• Jacob Carlborg (19/19) Sep 29 2020 This is looks pretty darn awesome. The Zig self-hosted compiler
• Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= (17/23) Sep 29 2020 I think it is a mistake to focus on raw batch compilation speed

Jacob Carlborg <doob me.com> writes:

This is looks pretty darn awesome. The Zig self-hosted compiler 
(which is far from complete in terms of language support) seems 
to be really fast. In the Zig compiler written in C++, most time 
is spent in LLVM optimizing and generating code. In the 
self-hosted compiler, everything is written in Zig: the frontend, 
the backend and linker.

It seems it supports incremental compilation of the semantic 
phase, the code generation and the linking phase. It doesn't seem 
like it's currently using incremental lexing or parsing. It 
doesn't even need to relink everything, it just patches the 
executable in-place.

This results in an incremental build taking around 0.5 
milliseconds. Here's a video showing this feature [1].

This is something that D should have, including incremental 
lexing and parsing as well. It's also a reason to keep the DMD 
backend.

[1] https://www.youtube.com/watch?v=R5FKgi9BYyU

--
/Jacob Carlborg

Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= <ola.fosheim.grostad gmail.com> writes:

On Tuesday, 29 September 2020 at 07:59:17 UTC, Jacob Carlborg 
wrote:
 This is looks pretty darn awesome. The Zig self-hosted compiler 
 (which is far from complete in terms of language support) seems 
 to be really fast. In the Zig compiler written in C++, most 
 time is spent in LLVM optimizing and generating code. In the 
 self-hosted compiler, everything is written in Zig: the 
 frontend, the backend and linker.

I think it is a mistake to focus on raw batch compilation speed 
though. What matters most is what kind of feedback you get while 
editing your code. Decent programmers that complicated languages 
like D/C++/Rust appeal to don't need to iterate that frequently 
in terms of batch-compilation (which also can be sped up by using 
a cluster if it really matters). Yes, I am aware that many will 
argue that it is very beneficial during testing, but I'd argue 
that if you cannot run tests in parallell, then maybe they should 
be written differently. CPU prices are going to fall faster than 
the labour it takes to rewrite a compiler from scratch and so are 
cloud computing solutions...

Where it does matter most is when you get fast compilation in 
combination with hot patching of running programs, like with 
Dart. Zig appears to be able to do that too using "hot swapping", 
but I have not tried Zig yet.