* Add Kondo linter to disallow hardcoded driver names in tests

* Fix all tests using hardcoded driver names

* Test fixes 

* Appease Kondo

* Remove normal-drivers-except

* Test fix 

* Test fixes 

* MongoDB test fixes

* Enable type info computation for fields coming from model source query

* Add test

* Add handling for fields coming from joins

* Handle column matching errors

* Fix test

* Fix test

* Fix test

* Fix test

* Fix test

* Add Timeseries insights even when Custom Columns exist

WIP

Fixes 46244

I don't know why it was (maybe still is) required that the :other columns be empty before providing insights.

* might be a solution to changing trendlines when data doesn't change

The insights code that powers the frontend trendline requires taking a sample of the dataset.

This sampling has some randomness associated by design, and should maintain its randomness. But it makes some sense to
use the exact same sample when the input has not changed at all, hence trying this memoize approach out.

Might not be the final solution, but it's a start

* add test to show that insights are computed

* use java.util.random with a seed to keep insights stable

* Add testcase from https://github.com/metabase/metabase/issues/43915



* Try a custom formatter if the default formatter for OffsetDateTime doesn't work

* Add testcase

* Make auxillary formatter private

* Catch a more specific exception

* Throw IllegalArgumentException on invalid dates

* Use strict resolver and fix test

* Empty commit to assign credit where it's due

---------

Co-authored-by: Chris Truter <chris@metabase.com>

* Dowload Row Limit Env Var

Adds `MB_DOWNLOAD_ROW_LIMIT` to enable changing the row limit on downloads and alert/subscription attachments (not the
rendered tables, but the .csv, etc.).

Based on: #44982 (Thanks, @r-kot)

Partially implements: #28144

The difference in this PR compared to #44982 is that the download limit applies to all downloads, the only exception
being when the limit is above `qp.i/absolute-max-results` (1048575, based on Excel's limitation); in such a case, the
user supplied limit is only used if the download is csv or json, and `qp.i/absolute-max-results` is used for xlsx.

This PR also fixes alert/subscription attachment limits; prior to this, they were set to the in-app limit of 2000
rows, but now they will follow the user supplied download-row-limit.

This PR also adds a test to the downloads-and-exports test namespace, confirming that they follow the supplied limit,
or the max limit if none is supplied.

* add test confirming the default limit works

* fix test to use the download-row-limit

* address review feedback

* Update src/metabase/public_settings.clj

Co-authored-by: Cal Herries <39073188+calherries@users.noreply.github.com>

* add test covering case where download-row-limit is unset

---------

Co-authored-by: Cal Herries <39073188+calherries@users.noreply.github.com>

* Export Formatting Correctness

WIP

Several bugs exist around the correctness of export formatting, this aims to fix them.

Fixes #27374

When a model has viz settings/metadata changed after save, the download won't incorporate said changes. At least it's
clear that scale doesn't work.

* number formatter merges settings from columns too

* column formatting on aggregated cols works in xlsx now too

* fixes 43040 by grabbing global-type-settings in csv formatter

* Last Used Param Value doesn't fill in with default value if cleared

Fixes #46368

The Last used parameter value is cleared when the frontend sends `nil` as the parameter's value. This works except in
cases where a user clears a parameter with a default value. In this case, we want to indicate that the parameter is
cleared, so we store `nil` in the appdb in such a case.

* Update src/metabase/query_processor/dashboard.clj

Co-authored-by: metamben <103100869+metamben@users.noreply.github.com>

* add assertion showing that 'nil' with default is still stored

---------

Co-authored-by: metamben <103100869+metamben@users.noreply.github.com>

Fixes: #46558

In #45575 we introduced casting uuid fields for drilldowns.

However we failed to include a check that the rhs was itself another
field of uuid type and so no casting should occur. This broke joins on
models when the database_type of the field may not be known.

* Try without broken refs

* Disable broken test

* Disable broken test

* fix: Handle tokens in strings when parsing params

Fixes #38051
Fixes #36262
Fixes #11755

* Fix handling snippets with single quotes in name

== Goal ==

Prevent admins of Metabase Cloud instances with Metabase Cloud Storage
from accidentally breaking their upload configuration, since in this
scenario that is automatically managed by Metabase Cloud to use the
attached data warehouse.

