson, Sept. 1999 Green: Other Topics Choice of programming paradigm (procedural, OO, functional, ...) Few studies comparing programming paradigms Programming medium Will new mediums affect design of languages and development of programs? Comprehension aid and documentation Different aids for different language types, different purposes Doxy: User definable browsers and filters Supply particular views of program Hide complexity irrelevant to user’s purpose. a0a12a4