* Migration adding 'width' to Dashboards

3 migrations:
 - 1st adding the width column with default value of 'fixed'
 - 2nd updating all existing dashboards to have width 'full', which corresponds to what the current behaviour is (will
 be the 'old' behaviour after the fixed-width project lands).
   - The rolloback here is necessary but we don't care what happens as the column will be dropped immediately in the
   next rollback anyway
 - 3rd sets the notNullableConstraint. DefaultNull is 'full' here, just in case there's an existing dashboard whose
 width value is not yet set from the 1st migration. Don't know how that could happen, but its here in case

* Dashboard PUT api endpoint accepts width changes and updates appdb

update-dashboard function now is aware of the :width key so those changes can end up in the transaction.

Also added a width test that asserts that the value's default is "fixed", it can be changed, eg. to "full", but cannot
be changed to other values.

* Add width to revision tests

* Fix dashboard revision tests.

:width key is now needed in some revision tests. As well we need a string communicating that the :width setting has
changed from 'full' to 'fixed' or vice-versa.

* Fix comments/remarks in migration to be accurate

* Attempt to fix default not working mysql/mariadb

* Set default in dashboard model

Signed-off-by: Adam James <adam.vermeer2@gmail.com>

* Revert default :width 'fixed' value.

* Explicitly add default value 'fixed' for MySQL/MariaDB

---------

Signed-off-by: Adam James <adam.vermeer2@gmail.com>

Co-authored-by: Ryan Laurie <30528226+iethree@users.noreply.github.com>

I misunderstood the desired behavior when I implemented this before. We
only want the limit to apply when the `add-default-userland-constraints`
middleware is applied to the query, not when we're downloading the data
directly.

This is a bit of a rat's nest. If there's a better way to approach the
problem, I'd be happy to hear it. I added documentation for the odd approach
as much as possible.

The main difficulty is that the userland middleware does not have access
to db-local settings. So if we want db-local settings to apply only to
userland, we have to hack our way around it.

The solution here is to divide the userland middleware into two parts:

- the actually-userland bit, which just marks the query as needing
default limits, and

- a middleware inserted deeper in the middleware stack (just before we
actually apply limits) that actually calculates those limits and
attaches them to the query.

CSV appends: Do not allow numbers with fractional component to be uploaded to an integer column (#38274)

Sometimes a test passes when run by itself, but fails due to ordering
issues. This is typically caused by another test failing to properly
clean up after itself, leaving the database in a dirty state. For
example, we might change permissions and forget to change them back, so
that another test hits an unexpected permissions failure.

Finding the cause of these has been a source of pain for me, so I wrote
a small snippet to do it for me.

If you have a test that passes when run by itself, but fails when run
along with all the other tests, you can run
`(dev/find-root-test-failure! #'my-ns/my-intermittent-failure-test')`.

It will run *all* tests. After each one, it'll run the test you passed
in. Once that test starts failing, it'll alert you about which test
caused it to fail.

* Support converting aggregation references to legacy references
* Pass query and stageIndex to legacyQuery

---------

Co-authored-by: Alexander Polyankin <alexander.polyankin@metabase.com>

Fixes #36974.

See the Slack [discussion](https://metaboat.slack.com/archives/C04CYTEL9N2/p1706193649444639)
for details.

* Chain filter search should check block perms

* Simplify test

* Field values cache needs to check permissions

If a user does not have permissions to the field, we should not return
the cached field values.

* Update default-get-or-create-field-values-for-current-user to check can-read too

* Revert params.field-values change

* Use query processor permissions checks before returning cached values

* Return true from check

* Fix test

* Check for ee config

For big databases autocomplete can be quite slow. At stats, query in regular form takes about 80 ms to run. Filtering out inactive tables and adding an index (pgsql-only unfortunately) for ordering clause brings that down to 10 ms.

Also I suspect that in a single session you'll hit similar autocomplete rules multiple time, so some client caching could help with performance.

* Fix NPE in schema pattern matching fns in sync

* Require PG

* adds info about what value validate got to the exception message

* test fixups

* fix the other test

* remove inline def

This regressed in 48 - it only works when either none of the
several filters has a value, or when all of them do.

This PR restores the 47 behavior: whatever subset of the filters have
values (including defaults) will be used, and the others are ignored.

The fix is on the BE but the change actually occurred in the FE -
previously only those parameters with values were sent in the request.
We might want that FE behavior back, but the new BE logic should work
either way.

Fixes #38012.

* add personal-only to collection endpoint

* address some comments

* fix test

* fix select-collections test

* Update src/metabase/api/collection.clj

Co-authored-by: bryan <bryan.maass@gmail.com>

* Update src/metabase/api/collection.clj

Co-authored-by: bryan <bryan.maass@gmail.com>

* address comments

---------

Co-authored-by: bryan <bryan.maass@gmail.com>

This reverts commit d78f324c, which reverted 7b0a906a.

* Fix flake in fingerprint test:

```
FAIL in metabase.sync.analyze.fingerprint-test/fingerprint-test (fingerprint_test.clj:287)

:sparksql
 Fingerprints should actually get saved with the correct values Text fingerprints
expected: {:global {:distinct-count 100, :nil% 0.0},
           :type {:type/Text {:average-length #<Fn@2ee5402f metabase.sync.analyze.fingerprint_test/fn__139755[fn/fn]>,
                              :percent-email 0.0,
                              :percent-json 0.0,
                              :percent-state 0.0,
                              :percent-url 0.0}}}
  actual: nil
```

* Typo fix

Fixes #36856.

* Filter out audit tables in search results

* add test

* Fix salt in pulse unsub hash
* Workaround for read-only settings not supporting temp overrides yet

- Introduce an "init" hook for application settings.
- Add guards against introducing race conditions again in other ways.
- Fix the usage of hygienic scope macros with read only settings.
- Lots of misc improvements to settings API and implementation.

This init hooks better ergonomics, and protects against race conditions. 
The previous pattern of calling setters inside getters almost caused a P0. 

See https://github.com/metabase/metabase/issues/37801.

Part of #37173.

Co-authored-by: Noah Moss <32746338+noahmoss@users.noreply.github.com>

These are helpers for going from an aggregation or breakout
clause to the relevant `:metadata/column`. (Nil for top-level
aggregations like `:count`.)

Fixes #37120.

That applies to both freestanding temporal buckets and to breakouts with
`:temporal-unit` set.

Fixes #36978.

Because aggregation refs are based on `:source-uuid` they are not
comparable between queries, which breaks some logic to find columns
added and removed between two versions of a query.

This adds a fallback based on the newly added `:source-name` option on
aggregation refs.

Fixes #37851.

* Remove references to shared/ directory which no longer exists

* Update shadow-cljs and pinned cider/nrepl versions

* Support legacy metadata and raw Dates for brush filter helper functions

* Fix fn name

* Apply suggestions from code review

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

---------

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

Part of #37173.

Exclude :card from the parameter types which require defaults to be considered fully parameterized (#37764)

* Add `align-temporal-unit-with-param-type-and-value` multimethod

* Fix tests and add a changelog entry

* Update the method docs

* Rearrange the default methods call order

* Revert formatting changes

* Move the clj-kondo ignore

* Update docs/developers-guide/driver-changelog.md

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

---------

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

Fixes #37659

* Use :skip-results-metadata? true in pulse query execution

Without this, the results-metadata middleware ends up replacing a model's metadata.

Using the correct flag causes that middleware to just be skipped, preserving whatever metadata already exists on the model.

* Add a testing form to explain the assertion as clearly as possible.