[
https://issues.apache.org/jira/browse/GROOVY-10535?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17531849#comment-17531849
]
Jochen Theodorou commented on GROOVY-10535:
-------------------------------------------
[~emilles]The line
https://github.com/apache/groovy/blob/6b6a3e8a6fec2826f13cc0e6a52dcf1f25b4d74b/src/main/java/org/codehaus/groovy/vmplugin/v8/Selector.java#L930
should add the test for the null case, 945 then builds it together. It really
is just a null check there and if it succeeds it will continue with "handle",
if not it will continue with "fallback". fallback is the general/slow path,
that should also reset the callsite. So what if it was not null and becomes
null. SAME_CLASS simply calls
https://github.com/apache/groovy/blob/6b6a3e8a6fec2826f13cc0e6a52dcf1f25b4d74b/src/main/java/org/codehaus/groovy/vmplugin/v8/IndyGuardsFiltersAndSignatures.java#L196
{code:java}
if (o == null) return false;
return o.getClass() == c;
{code}
You then have the null check here instead of a special null-check-methodhandle.
Of course this could be expressed with more method handles (maybe not the ==,
but I am not sure). So for example the code in theory could be simplified by
using the null-guard always and have the class equality check on it own.
Maybe in case it is not clear.... image a guardWithTest as a composition of 3
handles, on condition handle, one true-case handle and one handle for the
false-case. All of them operate on the given arguments. condition must fit the
input arguments and return boolean, true and false cases must have a signature
fitting to required input and output. You can very well have the true/false
cases as guards as well, allowing you to build a tree structure - a bit like a
decision tree. For Groovy it is more like a list, where the false-case is the
slow path and the true-case is whatever we want to test further or execute.
> IF condition on empty Collection has different behavior than null Collection
> ----------------------------------------------------------------------------
>
> Key: GROOVY-10535
> URL: https://issues.apache.org/jira/browse/GROOVY-10535
> Project: Groovy
> Issue Type: Bug
> Affects Versions: 3.0.10
> Environment: Groovy 3.0.10/OpenJDK 17.0.2 - Ubuntu 21.10
> Reporter: Rodolfo Yanke
> Assignee: Eric Milles
> Priority: Major
> Fix For: 5.0.0-alpha-1, 3.0.11, 4.0.3
>
>
> I believe this code should print "something" but doesn't work:
> {code:java}
> @CompileStatic
> class NotWorkingExample {
> static void main(String[] args) {
> Collection<String> values = null
> for (i in 0..<200_000) {
> printSomethingIfNotEmpty(values)
> }
> //never printed but it should
> values = ['A']
> printSomethingIfNotEmpty(values)
> }
> static printSomethingIfNotEmpty(Collection<String> values) {
> if(values) {
> println 'something'
> }
> }
> }
> {code}
> This one does print "something" because we pass an empty collection []
> instead of null:
> {code:java}
> @CompileStatic
> class ItWorks {
> static void main(String[] args) {
> Collection<String> values = []
> for (i in 0..<200_000) {
> printSomethingIfNotEmpty(values)
> }
> //it works because [] was passed in the previous 200k calls
> values = ['A']
> printSomethingIfNotEmpty(values)
> }
> static printSomethingIfNotEmpty(Collection<String> values) {
> if(values) {
> println 'something'
> }
> }
> }{code}
> Some optimization is done differently when the condition is skipped too many
> times. Both classes should output "something" on the last method call.
> Thank you for the support.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
