Given there's already a pg_hint_plan extension I really don't see why people are so angry about the developers not wanting to bless something they consider a footgun as a core feature.
"risk", because pg_hint is an external module, written by somebody as a hobby, therefore if it stops existing or working after some updates or with newer versions of Postgres then it becomes a huge problem for PROD systems that need it for any reason.
My point is that "core developers should add an unproven feature they think is a footgun and take on the support overhead of that without convincing evidence it would actually be a net win" is not a convincing argument.
Also given Amazon AWS support it for Aurora, I feel like it's not -that- hobby-ish.
If people use it and can provide evidence that it helps more than it hurts, that might be convincing.
Insulting the core developers for not yet being convinced seems rather less likely to help.
> Also given Amazon AWS support it for Aurora, I feel like it's not -that- hobby-ish.
Oh, didn't know that :)
About the rest: sure, I can agree as well about the indirect adverse effect of having them available (e.g. easy to misuse them as I saw in some apps using Oracle DBs), and it's for sure wrong to insult somebody because of this.
Still, personally, I think that the pros would outweight the cons of having that embedded in the app.
On one hand I remember some nights spent in the past trying to make some SQL work, hints were always at least a good temporary workaround.
On the other hand there will always be some SQL which confuse the optimizer (or more special cases about a lot of data changing distribution of values, etc..) and hints would be the only way to cover these cases.
Maybe an interesting question is on which level should hints act? I know mainly only Oracle & MariaDB, therefore I know hints of the type "use that index"/"query tables in this order"/"join these tables with this join type"/etc..., which are probably low-level hints. Maybe already just higher-level hints of the type "most important selectivity criteria comes from inline-view X"/"I want just the first row of the result"/"take into account only plans which select data by using indexes"/etc... would be as well interesting, not sure, just dumping here my thoughts.
