RFR (S): 8211394: CHECK_ must be used in the rhs of an assignment statement within a block (original) (raw)

David Holmes david.holmes at oracle.com
Mon Oct 8 21:19:37 UTC 2018

Hi Ioi,

On 9/10/2018 1:55 AM, Ioi Lam wrote:

I am not sure if this change is really required:

Symbol* MethodFamily::generatenodefaultsmessage(TRAPS) const { - return SymbolTable::newsymbol("No qualifying defaults found", THREAD); + Symbol* s = SymbolTable::newsymbol("No qualifying defaults found", CHECKNULL); + return s; I think the rationale is "since SymbolTable is a different class, we don't know what value it returns when an exception is thrown."

Yes. Seems important to try and establish some consistency on how to handle this.

However, the return value is a pointer type. Can we except anything other than NULL in case of an exception? I think using the CHECKNULL here will make the code bigger and slower, and also harder to read.

We shouldn't expect anything other than an "error return value" if an exception is pending, but that relies on who/how the target code was coded.

I don't think the code change will be at all noticeable in terms of size of speed. Readability is subjective.

Thanks, David

(BTW, we have at 150+ occurrencesof 'return.*THREAD[)];' in the code.)

Thanks - Ioi

On 10/7/18 3:08 PM, David Holmes wrote: Bug: https://bugs.openjdk.java.net/browse/JDK-8211394 webrev: http://cr.openjdk.java.net/~dholmes/8211394/webrev/

If a CHECK macro is used on a function call that is part of a return statement i.e. return foo(CHECKNULL); then it expands into an unreachable if-statement that checks the exception state: return foo(); if (EXCEPTIONOCCURRED) return NULL; This is obviously a programming error, but unfortunately not something our often pedantic compilers complain about. There are two ways to fix: 1. Convert to assignment: T* t = foo(CHECKNULL); return t; 2. If the method is local and its exception behaviour easily discernible and matches the expected behaviour, then change CHECK to THREAD return foo(THREAD); Both fixes are applied as appropriate. As per the bug report I also revisited an earlier fix in this area - JDK-8062808 - and made adjustments. Thanks, David

More information about the hotspot-runtime-dev mailing list