Re: dshash_find_or_insert vs. OOM

On Tue, Mar 17, 2026 at 9:34 PM Chao Li <li(dot)evan(dot)chao(at)gmail(dot)com> wrote:
> When OOM happens, Assert((flags & DSHASH_INSERT_NO_OOM) != 0); makes sense. But for resize(), the assert is inside resize(), while for insert_into_bucket(), the assert is in the caller. That feels a bit inconsistent to me, and I think it hurts readability a little. A reader might wonder why there is no corresponding assert after resize() unless they go read the function body.

Adjusted.

> Making this a nested block does have the benefit of keeping dsa_flags close to where it is used. But from my impression, this style is still fairly uncommon in the codebase. I worry it may implicitly signal to other hackers that this is an acceptable pattern. So unless we intentionally want to encourage that style, I would lean toward avoiding it here.

Yeah, that was dumb. Fixed.

Thanks for the review; here's v2.

--
Robert Haas
EDB: http://www.enterprisedb.com