C Panel Discussion Summary

1
C Panel Discussion Summary

• Jim Kowalkowski

2
Participants

• Amber Boehnlein
• Jim Kowalkowski
• Leo Michelotti
• Marc Paterno
• Liz Sexton-Kennedy
• Bjarne Stroustrup (distinguished panelist)

3
Day One

• Talks
• Frustration with the language
• Speed
• Persistence

4
Talks

• Interesting problems that have been addressed
• Abstraction and modularity are key concepts
• Experiences with templates and the Standard
  Library

5
Frustrations

• Questions and concerns
• Decomposing problem is difficult to do properly
• Requires design time
• Slower to produce results
• Appropriateness of algorithm/data separation
• Answers
• Good design yields flexible, maintainable, and
  correct code
• Prototyping is very important, distinguish it
  from production code
• Striking a balance in separating concepts is
  difficult, extremes are usually bad

6
Speed

• Questions and concerns
• Fortran is apparently faster
• Answers
• Performance testing is difficult, one must
  compare very similar things
• Cannot ignore efficiency when designing and
  implementing code
• Well designed code is generally easier to
  optimize
• Correctness is more important than speed
• Extremely fast programs that produce erroneous
  results, even part of the time, are incorrect

7
Day Two

• Missing features
• Use of language

8
Missing Features

• Questions and concerns
• Persistency
• Reflection
• Compiler optimization hints
• Answers
• Join committee
• No consensus (standards require this)
• Not pushed hard by groups in the committee

9
Use of Language Subset

• Questions and concerns
• Exceptions (cost)
• Templates (compiler support)
• Answers
• Take fresh look at exceptions and templates, much
  improved compiler support
• Turning off exceptions is not using Standard C
• Missing out on the strengths of C

10
C is Difficult

• Questions and concerns
• Abundant Fortran expertise, not so with C
• Difficult to read C
• Answers
• Fortran skill and code was not very good
• Poor C is hard to read. Good C code matches
  the concepts and is straightforward to read and
  maintain
• Rely on other code, concentrate on the
  application, do not need to understand every
  detail

11
Use of Standard Library

• Questions and concerns
• Algorithms are Weird
• Different style than many are accustomed to
• Answers
• Missing out on the strengths of C
• Algorithms allow direct expression of concepts in
  code
• Efficient
• Allow many optimization opportunities
• Best available algorithms

12
Session Highlights

• First day
• Design aspects
• Frustrations
• Second day
• Language features
• Language use

13
Stroustrups talk

• Majority of talk discussed techniques involving
  generic programming
• Concentrated on use of function objects and
  algorithms
• Classes and code should map clearly and cleanly
  onto the concepts and intention of the program
• Stressed importance of building classes that
  directly model a single concept.
• Stressed that class hierarchies should be kept as
  simple as possible (hierarchies of pure interface)

14
Summary

• Use the language features that are appropriate
  for solving the problem
• Code to the standard, not a subset
• Need for representation on the C committee
• Take a look at the generic programming aspects of
  the language.
• The combination of OO and generic programming is
  the strength of the language