On Jun 12, 2012, at 7:23 PM, David Holmes wrote: > On 13/06/2012 1:06 AM, Kelly Ohair wrote: >> if size is zero it returns null >> thats the cascading error that parfait is finding >> asserts turn into nothing with product builds so i need a dead stop here to >> shut up parfait :( > > Seems rather strange to me. Elsewhere where the assert is used a zero size > would cause us to index outside of allocated memory. Why isn't parfait > picking that up?
The info pointer is a field in the table element, when info_size==0, the info pointer is null anyway. So I think all this if test was doing was short circuiting returning a NULL info pointer, which would mean it had no extra info for the element. > > Not your fault but this seems very ad-hoc. Yeah, very ad-hoc. -kto > > David > >> >> Sent from my iPhone >> >> On Jun 11, 2012, at 22:56, David Holmes<david.hol...@oracle.com> wrote: >> >>> On 12/06/2012 12:48 PM, Kelly O'Hair wrote: >>>> >>>> Need reviewer. >>>> >>>> I was asked to look at some Parfait errors in hprof code: >>>> >>>> 7176138: Fixes for missing close() calls and possible null pointer >>>> reference instead of fatal error >>>> http://cr.openjdk.java.net/~ohair/openjdk8/parfait_hprof_fixes/webrev/ >>> >>> >>> hprof_table.c >>> >>> 211 if ( ltable->info_size == 0 ) { >>> 212 HPROF_ERROR(JNI_TRUE, "Table is empty and should never be."); >>> 213 return NULL; >>> >>> It is not obvious this should be a fatal error. Elsewhere this is handled >>> as an assert. >>> >>> David
