Myers, G. A controlled experiment in program testing and code walkthrough/inspections. Commun. ACM 21, 9(Sept. 1978), 760-768.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to check whether the author has resolved all defects. For this, one of the inspection participants verifies the defect resolution. Doolan reports that the moderator checks that the author has taken some remedial action for each defect detected [34] However, others do not report a follow up phase [95] [116] 122] They either did not perform one or did not Handbook on Software Engineering and Knowledge Engineering consider it important. Furthermore, many consider the follow up phase optional like the overview phase. Products The product dimension refers to the type of product that is ....

....60 for code inspections. Franz and Shih [43] present data from code inspection of a sales and inventory tracking systems project at HP. This was a batch system written in COBOL. Their data indicate that inspections had 19 effectiveness for defects that could also be found during testing. Myers [95] performed an experiment to compare program testing to code walkthroughs and inspections. This research is based on work performed earlier by Hetzel [53] The subjects were 59 highly experienced data processing professionals testing and inspecting a PL I program. Myers [95] reports an average ....

[Article contains additional citation context not shown here]

Myers, G. J., 1978. A controlled experiment in program testing and code walkthroughs /inspections. Communications of the ACM, 21(9):760-768. Handbook on Software Engineering and Knowledge Engineering

....expert systems in which thernoe was always one further important addition, one more significant and essential change. Working systems can contain multiple undetected errors. Preece and Shinghal [Preece Shinghal, 1992] document five fielded expert systems that contain numerous logical anomalies. Myers [Myers, 1977] reports that 51 experienced programmers could only ever find 5 of the 15 errors in a simple 63 line program, even given unlimited time and access to the source code and the executable. Potentially inaccurate and evolving theories must be validated, lest they generate inappropriate output for ....

Myers, G. (1977). A Controlled Experiment in Program Testing and Code Walkthroughs/Inspections. Communications of the ACM, 21:760--768.

....to run often enough. Bugs are found long after they re caused, so the cost of diagnosing and fixing them is much higher. Further, bugs may not be found at all: if the tests are judged manually, by a person looking at the output to see if it s correct, many failures will be missed ( Basili87] [Myers78]) While the long term benefits of an automated test suite are enormous, the startup cost can be high. For example, testing individual routines in isolation (unit testing) is too expensive except for the most critical routines. You spend too much time writing test drivers and the stubs that ....

Glenford J. Myers. "A Controlled Experiment in Program Testing and Code Walkthroughs/Inspections". Communications of the ACM, Vol. 21, No. 9, pp. 760-768, September, 1978.

....whether the author has resolved all defects. For this, one of the inspection participants verifies the defect resolution. Doolan reports that the moderator checks that the author has taken some remedial action for each defect detected [Doolan, 1992] However, others do not report a follow up phase [Myers, 1978], Russell, 1991] Shirey, 1992] They either did not perform or did not consider it important. Furthermore, many consider the follow up phase optional like the overview phase. 4.1.2 The Product Dimension The product dimension refers to the type of product that is usually inspected. Barry Boehm ....

....inspections. Franz and Shih [Franz and Shih, 1994] present data from code inspection of a sales and inventory tracking systems project at HP. This was a batch system written in COBOL. Their data indicate that inspections had 19 effectiveness for defects that could also be found during testing. Myers [Myers, 1978] performed an experiment to compare program testing to code walkthroughs and inspections. This research is based on work performed earlier by Hetzel [Hetzel, 1976] The subjects were 59 highly experienced data processing professionals testing and inspecting a PL I program. Myers reports an average ....

[Article contains additional citation context not shown here]

Myers, G. J. (1978). A controlled experiment in program testing and code walkthroughs / inspections. Communications of the ACM, 21(9):760--768.

....to improve software quality shows that formal technical review (FTR) provides unique and important benefits. Some studies provide evidence that FTR can be more effective at discovering errors than testing, while others indicate that it can discover different classes of errors than testing [19, 4]. In concert with other process improvements, Fujitsu found FTR to be so effective at discovering errors that they dropped system testing from their software development procedure [2] FTR forms an essential part of methods and models for very high quality software, such as Cleanroom Software ....

