D needs to publicize its speed of compilation

[Joakim]

On Friday, 22 December 2017 at 11:39:48 UTC, user1234 wrote:
> If this ever happens, what will be published will have to be 
> based on LDC2.
> If benchmarks based on DMD are published, the article will be 
> subject to the criticism that is that the shorter build time is 
> due to the optimization pass, since it's known not to be super 
> deep in DMD backend.

Sure, ldc is slower but not significantly so, and will still beat 
the pants off clang, Rust, or Swift.

On Friday, 22 December 2017 at 11:46:49 UTC, Chris wrote:
> D does not do itself any favors when it keeps accepting 
> mediocre results in benchmarks.

These are all runtime codegen benchmarks, whereas I was talking 
about build speed.

> https://www.techempower.com/benchmarks/previews/round15/
>
> Some years it does not complete the tests, in others it shows 
> results that are below mediocre.
>
> Vibe.d ( and other D web frameworks ) benching twice slower 
> then scripting languages like PHP/Ruby. That does not advertise 
> D.

Sonke and others have noted that there are issues that need to be 
worked out with that benchmark submission.

> How about:
>
> http://benchmarksgame.alioth.debian.org/
>
> No D there? Performance must be bad because its not listed at 
> all ( for a language that exist 20 years )?

It was on there for many years but was tossed out by a subsequent 
maintainer, who doesn't want it for some reason, as thedeemon 
pointed out:

http://forum.dlang.org/post/lspybogtpqorauausmmv@forum.dlang.org

> Another one:
>
> https://github.com/costajob/app-servers
>
> Single threaded ( look at CPU total ), losing to other 
> languages again.

I think vibe.d is meant to be run in multiple process instances 
instead, and it does second-best to Rust on a single core (maybe 
beat out by Crystal too?), which is pretty good.

> Impressions are everything when there is a wealth of languages 
> to pick from. Anybody stumbling over these results think: Well, 
> i am better going with Go, Rust, Crystal, ... for a web hosting 
> as they show more consistent high speed results.

I wouldn't say they're "everything," but they intrigue people 
enough to dig deeper into the language or just to move on without 
bothering.  Of course, you may get great benchmarks with Rust but 
once you try coding with it, you still may not like using it.

On Friday, 22 December 2017 at 14:48:22 UTC, thedeemon wrote:
> On Friday, 22 December 2017 at 11:46:49 UTC, Chris wrote:
>> http://benchmarksgame.alioth.debian.org/
>>
>> No D there? Performance must be bad because its not listed at 
>> all ( for a language that exist 20 years )?
>
> D is not there for the only reason of that benchmark maintainer 
> unwilling to include D. Technically you can take any C solution 
> there, translate it to D (mostly by renaming from .c to .d) and 
> with LDC get the same speed as C shows.

Better than C actually, if you use D libraries instead:

http://forum.dlang.org/post/olelrjiaiwdxhjhilagv@forum.dlang.org

Getting back to the original topic of build speed, highlighting 
that the dmd compiler itself now builds in seconds after the 
C++->D translation would seem to be a layup:

http://forum.dlang.org/thread/oknuas$2lfp$1@digitalmars.com