Don't infer package name from lex rule names.

[regisd]

Fix #104

Also, make ClassInfo immutable.

[lsf37]

This will still fail if the string package is mentioned on the same line as a semicolon in later sections of the lexer spec.

Test case:

  @Test
  public void guessPackageAndClass_givenClass_package_in_rule() throws Exception {
    String lex =
        "\n"
            + "%%\n"
            + "\n"
            + "%public\n"
            + "%class Foo\n"
            + "\n"
            + "%apiprivate\n"
            + "%int\n"
            + "\n"
            + "%%\n"
            + "\n"
            + "package X  { /* no action; */ }"
            + "\n"
            + "[^]  { /* no action */ }\n";
    assertThat(guessPackageAndClass(lex)).isEqualTo(new ClassInfo("Foo", null));
  }

The code should probably become more aware in which section it currently is, and scan for package name only up until the first %% and for class name only after the first %% until the second %%.

[regisd]

This will still fail if the string package is mentioned on the same line as a semicolon in later sections of the lexer spec.

Test case:

        + "\n"
        + "package X  { /* no action; */ }"

The code should probably become more aware in which section it currently is, and scan for package name only up until the first %% and for class name only after the first %% until the second %%.

You are absolutely right. There are 2 reasons why I dont' want to go for a proper fix:

• This only happens if the package is not defined in the Java user code ; which is never the case for production code ; i.e. we are only fixing examples.
• Changes may very well be confllcting with my refactoring of Out and Options.

In the end, there are only three robusts solutions:

• Only give the output directory ; but this breaks a lot of abstraction. I think the Emitter shuold only have an OutputStream (which is convenient for tests)
• Let the lexer do two passes.
• Output the java code in a temp file, and move it at the end of the generation, once the emitter can tell exactly what package it is

[regisd]

I'll submit this for now.

[lsf37]

Yes, that's alright. I'll make a low-priority issue so I don't forget coming back to this once you've merged the Out/Option refactor. (I think I can fix this properly reasonably easily).