....to the needs and characteristics of the organization and application domain. 5 Regardless of its prescriptive or descriptive nature, literature on FTR suffers from a lack of high quality, empirical studies comparing different review methods to each other. Past studies compare review to testing [19, 14, 3] or compare different factors within a single review method (usually Fagan s inspection) such as the effect of the number of participants or group composition on review effectiveness [5, 18] Unfortunately, these latter comparative studies do not specify the actual activities that occurred in ....

[Article contains additional citation context not shown here]

G. Myers. A controlled experiment in program testing andcode walkthrough/ inspection. Communications of the ACM, 21(9):760--768, September 1978.

....6 for new requirements. In addition, we revisit the compiler example of Chapter 7 to illustrate how the intuitive decompositions used in this example are in fact given by general coercion rules. 9. 1 Program Construction The Naur problem has drawn the interest of several researchers in the past [10, 32, 55, 56, 58, 72, 73, 85] The purpose of this section is not to solve the Naur problem as much as it is to illustrate our method. In order to illustrate the versatility of our notation and method, we will investigate alternative design choices and discuss their merits. If our purpose was to find a solution to the problem, ....

....then par k : #:fw:in, out : out ffl n rap else par k : k #:fw:in 1, out : out ffl b rap, out : out ffl fw:in, in : rw:in qes qes . If we provide an appropriate implementation of the type sequence using external files in Pascal, our solution is comparable to the solution of Myers in [72], both in terms of length and algorithmic complexity. Myers solution is 63 lines long (in PL I) and O(#:in) in complexity, versus 100 lines and O(#:in) for ours. The difference in length is explained by the fact that we have an additional layer of abstraction by implementing a sequence data ....

Myers, G.J.: A Controlled Experiment in Program Testing and Code Walkthroughs / Inspections. CACM 21(9) (1978) 760--768.

....experiment with 39 student subjects to compare black box testing, white box testing, and individual code reading. The testing based techniques were found to be equally effective, while code reading was found to be inferior. His subjects only observed about 50 of the revealed failures. Myers [Mye78] performed an experiment with 59 professional programmers who tested a single program using three techniques. The techniques are very similar to those used in our experiment with the exception that his code readers worked in 3 person teams. Similarities in the results were that a large variability ....

Glenford J. Myers. A controlled experiment in program testing and code walkthroughs / inspections. Communications of the ACM, 21(9):760--768, September 1978.

....compared to the usual technique was not deemed sufficiently important for the required investment. There is justification for this position in that the literature has consistently demonstrated that formalised reviews, that is, software inspection, to be better than no reviews [5] or walkthroughs [62]. Natural Assemblage of Groups. For logistical reasons, training could only be done in groups of 10 subjects. Therefore, for each of the original quasi experiment and the two replications we had two groups of 10 subjects each. This group size seemed reasonable in that it would have been difficult ....

G. Myers. A Controlled Experiment in Program Testing and Code Walkthroughs / Inspections. Communications of the ACM, 21(9):760--768, September 1978.

....size. They are particularly inappropriate in comparison with parallel applications, which typically involve tens or hundreds of thousands of lines and hundreds of files and are unlikely to exemplify structured, well modularized programming styles. 4) Limited to serial programming techniques (cf. [7, 14]) A high proportion of the studies dealt with program comprehension, but always in terms of serial code and in many cases, just code fragments. The complexity of the program development process alters drastically when parallelism is 4 introduced, and the problems are heightened by the ....

Myers, G. J.: A Controlled Experiment in Program Testing and Code Walkthroughs/Inspections. Communications of the ACM. 21 (9): 760-768 (1978).

....mutation coverage is worthwhile by measuring how many faults are detected. In the meantime, weak mutation coverage cannot be recommended. There is one exception. Relational operator faults ( for = and the like) are common; indeed, they are the motivation behind testing boundary conditions. As [Myers78] observes, testers often think they are doing a #################################### 4 Recall again that I wrote the specifications. The black box numbers are less reliable than the broken box numbers. Recall also that the percentages reported are of total conditions, not just the feasible ....

