Activity

1. cmgoffena13

   added a parent issue on Sep 4, 2026

   • Improve format, lint, and test commands to make pre-commit setup simple and more efficient #6020
2. tripleaceme commented on Sep 13, 2026

   

   tripleaceme

   on Sep 13, 2026

   Contributor

   More actions

   I'd like to take this one if it's free — unassigned with no linked PR right now.

   Plan: skip Context.load() when positional paths are given and format those files directly. _format only needs three things from the loaded object — _path, dialect and formatting — and all three are available from the file's own MODEL/AUDIT header plus config_for_path, which resolves a project config from a path without loading anything. is_meta_expression on the first parsed expression is what separates a model or audit file from macros/*.sql, so non-model SQL stays a no-op without needing the project graph.

   One behavioural question before I start. Today "is this file a model?" is answered by membership in the loaded project. Without the load it would be answered by parsing the header, and those two differ for a .sql file that has a MODEL header but is excluded from the project — sitting outside models/, or matched by ignore_patterns. Today sqlmesh format path/to/that.sql skips it; a pure header check would format it.

   I'd rather preserve today's behaviour and still honour ignore_patterns and the models//audits/ directories, since both are cheap to check against the config without loading. Say if you'd prefer the simpler "any file with a MODEL/AUDIT header gets formatted" instead.

   Also flagging that #5944 touches Context.format() as well. It's been quiet since August — I'll keep this change confined to the load path so the two stay separable.
3. tripleaceme commented on Sep 13, 2026

   

   tripleaceme

   on Sep 13, 2026

   Contributor

   More actions

   One finding while mapping this out, since it affects the spec above.

   The acceptance says to format a path "if [it is] a SQL model or standalone audit", but standalone audits aren't formatted today. Context.format iterates chain(self._models.values(), self._audits.values()), and an AUDIT (..., standalone true) is loaded into self._standalone_audits, which is a separate dict. So it's skipped.

   Reproduced on main with a project holding one model and two audit files:

   audits/ma.sql  (AUDIT(name ma, dialect 'duckdb'))                   -> reformatted
   audits/sa.sql  (AUDIT(name sa, dialect 'duckdb', standalone true))  -> byte-identical, untouched
   

   So "format only those paths that are models or standalone audits" is a behaviour change on top of the load change, not just a restatement of what happens now.

   Happy to include it — the header-based check I described treats both kinds of AUDIT the same, so standalone audits would start being formatted, which I think is what you want given the wording. But it's a separate change in effect, so tell me if you'd rather I preserve the current skip and let the standalone-audit gap be its own issue. I'll keep it isolated in the diff either way.
4. cmgoffena13

   assigned

   tripleaceme

   on Sep 15, 2026