Dict entries exclude the 'next' pointer when NULL #11746
Conversation
Whenever a 'next' pointer is set or unset, the dict entry may need to be reallocated. When this happens, the pointer to it needs to be updated and the entry metadata callback needs to be called.
This commit introduces a dictEntry with key and value, but no next field. The pointer to the dictEntry is tagged in the LSB bits.
zuiderkwast
force-pushed
the
dict-no-next
branch
from
c160ce9
to
a60a8a2
Compare
| /* plink points to the bucket. There's no prev entry. */ | ||
| *plink = dictGetNext(he); |
There was a problem hiding this comment.
is that comment right? it could be just a plain dictEntry (with no flags).
There was a problem hiding this comment.
This code doesn't care what kind of dictEntry it is. It matches how plink is created in dictTwoPhaseUnlinkFind. he and *plink can be any kind of dictEntry.
Thanks for looking at this. Any feedback can definitely help. I have split the commit into one refactoring commit (that doesn't change any behaviour) and a second commit that adds dictEntryNoNext. With only the first commit, all tests pass. I trying to narrow down the bug in this way. |
|
|
This commit introduces a dictEntry with key and value, but no next field, which is used to save memory when there is no next. The pointer to the dictEntry is tagged in the LSB bits.
Whenever a 'next' pointer is set or unset, the dict entry may need to be reallocated. When this happens, the pointer to it needs to be updated and the entry metadata callback needs to be called.
There's some crash that I haven't figured out. Help debugging is appreciated.