Glenford J. Myers. "A Controlled Experiment in Program Testing and Code Walkthroughs/Inspections". Communications of the ACM, Vol. 21, No. 9, pp. 760-768, September, 1978.

.... structural testing strategies generally suffer from analysis of either small programs or a small 2 number of test subjects (testers) Most suggest that functional testing is a useful preliminary to code testing, creating a base of useful test cases with generally high code coverage [POC93, BS87, Mye78, HH79, KL95, HLL94] It is agreed that tests performed later in the software development process are more expensive to correct [Bei90, Mar91] and hence functional testing coupled with heavy unit testing can be a cheap and effective strategy. Horgan et al. HLL94] also recommend that both ....

G.J. Myers. A Controlled Experiment in Program Testing and Code Walkthroughs/Inspections. Communications of the ACM, pages 760 -- 768, Sept 1978.

....in a one line model (which they may have studied extensively in high school) then they should be suspicious of the 645 RAS assumption that the truth status of larger models can be accurately determined by visual inspection. Other studies encourage us to be cautious about the RAS assumption. Myers [80] reports controlled experiments with a 63 line model. 59 experienced data processing professionals hunted for errors in a very simple text formatter (63 line of PL 1 code) Even with unlimited time and the 650 use of three different methods , the experts could only find (on average) 5 of the 15 ....

G.J. Myers. A Controlled Experiment in Program Testing and Code Walk- 1550 throughs/Inspections. Communications of the ACM, 21:760--768, 9, September 1977.

.... 2 while mw do if lfw:in then in : rw:in else if out = ffl then fcw(ffl)g else if filled(fw:in) then fcw(n )g else fcw(b )g end end end end If we provide an appropriate implementation of the type sequence using external files in Pascal, our solution is comparable to the solution of Myers in [20], both in terms of length and algorithmic complexity. Myers solution is 63 lines long (in PL I) and O(#:in) in complexity, versus 100 lines and O(#:in) for ours. The difference in length is explained by the fact that we have an additional layer of abstraction by implementing a sequence data ....

Myers, G.J. A Controlled Experiment in Program Testing and Code Walkthroughs / Inspections. CACM 21(9) (1978) 760--768.

.... experiments in which 59 experienced data processing professionals hunted for errors in a very simple text formatter (63 line of PL 1 code) Even with unlimited time and the use of three different methods 19 , the experts could only find (on average) 5 of the 15 errors in this 63 line model (Myers 1977). Feldman Compton used a technique called hypothesis testing to show that neuroendocrinological theories published in international referred journals contain a surprisingly high percentage of errors. In one study, 109 of 343 (32 ) of the known data points from six studied papers could not be ....

Myers, G.J. (1977). A Controlled Experiment in Program Testing and Code Walkthroughs/Inspections.

....logicians; presumably, the most rational of all human beings [5, p362] He observes: Human reasoning does not always correspond to the prescriptions of logic. People fail to see as valid certain conclusions that are valid, and they see as valid certain conclusions that are not [5, p264] Myers [69] reports controlled experiments with a 63 line program. 59 experienced data processing professionals hunted for errors in a very simple text formatter (63 line of PL 1 code) Even with unlimited time and the use of three different methods , the experts could only find (on average) 5 of the 15 ....

G.J. Myers. A Controlled Experiment in Program Testing and Code Walkthroughs /Inspections. Communications of the ACM, 21:760--768, 9, September 1977.

....a one line model (which they may have studied extensively in high school) then they should be suspicious of the RAS assumption that the truth status of larger models can be accurately determined by visual inspection. Other studies encourage us to be cautious about the RAS 760 assumption. Myers [87] reports controlled experiments with a 63 line model. 59 experienced data processing professionals hunted for errors in a very simple text formatter (63 line of PL 1 code) Even with unlimited time and the use of three different methods , the experts could only find (on average) 5 of the 15 errors ....

G.J. Myers. A Controlled Experiment in Program Testing and Code Walkthroughs /Inspections. Communications of the ACM, 21:760--768, 9, September 1977.