== How to test ==

On a subscription with Metabase Cloud Storage:
* The "Uploads" section in the left sidebar of `/admin/settings` does
  not show
* A "not found" page is shown at `/admin/settings/uploads` where one
  would otherwise expect the "Uploads Settings" page
* `PUT /api/setting` with an `uploads-settings` field in the body fails
  with `403 Forbidden` and the setting is not changed
* `PUT /api/setting/uploads-settings` fails with `403 Forbidden` and
  the setting is not changed
* `GET /api/setting` contains an element with `key: "uploads-settings"`
  and a value
* `GET /api/setting/uploads-settings` responds with the value
* `GET /api/session/properties` contains the `uploads-settings` field

Closes: https://github.com/metabase/harbormaster/issues/5121


Co-authored-by: Ryan Laurie <iethree@gmail.com>

* Add :relative-time-interval mbql function

* Add relativeDateFilterPartsRelativeTimeInterval

* Update display-name-method :relative-time-interval

* Update desugar-relative-time-interval

* Define relative-time-filter op

* Add or update tests

* Update tests

* Update display-name-method

* Update test

* Fix positive relative-time-interval shift

* Add checkins:1-per-day:60 dataset

* Update var name

* Bail out from optimization of relative-time-interval desugared filters

* Update dataset kondo hook

* Generate cache required for kondo hooks in kondo.sh

* Rename dataset

* perf: Optimize validation

* Bump Malli version

* perf: Optimize validation

### Description

This change should fix some explosions in stats due to phantom fields, while also making things a bit more rigorous.

- Record the table schema in query_field - especially useful for unknown tables
- Allow null table names in query_field - they might be phantoms, or they might not
  - We still have the option to filter them out in tools we build on top
- Tighten up matching logic to consider schema

* Fix coll permissions for audit collection

We've had a function in `models.collection` for a while that has taken a
permissions set and returned a set of collection IDs that the user has
permissions on. I refactored this recently, but didn't notice that it
was actually doing the permissions checks slightly incorrectly.
Specifically, because it only looks at the user's permissions and the
collection IDs and doesn't use `mi/can-read?` or `mi/can-write?`, it's
completely indifferent to whether a collection is an audit collection or
not.

This is arguably not a *permissions* issue: the two errors that could
come about here are:

- someone sees the audit collection even though the audit feature is
disabled, or

- someone is presented with the audit collection in a context where only
writable collections should be present - when they try to actually write
to it, it fails (since then we're doing the real permissions check).

The primary motivation for this fix was to prevent audit dashboards and
cards from appearing in the list of stale items.

* Fix hardcoded collection ID

We were creating a timeline in a fixed collection ID, that in tests
happened to be the ID of the Metabase Analytics collection.

* Update values-from-card-query

* Add test

* Update values-from-card-query

* Add missing issue no

* Avoid adding temporal-unit to lhs cols

* Use :default temporal-unit

* Simplify logic, update comments

* Update substitute-field-filter-test

* Update align-temporal-unit-with-param-type-test

* Update field-filter-date-test

* Use end-excludding gte lt filter for DateTime fields

* Update substitute-field-filter-test

* Update align-temporal-unit-with-param-type-test

* Update field-filter-date-test

* Update bigquery test + comment

* Update date-str->qp-aware-offset-dt

* Add guard for unexpected date string format

This is just for the completeness, I haven't encountered it.

* Address review remarks

* Update comment

* Squash

* Clean up query modifier for validating card references (#46283)

* Fixup bad git merge on test expectation

* Tweak docstrings and de-densify logic

This reverts commit 72a20181.

* Add :relative-time-interval mbql function

* Add relativeDateFilterPartsRelativeTimeInterval

* Update display-name-method :relative-time-interval

* Update desugar-relative-time-interval

* Define relative-time-filter op

* Add or update tests

* Update tests

* Update display-name-method

* Update test

* Fix positive relative-time-interval shift