Remove stream usage from Symtab.lookupPackage()

[Jonathan Gibbons]

In times past, there were places in javac when `instanceof` was *not* a 
substitute for checking the tag.

We've probably gotten rid of those cases by now, but as a general coding 
pattern, I note that for end users writing code like annotation 
processors, the problem still remains if using other compilers.

-- Jon

On 2/21/23 7:30 PM, Archie Cobbs wrote:
> On Tue, Feb 21, 2023 at 6:24 PM christoph.dreis at freenet.de 
> <christoph.dreis at freenet.de> wrote:
>
>     I'm currently profiling some compilation phases of internal
>     projects and noticed in allocation profiles that
>     |Symtab.lookupPackage| takes up ~2% overall. The majority of this
>     is spent in |.stream().anyMatch()| usages to find out if the given
>     module symbol depends on the unnamed module.
>
>
> Straightforward patch, but it's disappointing that unrolling a Stream 
> into the equivalent for loop makes it noticeably faster (how much?).
>
> This also brings up larger philosophical questions about code style 
> vs. performance. If using a Stream somewhere makes for more 
> clean/elegant/understandable code, but is slower, how to weigh the 
> trade-off? Maybe this has been discussed before on this list before I 
> joined. Maybe we should assign this bug to the hotspot group :)
>
> On this same topic, this pattern which appears constantly in the 
> compiler always makes me kind of quizzical:
>
> JCExpression expr = TreeInfo.skipParens(tree.expr);
> if (expr.hasTag(APPLY)) {
>     JCMethodInvocation apply = (JCMethodInvocation)expr;
>     ...
> }
>
> Is instanceof really that much slower? Maybe it was once, long ago, 
> and this is a legacy thing?
>
> -Archie
>
> -- 
> Archie L. Cobbs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/compiler-dev/attachments/20230222/7b53cd6c/attachment-0001.htm>