Paul King created GROOVY-12706:
----------------------------------
Summary: RegexChecker fails inside the compiler on script binding
variables
Key: GROOVY-12706
URL: https://issues.apache.org/jira/browse/GROOVY-12706
Project: Groovy
Issue Type: Bug
Components: groovy-typecheckers
Affects Versions: 6.0.0, 5.1.3
Reporter: Paul King
Assignee: Paul King
h2. Problem
With {{groovy.typecheckers.RegexChecker}} enabled, a script that uses a binding
variable (one assigned without {{def}}) in a regex subscript or a
{{findGroups}} call makes the compiler fail with a general error, instead of
reporting the normal type-checking error for an undeclared variable.
Reproducer, run with {{groovy --configscript config.groovy subscript.groovy}}:
{code:groovy|title=config.groovy}
import groovy.transform.TypeChecked
withConfig(configuration) {
ast(TypeChecked, extensions: 'groovy.typecheckers.RegexChecker')
}
{code}
{code:groovy|title=subscript.groovy}
finder = 'groovy' =~ /(oo)/
finder[0]
{code}
Actual:
{noformat}
General error during canonicalization: No signature of method: getNodeMetaData
for class: org.codehaus.groovy.ast.DynamicVariable is applicable for argument
types: (String) values: [RegexChecker_INFERRED_GROUP_COUNT]
{noformat}
Expected: the same result as without the extension:
{noformat}
[Static type checking] - The variable [finder] is undeclared.
{noformat}
||Script||4.0.32||5.0.6||5.1.3||6.0.0||master||
|{{finder = 'groovy' =~ /(oo)/; finder[0]}}|general error|general error|general
error|general error|general error|
|{{p = ~/(oo)/; 'groovy'.findGroups(p)}}|undeclared (correct)|undeclared
(correct)|undeclared (correct)|general error|general error|
|{{p = ~/(oo)/; p.matcher('groovy')}}|undeclared (correct)|undeclared
(correct)|undeclared (correct)|undeclared (correct)|undeclared (correct)|
|{{m = 'groovy' =~ /(oo)/; m.group(1)}}|undeclared (correct)|undeclared
(correct)|undeclared (correct)|undeclared (correct)|undeclared (correct)|
h2. Cause
For a binding variable, {{CheckingVisitor.findTargetVariable}} returns an
{{org.codehaus.groovy.ast.DynamicVariable}}, which is not an {{ASTNode}} and
has no node metadata. RegexChecker reads its group-count metadata from whatever
that lookup returns ({{var?.getNodeMetaData(...)}} /
{{source?.getNodeMetaData(...)}}), so the dynamic {{getNodeMetaData}} call
fails:
* the {{matcher[...]}} subscript inference in {{beforeVisitMethod}} (present
since RegexChecker was added);
* the {{findGroups}} fallback that takes the group count from a pattern
variable (6.0.0 onwards);
* the {{Pattern.matcher(...)}} propagation and the {{Matcher.group(...)}}
check. These are only reached when the variable's type is {{Pattern}} or
{{Matcher}}, so they don't fail today. They would once binding variables carry
checked types (GROOVY-12462).
The other checkers in groovy-typecheckers that look up variables (NullChecker,
PurityChecker, ModifiesChecker, SqlInjectionChecker) already handle a
{{DynamicVariable}} safely.
h2. Proposed fix
Treat a lookup result that is not an AST node as an untracked variable: skip
the regex metadata reads and writes, and leave the variable to normal type
checking. Cover all four lookup sites, including the {{copyRegexMetaData}}
helper added by GROOVY-12697.
h2. Tests
* Each row of the table above reports only the undeclared-variable error, with
no general error.
* A regex declared with {{def}} in the same script is still checked ({{def m =
'groovy' =~ /(oo)/; m[0][2]}} reports "Invalid group count 2 for regex with 1
group").
Static compilation of scripts that use binding variables is uncommon today, but
GROOVY-12462 (typed binding for REPL hosts) is intended to make it routine, so
this should be fixed before that work.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
