[
https://issues.apache.org/jira/browse/SOLR-13658?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16896139#comment-16896139
]
Martin Grigorov commented on SOLR-13658:
----------------------------------------
Hi [~erickerickson],
You didn't provide full snippets so I might have done something in a different
way but the following code compiles and runs fine:
{code:java}
import java.util.Map;
public class Solr13658 {
public static void main(String[] args) {
// with var
var kidding = getComplex();
var blivet = kidding.get("one");
var blah = kidding.get("two").get(2);
blah = 55;
System.err.println("blah: " + blah);
// without var
Map<String, Map<String, Integer>> kidding1 = getComplex();
Map<String, Integer> blivet1 = kidding1.get("one");
Integer blah1 = kidding1.get("two").get(2);
blah1 = 55;
System.err.println("blah1: " + blah1);
}
private static Map<String, Map<String, Integer>> getComplex() {
return Map.of(
"one", Map.of("1", 1, "11", 11, "111", 111),
"two", Map.of("2", 2, "22", 22, "222", 222)
);
}
}
{code}
and it prints :
{code:java}
blah: 55
blah1: 55
{code}
in both cases _blah_s types are java.lang.Integer
> Discuss adding the new "var" construct to the forbidden API list.
> -----------------------------------------------------------------
>
> Key: SOLR-13658
> URL: https://issues.apache.org/jira/browse/SOLR-13658
> Project: Solr
> Issue Type: Wish
> Security Level: Public(Default Security Level. Issues are Public)
> Affects Versions: master (9.0)
> Reporter: Erick Erickson
> Assignee: Erick Erickson
> Priority: Major
>
> Personally, I'm strongly against allowing the "var" construct in Lucene/Solr
> code. I think it's a wonderful opportunity to introduce bugs that won't be
> found until runtime as well as making maintainence significantly harder. I
> don't even think for a project like Solr it would save any time overall...
> So let's discuss this ahead of time and see if we can reach a consensus. I'll
> start the discussion off:
> My baseline argument is that for a large complex project, especially ones
> with many different people coding, I want the compiler to give me all the
> help possible. And the "var" construct takes away some of that help.
> I’ve seen this argument go around at least 4 times in my career. The argument
> that “it takes longer to write if you have to type all this stuff” is bogus.
> Last I knew, 80% of the time spent is in maintaining/reading it. So the
> argument “I can write faster” means I can save some fraction of the 20% of
> the time writing the original code but spend many times that figuring out
> what the code is actually doing the other 80% of the time.
> The IDE makes _writing_ this slightly faster, admittedly.
> {code:java}
> Whatever what = new Whatever();
> var kidding = what.getComplex();
> var blivet = kidding.get("stuff");
> {code}
> But once that’s done, if I’m reading the code again I don't have any clue what
> {code:java}
> kidding or blivet
> {code}
> are. Here's the signature for getComplex:
> {code:java}
> Map<String, Map<Integer, Integer>> getComplex()
> {code}
> I have to go over to the definition (which I admit is easier than it used to
> be in the bad old days, but still) to find out.
> HERE'S THE PART I REALLY OBJECT TO!
> The above I could probably live with, maybe we could get the InteliJ
> developers and see if they can make hover show the inference. What I will
> kick and scream about is introducing bugs that are not found until runtime.
> Even this obvious stupidity fails with a ClassCastException:
> {code:java}
> var corny = new TreeMap<String, String>();
> corny.put("one", "two");
> corny.get(1);
> {code}
> But it's much worse when using classes from somewhere else. For instance,
> change the underlying class in the first example to return
> {code:java}
> Map<String, Map<String, Integer>>{code}
> .
> This code that used to work now throws an error, _but it compiles_.
> {code:java}
> var kidding = what.getComplex();
> var blivet = kidding.get("stuff");
> var blah = kidding.get("stuff").get(1); // generates ClassCastException:
> class java.lang.String cannot be cast to class java.lang.Integer
> {code}
> So in order to save some time writing (that I claim will be lost multiple
> times over when maintaining the code) we'll introduce run-time errors that
> will take a bunch _more_ time to figure out, and won’t be found during unit
> tests unless and until we have complete code coverage.
> If there's a way to insure that this kind of thing can't get into the code
> and we implement that, I could be persuaded, but let's make that an explicit
> requirement (and find a suitable task for the build system, precommit or
> whatever).
> The floor is open...
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
