Repository navigation
fix(postgres): tell a self-join's alias.* apart by its alias - #92
Merged
Merged
Conversation
8000
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Contributor
There was a problem hiding this comment.
ℹ️ No critical issues. One edge case is inline: the check that each target is one column is gone.
Reviewed changes
I reviewed the whole diff. It teaches aliased_columns to expand a top-level alias.* over a table reading, so that each half of a self-join keeps its own nullability.
alias.*expansion: a star over an alias whose FROM item is a table now addswidth(table)entries of(table, nulled). A bare*, a star over a subselect, function or WITH item, and a table whose width is unknown or ambiguous still give nothing.width: this is the column count of the oneself.relationsentry with that name.RELATION_COLUMNS_QUERYfills it with attnum > 0, non-dropped,pg_table_is_visiblecolumns, which matches whata.*expands to for a visible table.- Count guard moved: the early
target_list.len() != countcheck is replaced byout.len() != countafter expansion. - Oracle and artifact:
whole_row_json_self_join's preserved side is no longer nullable. Three newinference_precisioncases cover a full self-join,a.*on the preserved side andb.*on the nulled side.engine.wasmis rebuilt.postgres.mdx:444already states the rule ("told apart by its alias"), so the docs need no change.
claude-opus-5-5 ｜ 𝕏
colinhacks
force-pushed
the
fix/pg-self-join-star
branch
from
October 23, 2026 22:35
53aabcd to
19e830d
Compare
Contributor
There was a problem hiding this comment.
✅ No new issues found. The edge case from the last review is closed.
Reviewed changes
Since the last review, one commit adds two guards to aliased_columns, so a misjudged alias.* width can no longer line up with the wrong columns.
(x).*gives up: a target that is anA_IndirectioncontainingA_Starnow returns no readings at all. Every target other thanalias.*is therefore exactly one column, and the finalout.len() != countcheck catches a wrong width on its own.- Schema-named stars give up:
itemsnow collects intoqualifiedthe alias of eachRangeVarwritten with a schema, and a star over one of those aliases returns no readings.widthonly ever answers for the visible table, so this keeps it from reporting the visible table's width for a different table of the same name. - Unit tests: the counter-example from the last review, a variant without the composite and a lone
(b.addr).*are now asserts ina_star_of_a_reading_is_its_tables_columns_by_that_reading.
claude-opus-5-5 ｜ 𝕏
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
In a table joined to itself,
alias.*lost track of which reading each column came from. So every column of the table came out nullable, including on the preserved side. The engine tells a self-join's readings apart by alias foralias.columntargets, but it gave up as soon as a*appeared in the select list. Ato_jsonb(a)is read through ana.*probe, so it was affected too:earlier: { id: int4?, name: text?, email: text? }?earlier: { id: int4, name: text, email: text? }, withlaterunchanged as{ id: int4, name: text, email: text? }?The rule
alias.*over a table's reading stands for that table's columns, in order, by that reading. Each column is nullable where an outer join nulls that reading, and otherwise it is as the catalog says. A bare*, a star of an item that isn't a table, and a column count that doesn't add up still say nothing, as before.Oracle (
inference_precision)whole_row_json_self_join:earliergoes from{ id: int4?, name: text?, email: text? }?to{ id: int4, name: text, email: text? }.whole_row_json_self_full_join: both objects stay nullable, and their fields keep their own nullability.whole_row_star_self_join(a.*, b.name):{ id: int4, name: text, email: text?, later: text? }.whole_row_star_self_join_nulled(a.name, b.*):{ earlier: text, id: int4?, name: text?, email: text? }. The nulled side stays nullable.