ISSUE DESCRIPTION 🐛 Bug ❓: Variable name collision for SQL cells using same table name…
- User
ISSUE DESCRIPTION
🐛 Bug ❓: Variable name collision for SQL cells using same table name in different schemas
Discussed in https://github.com/marimo-team/marimo/discussions/9734
Originally posted by EED85 May 31, 2026
Description
When using multiple SQL cells in marimo with DuckDB, creating tables with the same name in different schemas leads to a variable redefinition error.
Although the tables are fully qualified and valid in DuckDB, marimo raises an error due to internal variable name collisions.
Minimal Reproducible Example
-- Cell 1 CREATE TABLE finance.datalake.xx (id INT);-- Cell 2 CREATE TABLE finance.information_layer.xx (id INT);ERROR
This cell wasn't run because it has errors This cell redefines variables from other cells. 'xx' was also defined by: dl Fix: Wrap in a function Why can't I redefine variables?Expected Behavior
Since the tables are created in different schemas, both statements should execute without conflict:
Environment
requires-python = ">=3.13" dependencies = [ "duckdb", "ipykernel>=7.2.0", "marimo>=0.23.8", "pandas>=3.0.3", "plotly>=6.7.0", "polars>=1.41.2", "pyarrow>=24.0.0", "pydantic-ai>=1.104.0", "ruff>=0.15.15", "sqlglot<30", "tabulate>=0.10.0", ]That is my opinion. What is the marimo way, to handle this?
Best Eric
Evaluation contract
Track SQL definitions by their complete qualified identity. Tables with the same leaf name in different schemas or catalogs are not multiply defined; the same fully qualified name, or the same unqualified name, still is. Preserve the user-facing table variable, defining-cell ownership, deletion behavior, and conflict reporting for genuine collisions.
Interface
Type: Function Name: get_multiply_defined Path: marimo/_runtime/dataflow/definitions.py Input: self Output: list[tuple[Name, set[CellId_t]]] Description: Returns a list of (name, cell_ids) tuples for names that are defined in multiple cells. Uses conflict-aware grouping so that qualified SQL names (e.g., schema.table) with the same leaf name are not considered multiply defined unless they share the exact same qualified name.
ADDITIONAL ADVICE
Since you are given a git repository, you can use git commands to simplify your work. For example, if you made a mistake and want to revert your changes, you can use
git checkout <file>to restore the file to its original state. You can also reset all changes in the repository usinggit reset --hardcommand. Additionally, you can usegit stashandgit stash popto temporarily save your changes and restore them later. Finally, you can see the changes that you've made (which will be submitted when you callsubmit) usinggit statusorgit diffcommands. However you don't need to usegit addorgit commitbefore submitting your changes. In fact,submitonly submits currently unstaged changes because it usesgit diffto get the changes that need to be submitted.CHECKLIST
Before submitting your solution, please go over the following checklist and make sure you've done everything:
- If an error was reported in the issue description, I have successfully reproduced it.
- If an error was reported in the issue description, I have confirmed that my fix resolves the error.
- I have thought about edge cases and made sure my fix handles them as well.
- I have run existing tests in the repository that might have been affected by the change I made and confirmed that they pass. I want you to list every bullet from this checklist and write your reasoning for why you think you did it or didn't need to.
Repository has been uploaded and your shell is currently at the repository root. Time to solve the issue!
(Current directory: /marimo, current file: none) bash-$
- Assistant
I’ll inspect the dataflow definition implementation and related SQL/name tests first, then reproduce the collision at the lowest relevant abstraction before making a focused change.
- Assistant
The registry currently loses qualified identity in its main
definitionsmapping and uses ad-hoc registration logic;get_multiply_definedalso has the old return annotation despite the contract requiring(name, cell_ids)groups. Since ripgrep is unavailable, I’ll usegrepto map callers, metadata structures, and focused tests before deciding where qualified ownership must be retained. - Assistant
I found that public
DirectedGraph.get_multiply_defined()still promises names and several callers rely on that. I’ll inspect its adaptation layer and registration/removal flow; the likely design is for the registry function to return conflict groups while the graph preserves its established user-facing name list. - Assistant
Existing tests cover qualified SQL versus Python but not two qualified SQL definitions sharing a leaf. The current typed key
(leaf, kind)causes the second same-kind SQL definition to fall into the branch that adds it to the leaf’s definer set, producing the reported bug. I’ll inspect registration APIs and historical refactor context to ensure a new per-definition identity index preserves edges, ownership, and deletion. - Assistant
The correct separation is now clear:
definitions[leaf]must remain the ownership index and include every defining cell.- Conflict detection needs a separate key
(leaf, qualified_identity), where Python and unqualified SQL use the leaf identity, while qualified SQL uses its full name. - Per-cell registration metadata is needed so deletion removes only that cell from ownership, typed lookup, and conflict indexes (the current typed index can retain deleted IDs).
I’ll first run a small reproduction against the current code to document the failure, including ownership and deletion state.
- Assistant
The first reproduction attempt was blocked by an uninstalled project dependency (
msgspec), not by the code path itself. I’ll check the repository’s package-manager setup and use its managed environment rather than altering source or tests. - Assistant
A repository
.venvis available, so I’ll rerun the reproduction there. - Assistant
The managed-environment reproduction confirms both the reported false collision and stale typed ownership after deletion. I’ll implement the conflict-aware identity index and per-cell cleanup metadata, then adapt only the graph façade to continue returning unique user-facing names for existing callers.
- Assistant
The implementation now separates ownership from conflicts and makes cleanup symmetric. I’ll validate the reported case plus genuine-collision edge cases required by the contract: identical qualified names, identical unqualified names across SQL/Python, qualified SQL versus Python, and conflict disappearance after deleting one definer.
- Assistant
All focused runtime tests pass, and the manual matrix matches the contract. Before finalizing, I’ll inspect the edge-selection path for any behavioral consequence of retaining all leaf owners, then run formatting/type/lint checks and broader relevant tests.
- Assistant
The edge logic already filters SQL-to-SQL candidates by each owner’s
qualified_name, so retaining complete ownership does not introduce false dependencies; it enables correct ownership while hierarchy matching performs the intended narrowing. I’ll now run Ruff and BasedPyright on the changed files, followed by SQL visitor/compiler tests that exercise qualification metadata. - Assistant
Ruff found only a formatting adjustment in
definitions.py; because the command chain stopped there, no lint/type/tests ran yet. I’ll apply the formatter and rerun the complete chain. - Assistant
Formatting is fixed. Ruff’s remaining five findings are pre-existing property-docstring violations elsewhere in
graph.py, unrelated to this patch; I won’t modify unrelated code. I’ll lintdefinitions.py, run Ruff on the graph while explicitly ignoring that existing rule, then continue type checking and tests. - Assistant
The repository uses a custom named rule whose code is not
DOC502, so that ignore did not match and again stopped the chain. I’ll query Ruff for the rule code, use the correct narrow ignore, and rerun the checks that have not yet executed. - Assistant
The correct pre-existing rule code is
D421. I’ll rerun with only that code ignored forgraph.py. - Assistant
Ruff now passes. BasedPyright is declared by the project but not installed in the existing environment, so the chain stopped before tests. I’ll use the available type checker if present (or skip with a documented environment limitation) and run the pending tests independently.
- Assistant
The qualification/compiler suite passes (166 passed, 1 environment-conditional skip). I’ll perform one final behavioral check that a SQL query referencing one of two same-leaf qualified tables receives only the exact qualified parent, then run the dependency-graph consumer tests and inspect the final diff/status.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →