[ 
https://issues.apache.org/jira/browse/GROOVY-12762?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18144171#comment-18144171
 ] 

ASF GitHub Bot commented on GROOVY-12762:
-----------------------------------------

testlens-app[bot] commented on PR #3296:
URL: https://github.com/apache/groovy/pull/3296#issuecomment-6843215034

   ## ✅ All tests passed ✅
   
   🏷️ Commit: 4113f621a154d0104955af4a7c97f519ded0d90e
   ▶️ Tests:  131773 executed
   ⚪️ Checks: 26/26 completed
   
   ---
   _Learn more about TestLens at 
[testlens.app/docs](https://testlens.app/docs/features/pr-comment/)._
   




> @NamedVariant: default values aren't handled correctly by the named-arg 
> variant
> -------------------------------------------------------------------------------
>
>                 Key: GROOVY-12762
>                 URL: https://issues.apache.org/jira/browse/GROOVY-12762
>             Project: Groovy
>          Issue Type: Bug
>    Affects Versions: 6.0.0
>            Reporter: Paul King
>            Assignee: Paul King
>            Priority: Major
>
> The named-arg variant generated by {{@NamedVariant}} mishandles default 
> parameter values in several ways. All examples below use:
> {code:groovy}
> import groovy.transform.*
> @ToString(includeNames=true)
> class Color { int r, g, b }
> {code}
> h3. 1. A positional parameter's default is never used
> When named arguments ({{@NamedParam}}/{{@NamedDelegate}}) are mixed with a 
> positional parameter that has a default value, the named-arg variant has no 
> default-argument variants, so the default can never be used by leaving the 
> argument out:
> {code:groovy}
> @NamedVariant
> String m(@NamedDelegate Color color, Number alpha = 4.5) { "$color $alpha" }
> assert m(r:1, g:2, b:3, 0.0) == 'Color(r:1, g:2, b:3) 0.0'
> assert m(r:1, g:2, b:3) == 'Color(r:1, g:2, b:3) 4.5'  // 
> MissingMethodException: no m(Map)
> {code}
> Only {{m(Map, Number)}} is generated. The named-arg variant reuses the 
> original method's {{Parameter}} objects, and the Verifier clears each default 
> once it has expanded the original method, so nothing is left for the variant. 
> This is the {{// TODO account for default params giving multiple signatures}} 
> in {{NamedVariantASTTransformation.createMapVariant}}. It was raised in 
> review of GROOVY-10498 (PR #3295), where the suggested assertion {{m(r:1, 
> g:2, b:3)}} couldn't be added for this reason.
> h3. 2. A default that references a {{@NamedDelegate}} fails at runtime
> {code:groovy}
> @NamedVariant
> String s(@NamedDelegate Color c, @NamedParam String z = c.toString()) { z }
> assert s(r:1, g:2, b:3) == 'Color(r:1, g:2, b:3)'  // 
> MissingPropertyException: No such property: c
> {code}
> This also happens under {{@CompileStatic}}. The default is copied into the 
> named-arg variant unchanged, where {{c}} doesn't exist.
> h3. 3. A nested reference to an earlier parameter in a default isn't rewritten
> {code:groovy}
> @NamedVariant
> String n(@NamedParam int x, @NamedParam int y = x * 2 + 1) { "$x $y" }
> assert n(x: 5) == '5 11'  // MissingPropertyException: No such property: x
> {code}
> {{getDefaultValue}} rewrites references to earlier parameters (GROOVY-10561, 
> GROOVY-10889, GROOVY-11325) only one level deep: {{x + 1}} works, {{x * 2 + 
> 1}} doesn't.
> h3. 4. A closure in a default that references an earlier parameter crashes 
> the compiler
> {code:groovy}
> @NamedVariant
> String s(@NamedParam int x, @NamedParam Closure z = { "v$x" }) { z() }
> s(x: 1)
> {code}
> {noformat}
> BUG! exception in phase 'class generation' ... tried to get a variable with 
> the name x as stack variable, but a variable with this name was not created
> {noformat}
> The same happens for {{@NamedParam String z = { "v$x" }()}} and for 
> references to a {{@NamedDelegate}}.
> h3. 5. A {{@NamedVariant}} constructor with {{@NamedDelegate}} fails to 
> compile without {{@CompileStatic}}
> {code:groovy}
