On Sunday, 25 October 2015 at 03:22:39 UTC, Joakim wrote:
On Saturday, 24 October 2015 at 15:40:41 UTC, Jack Stouffer
wrote:
That's surprising given that many were worried that switching
to ddmd would slow compilation speeds down by at least 30%.
Also, this does not seem to be using any of ldc's optimization
flags.
Well, all three of those are ddmd: the only difference is
whether ddmd is compiled by dmd, gdc, or ldc. The 30%
measurement was based on comparing the previously completely
C++ dmd with ddmd:
http://forum.dlang.org/post/55c9f77b.8050...@dawg.eu
Whoops, posted before I was done writing.
The Travis CI run combines the time spent compiling ddmd, time
spent compiling the druntime/phobos tests, and then running the
tests. The original 30% comparison was only for time spent
compiling a D codebase, like phobos or vibe.d.
It's possible ldc takes longer to compile ddmd, but then the
resulting ddmd takes less time to compile phobos. That would
have to be separated out. It's also possible the backend is not
the issue and the D frontend itself is slower than the C++
frontend, in which case using ldc to compile ddmd won't make a
difference.