....experiment with 39 student subjects to compare black box testing, white box testing, and individual code reading. The testing based techniques were found to be equally effective, while code reading was found to be inferior. His subjects only observed about 50 of the revealed failures. Myers [Mye78] performed an experiment with 59 professional programmers who tested a single program using three techniques. The techniques are very similar to those used in our experiment with the exception that his code readers worked in 3 person teams. Similarities in the results were that a large variability ....

Glenford J. Myers. A controlled experiment in program testing and code walkthroughs / inspections. Communications of the ACM, 21(9):760--768, September 1978.

.... ISERN Report ISERN 98 10 the form of controlled experiments compared the (cost )effectiveness of inspection and testing when applied independently of each other, i.e. the subjects of these experiments applied either inspection or testing on a software artifact but not both [3] 12] 17] [21], 28] However, as pointed out in [28] there is little clear, consistent evidence that one defect detection technique is more (cost )effective than another when applied in isolation. The existing evidence to date rather suggests that each defect detection technique has its own merits, e.g. by ....

....in isolation. The existing evidence to date rather suggests that each defect detection technique has its own merits, e.g. by catching different classes of defects. It is therefore often suggested that inspection and testing should be applied in combination rather than in isolation [3] 4] [21], 22] 28] However, few controlled empirical studies investigate this approach in more detail. Hence, its merits are rather unclear, i.e. what percentage of defects are detected by inspection, what percentage of defects are detected by testing, and what percentage of defects are not detected ....

[Article contains additional citation context not shown here]

G. J. Myers. A controlled experiment in program testing and code walkthroughs / inspections. Communications of the ACM, 21(9):760--768, Sept. 1978.

....important task in the debugging process. Different strategies for locating faults would therefore affect the performance of debugging. SCML79] A representative paragraph from Martin [MM83] states: Traditionally, programmers have spent too much time looking for errors in the wrong places. Myers [Mye78] found that programmers focused their attention on normal processing at the expense of considering special processing situations and invalid inputs. Weinberg [Wei71] found that programmers have difficulty finding errors because their conjectures become prematurely fixed, blinding them to other ....

Glenford J. Myers. A controlled experiment in program testing and code walkthrough /inspections. Communications of the ACM, 21(9):760--768, September 1978.

....experiment with 39 professional subjects to compare black box testing, white box testing, and individual code reading. The testing based techniques were found to be equally effective, while code reading was found to be inferior. His subjects only observed about 50 of the revealed failures. Myers [16] performed an experiment with 59 professional programmers who tested a single program Appeared in Proc. 5th European Software Engineering Conference, September 1995. 17 0 1 2 3 4 5 6 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Fault isolation rate (faults hour) Subject number ....

Glenford J. Myers. A controlled experiment in program testing and code walkthroughs / inspections. Communications of the ACM, 21(9):760--768, September 1978.

....Kamsties . 43.9 44.2 35.0 Roper . 43.4 55.1 57.9 Table 1: Average Percentage of defects found in each experiment tural testing. His results implied that the two testing techniques were equally effective, with code reading seeming to be inferior. ffl The next study was undertaken by Myers[28]; he compared informal versions of code reading, functional testing and structural testing; this time within a team oriented setting. He reported that all the techniques were of similar effectiveness, but that effectiveness seems to be highly dependent on defect type . He further demonstrated ....

G.J. Myers. A controlled experiment in program testing and code walkthoughs /inspections. Communications of the ACM, 21(9):760--768, 1978.

No context found.

G. J. Myers. A controlled experiment in program testing and code walkthroug hs/inspections. Communications of the ACM, 21(9):760--768, September 1978.

No context found.

Myers, G. A controlled experiment in program testing and code walkthrough/inspections. Commun. ACM 21, 9(Sept. 1978), 760-768.

No context found.

G.J. Myers. A controlled experiment in program testing and code walkthoughs /inspections. Communications of the ACM, 21(9):760--768,1978.

No context found.

Glenford J. Myers. A controlled experiment in program testing and code walkthroughs/- inspections. Communications of the ACM, 21(9):760--768, September 1978.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC