Overcompilation when no files have changed

[keynmol]

Version(s)
1.5.0

Describe the bug
When doing scala-cli run . --native, it seems that the project is being recompiled (I can see warnings being re-emitted) even though nothing has changed:

/v/f/j/4/T/tmp.d938Ip9W80 > scli run . --native
Compiling project (Scala 3.5.1, Scala Native 0.5.5)
[warn] ./test.scala:6:7
[warn] unused local definition
[warn]   val x = 25
[warn]       ^
Compiled project (Scala 3.5.1, Scala Native 0.5.5)
[info] Linking (multithreadingEnabled=true, disable if not used) (1046 ms)
[info] Discovered 888 classes and 5362 methods after classloading
[info] Checking intermediate code (quick) (41 ms)
[info] Multithreading was not explicitly enabled - initial class loading has not detected any usage of system threads. Multithreading support will be disabled to improve performance.
[info] Linking (multithreadingEnabled=false) (317 ms)
[info] Discovered 510 classes and 2557 methods after classloading
[info] Checking intermediate code (quick) (23 ms)
[info] Discovered 487 classes and 1958 methods after optimization
[info] Optimizing (debug mode) (396 ms)
[info] Produced 9 LLVM IR files
[info] Generating intermediate code (395 ms)
[info] Compiling to native code (932 ms)
[info] Linking with [pthread, dl]
[info] Linking native code (immix gc, none lto) (138 ms)
[info] Postprocessing (0 ms)
[info] Total (3314 ms)
yes
/v/f/j/4/T/tmp.d938Ip9W80 > scli run . --native
[warn] ./test.scala:6:7
[warn] unused local definition
[warn]   val x = 25
[warn]       ^
yes
/v/f/j/4/T/tmp.d938Ip9W80 > scli run . --native
[warn] ./test.scala:6:7
[warn] unused local definition
[warn]   val x = 25
[warn]       ^
yes

This adds overhead to the command, where actually running the binary takes 2ms on my laptop, but running scala-cli run . --native takes 30ms, even though it literally just runs the binary (or at least should). A user reports even higher overhead on a non-trivial program.

I don't know if Bloop actually compiles anything, but this certainly takes quite a bit of time (in relative terms)

To Reproduce

//> using toolkit default
//> using option -Wunused:all
//> using scala 3.5.1

@main def hello = 
  val x = 25
  println("yes")

Then run scala-cli run . --native a bunch of times. Warnings are re-emitted every time. Running with -v -v actually shows the bloop request being sent.

Expected behaviour

When no input files or parameters have changed, I expect the overhead of scala-cli run to be much lower.

[tgodzik]

I think we always send last warnings and do not compile anything inside Bloop. This is a correct behaviour in my opinion.

On the other hand though, I don't really understand the overhead. If we have no warnings present is it still an issue?

[keynmol]

Heh, the overhead is even larger :D

/v/f/j/4/T/tmp.d938Ip9W80 > time scli run . --native
yes

________________________________________________________
Executed in  398.86 millis    fish           external
   usr time   46.06 millis    0.18 millis   45.89 millis
   sys time   36.60 millis    1.74 millis   34.86 millis

Actually here's a better view:

Benchmark 1: scala-cli run . --native
  Time (mean ± σ):     425.3 ms ±   8.0 ms    [User: 45.1 ms, System: 32.2 ms]
  Range (min … max):   413.3 ms … 438.0 ms    10 runs

Benchmark 2: ./.scala-build/tmp_e396f26dc8/native/main
  Time (mean ± σ):       1.7 ms ±   0.6 ms    [User: 0.6 ms, System: 0.6 ms]
  Range (min … max):     0.7 ms …   4.8 ms    561 runs

  Warning: Command took less than 5 ms to complete. Note that the results might be inaccurate because hyperfine can not calibrate the shell startup time much more precise than this limit. You can try to use the `-N`/`--shell=none` option to disable the shell completely.

Summary
  ./.scala-build/tmp_e396f26dc8/native/main ran
  244.94 ± 85.66 times faster than scala-cli run . --native

[tgodzik]

This might be related to #3158

[Gedochao]

Seems to me this is a duplicate of #3158. Let's track it there.