Skip to content

Query: Always use INNER join to establish the outer query link - #173

Draft
nilmerg wants to merge 5 commits into
fix/proper-reversal-of-relations-in-sub-queries-170from
always-use-inner-joins-for-sub-queries
Draft

Query: Always use INNER join to establish the outer query link#173
nilmerg wants to merge 5 commits into
fix/proper-reversal-of-relations-in-sub-queries-170from
always-use-inner-joins-for-sub-queries

Conversation

@nilmerg

@nilmerg nilmerg commented Aug 26, 2026

Copy link
Copy Markdown
Member

A subquery always needs to establish a link, be it part of a WHERE or solely used to eager load (using Query::derive) a relation. Using an INNER join allows the database to apply optimizations it wouldn't be able to otherwise.

Since this affects Query::derive(), any eager loaded relation will now return no result. But this should be usually not a problem, as a caller that iterates or calls first() without handling the never entered loop or null-return is already bugged, IMHO. I've checked this with Icinga DB Web and found exactly two cases: Icinga/icingadb-web#1408

Still attempting to prove this is actually of advantage, though…

This is required to establish type symmetry as the base's
methods also accept `NULL` to be able to direcly pass a
getters return value to the appropriate setter.
There is now `Relation::bindTo(Model, string, Resolver)`
in order to pass control to relations how they're
prepared. Since the introduction of `BelongsToMany`,
it is established that a relation may resolve to
multiple hops and thus needs to perform steps
n-times rather than a single time. It's this reason
because registering the alias and resolving the filter
is now a responsibility of a relation rather than
the resolver.

Relations know it better how to and the override of
`bindTo` in `BelongsToMany` proves it as it turned
out that it is necessary to allow referencing the
junction table in either the filter or the through
filter in order to be able to better reverse relations.

Qualification must be done by a relation in turn as well,
as otherwise there's a mis-match with what's allowed to
reference and what can be qualified. My initial attempt
was to teach `Relation::resolve()` this, but without
passing it the resolver and changing the return value
this doesn't make sense. Sadly, this is out of the
question as this is a breaking change. Say hello to
`Relation::setFilterSubjects()` due to this.
Changes the way relations can be reversed drastically
as it is now possible to influence the relation to use
during reversal with `::setReverseName(string)` which
allows Icinga DB Web to drop the error-prone `to.from`
and `from.to` relations. An additional change is that
it is now not mandatory anymore to define relations
that are solely being required because of sub-queries.
Missing relations on the reversed path are automatically
registered. For this, each relation type now has its
specific counterpart which is possible to override
with `::setReverseClass(class-string)`. The default
however, is to use the same type which is the case
for `BelongsToOne` and `BelongsToMany`. For `BelongsTo`
a sane override has been chosen that is based on how
it's used at the moment in our products, as `HasOne`
and `HasMany` may both be appropriate. But the latter
clearly is used more often.
Since `::reverse()` uses the source's table alias by default
as reverse name, a deprecation notice is triggered if the
original forward path uses a different name, indicating that
it is necessary to use this name as explicit reverse name.

fixes #170
A subquery always needs to establish a link, be it part of a
WHERE or solely used to eager load (using `Query::derive`)
a relation. Using an INNER join allows the database to apply
optimizations it wouldn't be able to otherwise.
@nilmerg nilmerg added this to the v1.0.0 milestone Aug 26, 2026
@nilmerg nilmerg self-assigned this Aug 26, 2026
@cla-bot cla-bot Bot added the cla/signed label Aug 26, 2026
@nilmerg
nilmerg force-pushed the fix/proper-reversal-of-relations-in-sub-queries-170 branch from 177dcc4 to bcd7fd8 Compare September 2, 2026 08:52
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant