RFR: 8345263: Make sure that lint categories are used correctly when logging lint warnings

[Maurizio Cimadamore]

On Wed, 4 Dec 2024 17:11:26 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

> This PR tightens up the logic by which javac reports lint warnings. Currently, lint warnings are typically emitted using the following idiom:
> 
> 
> if (lint.isEnabled(LintCategory.DIVZERO) {
>     log.warning(LintCategory.DIVZERO, pos, Warnings.DIVZERO);
> }
> 
> There are some issues with this approach:
> 
> * logging a lint warning has to be preceded by the correct `isEnabled` check
> * the check and the `log::warning` call must share the same `LintCategory`
> * the selected warning key in the `Warnings` class must also make sense for the selected `LintCategory`
> 
> This PR addresses these issues, so that the above code is now written as follows:
> 
> 
> lint.logIfEnabled(pos, LintWarnings.DIVZERO);
> 
> 
> The new idiom builds on a number of small improvements:
> 
> * the lint category is now tracked directly in the `compiler.properties` file;
> * a new `LintWarning` class is added to `JCDiagnostic` to model a warning key that is also associated with a speicfic `LintCategory` enum constant;
> * the `CompilerProperties` class has a new group of compiler keys, nested in the new `LintWarnings` class. This class defines the `LintWarning` objects for all the warning keys in `compiler.properties` that have a lint category set
> * A new method `Lint::logIfEnabled(Position, LintWarning)` is added - which simplifies the logging of lint warnings in many common cases, by merging the `isEnabled` check together with the logging.
> 
> As bonus points, the signatures of some methods in `Check` and `MandatoryWarningHandler` have been tightened to accept not just a `Warning`, but a `LintWarning`. This makes it impossible, for instance, to use `Check::warnUnchecked` with a warning that is not a true lint warning.
> 
> Many thanks @archiecobbs for the useful discussions!

src/jdk.compiler/share/classes/com/sun/tools/javac/code/Lint.java line 398:

> 396:      */
> 397:     public void logIfEnabled(DiagnosticPosition pos, LintWarning warning) {
> 398:         if (isEnabled(warning.getLintCategory())) {

I'm not 100% sure about this method. On the one hand it makes clients simpler - but it does require `Lint` keeping track of a `log`. An alternative could be to add a method in `AbstractLog`, e.g.:

warningIfEnabled(DiagnosticPosition pos, Lint lint, LintWarning key)

src/jdk.compiler/share/classes/com/sun/tools/javac/comp/Attr.java line 4472:

> 4470:             // a warning. Make allowance for the class of an array type e.g. Object[].class)
> 4471:             if (!sym.owner.isAnonymous()) {
> 4472:                 chk.lint.logIfEnabled(tree, LintWarnings.StaticNotQualifiedByType(sym.kind.kindName(), sym.owner));

This is an odd one - the rest of `Attr` uses `env.info.lint` but if I do that here I get some NPE in some unrelated JVM test. I wonder how robust using two different lints is... but for now I'll keep compatibility with what we have.

src/jdk.compiler/share/classes/com/sun/tools/javac/comp/Check.java line 4664:

> 4662:         if ((rd.module.flags() & Flags.AUTOMATIC_MODULE) != 0) {
> 4663:             deferredLintHandler.report(_ -> {
> 4664:                 if (rd.isTransitive() && lint.isEnabled(LintCategory.REQUIRES_TRANSITIVE_AUTOMATIC)) {

This one is a bit odd - the warning can be reported under two categories, so the code makes sure we don't report twice. This is deliberate I've checked with @lahodaj

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

PR Review Comment: https://git.openjdk.org/jdk/pull/22553#discussion_r1870145545
PR Review Comment: https://git.openjdk.org/jdk/pull/22553#discussion_r1870139558
PR Review Comment: https://git.openjdk.org/jdk/pull/22553#discussion_r1870141200