[ 
https://issues.apache.org/jira/browse/GROOVY-12709?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18139320#comment-18139320
 ] 

ASF GitHub Bot commented on GROOVY-12709:
-----------------------------------------

paulk-asert opened a new pull request, #3239:
URL: https://github.com/apache/groovy/pull/3239

   …ch other
   
   CheckingVisitor.findConstExp followed field initializers with no record of 
the fields already visited, so fields whose initializers formed a cycle (static 
String A = B; static String B = A) made RegexChecker and FormatStringChecker 
overflow the stack. Track the variables followed during a lookup and treat a 
repeat as having no constant value.




> Type checkers overflow the stack on fields whose initializers refer to each 
> other
> ---------------------------------------------------------------------------------
>
>                 Key: GROOVY-12709
>                 URL: https://issues.apache.org/jira/browse/GROOVY-12709
>             Project: Groovy
>          Issue Type: Bug
>    Affects Versions: 5.1.3, 6.0.0
>            Reporter: Paul King
>            Priority: Major
>
> h2. Problem
> With {{groovy.typecheckers.RegexChecker}} or 
> {{groovy.typecheckers.FormatStringChecker}} enabled, compiling code that uses 
> a field whose initializer refers, directly or through other fields, back to 
> itself makes the compiler throw {{java.lang.StackOverflowError}}. The error 
> isn't turned into a compilation error, so the compile is aborted.
> Reproducer, run with {{groovy --configscript config.groovy cycle.groovy}}:
> {code:groovy|title=config.groovy}
> import groovy.transform.TypeChecked
> withConfig(configuration) {
>     ast(TypeChecked, extensions: 'groovy.typecheckers.RegexChecker')
> }
> {code}
> {code:groovy|title=cycle.groovy}
> import java.util.regex.Pattern
> class Cycle {
>     static final String R = S
>     static final String S = R
>     static Pattern compile() { Pattern.compile(R) }
> }
> println 'compiled'
> {code}
> Actual:
> {noformat}
> Caught: java.lang.StackOverflowError
>         at 
> org.apache.groovy.typecheckers.CheckingVisitor.findTargetVariable(CheckingVisitor.groovy)
>         at 
> org.apache.groovy.typecheckers.CheckingVisitor.findConstExp(CheckingVisitor.groovy)
>         at 
> org.apache.groovy.typecheckers.CheckingVisitor.findConstExp(CheckingVisitor.groovy)
>         ...
> {noformat}
> Expected: the script compiles and prints {{compiled}}, as it does without the 
> extension. The fields have no constant value to find, so there is nothing to 
> check (both fields are {{null}} at runtime).
> The same happens with {{FormatStringChecker}}, for example with 
> {{String.format(F, 1)}} where {{static final String F = G}} and {{static 
> final String G = F}}.
> ||Version||RegexChecker||FormatStringChecker||
> |4.0.32|StackOverflowError|(checker not available)|
> |5.1.3|StackOverflowError|StackOverflowError|
> |6.0.0|StackOverflowError|StackOverflowError|
> h2. Cause
> {{CheckingVisitor.findConstExp}} finds the constant value of an expression, 
> and for a variable that resolves to a field with an initializer it calls 
> itself on that initializer. Nothing records which fields it has already 
> followed, so a chain of initializers that leads back to an earlier field 
> recurses until the stack overflows.
> {{findConstExp}} is shared by RegexChecker and FormatStringChecker; the other 
> checkers in groovy-typecheckers don't call it.
> h2. Proposed fix
> Track the variables already followed during one lookup, and return {{null}} 
> (no constant value) when a variable is reached a second time. This matches 
> how {{findConstExp}} already treats any other value it can't determine.
> h2. Tests
> * The reproducer above, for both RegexChecker and FormatStringChecker: 
> compiles, and runs.
> * A longer cycle ({{A = B}}, {{B = C}}, {{C = A}}).
> * A field initialized from another field that does have a constant value 
> ({{static final String R = S}}, {{static final String S = /(a)/}}) is still 
> followed and checked, for example a bad group number is still reported.



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

Reply via email to