> class K {
>     String s
>     @NamedVariant
>     K(@NamedDelegate Color c, @NamedParam String z = 'q') { s = "$c $z" }
> }
> new K(r:1, g:2, b:3)
> {code}
> {noformat}
> Cannot reference 'subMap' before supertype constructor has been called.
> {noformat}
> The {{subMap}} call that builds the delegate is created with 
> {{implicitThis=true}}, so the Verifier treats it as an instance call before 
> {{this(...)}}. The neighbouring {{containsKey}} helper already does 
> {{setImplicitThis(false)}} for this reason. It works under {{@CompileStatic}}.
> h3. Proposed fix
> # *Default-argument variants for the named-arg variant (part 1).* Give the 
> named-arg variant its own copies of positional parameters, keeping their 
> default values, so the Verifier generates {{m(Map)}} etc.
> # *Clashes are a compile-time error.* If a default-argument variant of the 
> named-arg method/constructor duplicates a declared member, report an error, 
> as {{@TupleConstructor(namedVariant=true)}} does (GROOVY-11644), e.g.
> {noformat}Class W already has method m(java.util.Map) which duplicates a 
> default-argument variant of the named-arg method{noformat}
> Code that declares its own {{m(Map)}} alongside such a {{@NamedVariant}} 
> compiles today, so this is a deliberate behaviour change.
> # *A {{@NamedDelegate}} referenced by a default is built once and shared 
> (part 2).*
> #* Methods: bind the delegate to a local in the named-arg variant and use it 
> for both the argument and the default(s) that reference it.
> #* Constructors, where no statements are allowed before {{this(...)}}: the 
> named-arg constructor chains to a private helper constructor {{(Map, 
> <positional...>, <referenced delegates...>)}}. It is generated only when a 
> default references a delegate, and it is a compile-time error if that 
> signature is already declared.
> In both cases the method receives the same instance the default saw.
> # *A positional default that references a {{@NamedDelegate}} is a 
> compile-time error*, for methods and constructors:
> {noformat}The default value of positional parameter 'd' can't reference 
> @NamedDelegate parameter 'c'.{noformat}
> Such a default is evaluated in a Verifier-generated variant where only the 
> map is in scope, so the shared instance isn't available. For methods this 
> could later be relaxed by having the transform generate those variants 
> itself; for constructors that would need extra private constructors with 
> synthetic marker parameters to avoid signature clashes. {{@NamedParam}} 
> defaults are unaffected.
> # *Rewrite references to earlier parameters at any depth (part 3).*
> # *A default that references an earlier parameter from within a closure is a 
> compile-time error (part 4).* Supporting it would mean copying the closure 
> and redoing variable scoping for the generated member.
> # *Create the delegate's {{subMap}} call with {{setImplicitThis(false)}} 
> (part 5).*
> Parts 3 and 5 are small and independent of the rest, so they could be 
> backported on their own.
> h3. Example after the fix
> {code:groovy}
> @NamedVariant
> String s(@NamedDelegate Color c, @NamedParam int x, @NamedParam int y = x * 2 
> + 1, @NamedParam String z = c.toString()) {
>     "$c $x $y $z"
> }
> assert s(r:1, g:2, b:3, x:5) == 'Color(r:1, g:2, b:3) 5 11 Color(r:1, g:2, 
> b:3)'
> class K {
>     boolean same
>     @NamedVariant
>     K(@NamedDelegate Color c, @NamedParam Color d = c) { same = c.is(d) }
> }
> assert new K(r:1, g:2, b:3).same
> {code}
> h3. Notes
> * With default-argument variants, an explicit {{null}} for a positional 
> parameter is passed through, as for ordinary Groovy default parameters; only 
> leaving the argument out selects the default. This matches GROOVY-10498.
> * A prototype of all of the above passes the existing 
> {{org.codehaus.groovy.transform.*}} tests (1336).



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to