On Wed, 16 Nov 2022 14:06:16 GMT, Erik Joelsson <er...@openjdk.org> wrote:

>> The sjavac ("smart javac") was an ambitious project. It should parallelize 
>> java compilation, create a background daemon process that kept the JVM "hot" 
>> with the JITted javac code, define a public api so only noticeable changes 
>> in a class caused further dependency compilations, etc etc.
>> 
>> Some of this never came to fruition. Other were implemented differently, as 
>> the `depend` plugin. The only thing we're currently using (and have been for 
>> the last few years) is the "server" functionality, that is, the ability to 
>> keep a single JVM process alive, and reuse the JITted code.
>> 
>> This code does not belong in the jdk.compiler module. It is a buildtool, 
>> pure and simple. Let's move it to it's proper place.
>
> Would it be worth keeping the relevant tests around in a separate test root 
> for buildtools, given that this buildtool is rather big and complex? Or is 
> the replacement so imminent that it's not worth the effort?

@erikj79 I'd also like to add that, at this point, the buildtool is not very 
complex. Right now it just listens to a port, and kicks off a javac compilation 
for each incoming call. It looks more complex since we've come here the long 
way around from a more complicated tool, and I did not spend too much time 
simplifying the code.

-------------

PR: https://git.openjdk.org/jdk/pull/11185

Reply via email to