Excellent, thanks for collecting and reporting that data, Sergey. So it's not too bad, but still that says to me that caching dependencies would be beneficial for bigger programs. But anyway, you it would certainly be hard to beat your script in terms of simplicity.
Cached dependencies can be left as an exercise for the reader. :-) --bb On Sun, Dec 14, 2008 at 3:35 AM, Sergey Gromov <snake.sc...@gmail.com> wrote: > Sat, 13 Dec 2008 07:37:18 +0900, Bill Baxter wrote: > >> So how fast is dmd -o-? Does this pretty much double the compile time? >> An option to reuse a previously generated .deps might be nice. > > Here's the benchmark. > > I'm compiling dwt-samples\examples\helloworld\HelloWorld5.d from dwt > trunk rev. 119, without any pre-compilation. The system is a 1.83 GHz > Core2 Duo notebook under Win XP. The compiler is DMD 1.033/Tango. > > The -c -o- -v stage takes 0.97 seconds (best of five, all files in > memory, no swap). The result is a list of 202 files in win-dwt. That's > more than a half of dwt (357 .d files). > > The actual build phase takes 3.27 seconds, best of five, etc. The > result is a working 1.7 MB executable. > > So the -v phase is at least 3 times faster than the actual compilation. > Also note that in -o- mode the compiler misses some semantic errors > which means it doesn't do full analysis. >
