DirectiveStrangeLove,OrHowILearnedToStopWorrying.ppt

Directive Strange Loveor: How I Learned To Stop Worrying And Love the Macro Bill McCloskey with Rob von Behren Jeremy Condit Feng Zhou starring Eric Brewer as Gen. “Buck” Turgidson George Necula as Dr. Strangelove Introduction The C preprocessor has serious problems Goal: To replace it with something better that has comparable functionality This talk will focus on macros, but includes and conditional compilation are future work Quick Overview The C preprocessor (cpp) operates on streams of tokens, not on syntax trees Leads to strange possibilities: This example is contrived, but actual code is at least as inscrutable Consequently, analysis tools ignore cpp Quick Overview Any analysis that ignores cpp will give inadequate error messages Same as with C++ templates Even if error messages are improved, compiler will never understand the semantics of a macro Side effects? Context (expression, statement, type)? Argument types? Quick Overview Consequently, macros are error-prone According to Ernst et al., ~23% of macros contain possible errors Unparenthesized formals Multiple formal uses Free variables Unparenthesized bodies Dangling semicolons Quick Overview Although macros present difficulties, they are useful Perl source code contains ~0.6 macros per line, and it would be unreadable without them Quick Overview Why not create a new language that is less error-prone and easier to analyze? Operate on syntax trees instead of tokens Eliminate weird semantics C++ made an effort in this direction Constants Inline functions But need something more expressive Overview The preprocessor is bad! How is the preprocessor used? What language features could replace it? How can we make use of old programs? Preprocessor Patterns Ernst et al. examine common macro uses Constants and expressions are 75% of definitions Statements, types, declarations account for most of the rest Goal: To design a replacement macro language that can express almost everything that cpp can expr