Annotation processor option with space breaks header compiler parsing

[guw]

Description of the bug:

Turbine header compiler fails to compile java_library with annotation processor because annotation processor options from javacopts are not passed to the header compiler.

java_library(
  name = "A",
  plugins = ["//some/plugin"],
  javacopts = [
    "-Asome.annotation.processor.option=vale1,value2, value3",
  ]
)

Bazel fails with:

ERROR: .. Compiling Java headers ... and running annotation processors (MyGenerator, ..., ...) failed: (Exit 1): java failed: error executing command ...

Use --sandbox_debug to see verbose messages from the sandbox and retain the sandbox build root for debugging
<>: error: An exception occurred in MyGenerator:
java.lang.IllegalArgumentException: Property some.annotation.processor.option invalid 
	at MyGenerator.init(...)
	at com.google.turbine.binder.Processing.process(Processing.java:131)
	at com.google.turbine.binder.Binder.bind(Binder.java:107)
	at com.google.turbine.main.Main.bind(Main.java:270)
	at com.google.turbine.main.Main.compile(Main.java:162)
	at com.google.turbine.main.Main.compile(Main.java:133)
	at com.google.turbine.main.Main.main(Main.java:89)

Inspecting the .param files show the javacopt are actually passed to the header compiler. It looks like the header compiler is not passing them to the annotation processor.

cat bazel-out/....-hjar.jar-0.params
...
--javacopts
-Asome.annotation.processor.option=vale1,value2, value3
...

This looks to be related to the space in the annotation processor option.
It's added as a single line in the .params file but the Turbine header compiler seems to parse it incorrectly.

Which category does this issue belong to?

Java Rules

What is the output of bazel info release?

6.3.2

[hvadehra]

Considering that the java rules tokenize javacopts, I would expect

  javacopts = [
    "-Asome.annotation.processor.option=vale1,value2, value3",
  ]

to fail with an error like error: invalid flag: value3

@guw Does this need something more to reproduce?

[guw]

@hvadehra Not that I am aware of. Which tokenization are you referring to? Do you have a link to the code?

[hvadehra]

https://github.com/bazelbuild/bazel/blob/5932bee129b189baa5e4cbec116176b2f15d723e/src/main/java/com/google/devtools/build/lib/rules/java/JavaHelper.java#L98C39-L98C58

[guw]

Ah ... thanks for sharing that code. I omitted a tiny detail: '. Here is how the annotation processor option is actually being populated in our macro.

   javacopts.append("-Asome.annotation.processor.option='%s'" % generator.get("option"))

The correct value is:

  javacopts = [
    "-Asome.annotation.processor.option='vale1,value2, value3'",
  ]

This is handled/supported by the tokenization.

[hvadehra]

Ah, thanks, that makes sense. Looks like JavaBuilder behaves as expected and it's only turbine that is splitting this incorrectly.

Presumably this has been fixed by google/turbine#272, so all we need now is a turbine release and updating that in Bazel.