RSS / Atom

D Language

2014-04-19 , , ,

I recently had to write a moderate amount of C around vendor libraries and get things working in three days. The effort resulted in about 3200 lines of ugly, hastily written code that compiled under gcc and clang with -Wall -Werror -pedantic and looked reasonably leak-free in Valgrind. I whittled it down over the following week as time permitted to a bit over 2000 lines.

Lines of code is a lousy metric, but it’s easy to extract using available tools. Here are the counts from cloc for the original two projects in C:

Language                     files          blank        comment           code
C                                3            298            193           1869
C/C++ Header                     3             32              0            178
SUM:                             6            330            193           2047

Quick passes with grep counts 11, 13, and 16 functions.

A better metric might be complexity. Running it through the cyclomatic complexity checker cyclo results in scores of 79, 99, and 110. These seem completely off the charts (10-15 is the recommended range for a “module”). Quick runs against some code I had lying around shows that these numbers are high (OpenBSD’s strncpy and a naive CRC-16 implementation I had both score a 5, SolarDesigner’s concat function scores a 12) but we can use them for comparison. Let’s torture the numbers by averaging, 13.3 functions and a score of 96, for about 7.2 complexity per function.

I’m always on the look out for an advantage (I need them…) and saw a recent announcement of a release of the D Language compiler. The features piqued my interest: native compilation, garbage collection and dynamic memory, types, classes, interfaces, mixins, exceptions, extensive standard library, and an easy interface to C. Walter Bright, the language designer and writer of the reference compiler, is well known. There’s a decent website, a tutorial, and C++ guru Andrei Alexandrescu wrote the book on the language. Cool, I thought, a better C++. I bought the book, started coding familiar tasks, and dug in during my little free time.

It went easily and I had a satisfactory result in a few days. The equivalent in D:

Language          files     blank   comment      code    scale   3rd gen. equiv
D                     2       109        96       709 x   1.70 =        1205.30
SUM:                  2       109        96       709 x   1.70 =        1205.30

Notice that I’ve reduced the number of files, I’ll get to that. The pass through grep counts 8 and 13 functions, of which 3 of them in each source file are two-line logging routines to replace the classic C preprocessor macros dbgprintf and errprintf. cyclo doesn’t perfectly handle D, but it’s reasonably close in syntax to C/C++, producing scores of 45 and 61. Twisting numbers again gives 10.5 and 53 for a score of about 5 per function. The improvement in complexity is on par with the reduction in lines of code.

The kicker is that not only are the D versions are shorter and easier to comprehend, they do more and have more robust error-checking and exception-handling. Overall I’m more confident they behave correctly. I replaced two independent programs with a single one. I replaced a throttled fork()-ing server with a fixed pool of workers. I leveraged associative arrays, JSON, and Base64 without rolling my own or borrowing from other Open Source libraries. None of this is stuff I couldn’t implement or find somewhere but I didn’t have to. Classic force multiplier of a programming language.