Seems like a patch to make a pragma that makes autovivification a fatal error would be rather easy to write and would be useful here and for other uses.

Don't trust people who know more than me ;-)

I think the point here is that without knowing Mr. Huckaby personally its difficult to asses his knowledge level. By the look of the module it would appear he is a competent fellow, but that section on Tie's suggests he might not be so well informed about some of perls esoterica. At the very least it would appear that he is discounting comments in MLDBM, something that is IMO unwise as the author, Gurusamy Sarathy, is a former pumpking (responsible for 5.6) which makes his opinion carry considerable weight.

---
$world=~s/war/peace/g

Be careful not to mix apples and oranges here. MLDBM uses one strategy. DBM::Deep uses a different strategy that does indeed work to track changes to deep structures. However, there is a bug with autoviv: it doesn't call the STORE routine of the tied hash when a value is autovivied for a deeper reference.

If you carefully extend your DBM::Deep hash, it works as advertised, and in a way that MLDBM cannot touch, and in a way that requires no binary DBM module. This is a remarkable achievement. The horsing around required by MLDBM referenced in Re^3: DBM::Deep Problem is not required in DBM::Deep.... you just have to not autovivify.

The tie bug is sad. I wish it would get fixed.

-- Randal L. Schwartz, Perl hacker
Be sure to read my standard disclaimer if this is a reply.