* Add some dox

* Rename test-runner & cloverage -> metabase.test-runner & metabase.cloverage-runner

* Move JUnit-related test runner code to separate namespace

* Bespoke JUnit-output code

* Delete unused .lein-classpath file

* Do JUnit writes in a cached thread pool

* Add failing test to make sure JUnit output works as expected...

* Move bad-test to metabase.bad-test so the FAILING tests actually run

* FASTER TEST LOADING 

* Test CircleCI :system-out

* Final cleanup (hopefully)

* Sort some namespaces; allow initializing :plugins when namespaces get loaded for now.

* Need those [[tags]] in docstrings

* Fix bad docstring

* rmv old bucm icons and remove verified fill color

* add moderation action section to sidebar

* add moderation review icon to the saved question header button

* hide moderation section when is not a moderator

* add UI for ModerationReviewBanner

* Backend for moderation-review

- create table moderation_review. Same as before but also has a
  "most_recent" boolean flag for the most recent moderation for easy
  lookup
- POST /moderation-review/ . Status can be "verified" or nil
- must be an admin to post
- No PUT or edit route yet. Not sure if this is even
  necessary. _MAYBE_ to edit the text, but certainly not for the
  status, ids, etc. If there's to be history, let's build some history
- Ensure we never have more than 10 reviews. Adding a new review will
  delete the older ones, mark all old ones as not `most_recent`, and
  add the newest one as `most_recent true`
- Ensure the card actually exists before creating the mod review
- Since admin only at this time, don't need to check moderate
  permission or view permission
- When hydrating ensure reviews are ordered by id desc. Should mimic
  the created_at desc

* fix moderation review banner tooltip offset

* disable verification button when already verified

* rmv iconOnly prop because it seems to do nothing

* update getLatestModerationReview to rely on most_recent boolean

* Return 400 on invalid status to post /moderation-review

the schema was using keywords on the left hand side rather than the
symbols. Required a change to the docstring generator, when it made a
docstring for enums, it would call (sort (:vs enum)) and need to
string em.

* Add ModerationReview model to models.clj and copy infra

* hydrate moderation reviews on cards

* clean up + wire up to BE + ensure mod buttons don't show for normal users

* rmv unused moderation redux logic from QuestionDetailsSidebarPanel

* finish writing unit tests for FE

* ensure getIconForReview returns an object

* enable/disable verify button tooltip when unverified/verified

* add e2e tests

* fix tests

* styling tweaks

* more styling on moderationReviewBanner

* add function for abbreviated timestamp

* increase fontsize of timestamp back to 12

* fix tooltip offset

* ensure custom locale is separate from 'en' and not used for other languages

* Deletion moderation reviews when deleting cards

i had actually thought this was a much larger problem. But it turns
out we almost never delete cards (thanks comment!). And so we won't
really generate a lot of garbage.

I was worried that since we aren't using actual foreign keys but just
`moderated_item_type "card"` and `moderated_item_id 2` we would have
deleted cards with these moderation reviews but that is not the case
as the cards aren't deleted.

* hide verify disabled button when a question is verified

* update test to use queryByTestId

* Hydrate moderation reviews on cards on ordered cards

* Handle mysql's lack of offset functionality

mysql cannot handle just a `offset` clause, it also needs a limit

clause grammar from
https://dev.mysql.com/doc/refman/8.0/en/select.html:

[LIMIT {[offset,] row_count | row_count OFFSET offset}]

select id, name from metabase_field offset 5;         -- errors
select id, name from metabase_field limit 2 offset 5; -- works

Since our numbers are so small here there is no worry and just do the
offset in memory rather than jump through hoops for different dbs.

* Batch hydrate moderation reviews

* Don't let /api/user/:userId failure conceal moderation banner

* fix moderation cy tests

* work around possible bug in toucan hydration

dashboards hydrate ordered cards
(hydrate [:ordered_cards [:card :moderation_reviews] :series])

Ordered_cards are dashboard_cards which have an optional card_id. But
toucan hydration doesn't filter out the nils as they go down. It seems
toucan returns a nil card, and then when hydrating the
moderation_review, passes the collection of all "cards" including the
nil ones into the hydration function for moderation_reviews

This feels like a bug to me

* Cleanup moderation warnings

* Docstring in moderation review

* include hoisted moderated_status on cards in collections api

* Expect unverified in test 



Co-authored-by: dan sutton <dan@dpsutton.com>
Co-authored-by: Maz Ameli <maz@metabase.com>
Co-authored-by: alxnddr <alxnddr@gmail.com>

* Revert changes to activity and activity-test namespaces

* Add some sanity-checks to help debug the failing test

* Delete accidental newline

Change default behavior of pagination to be what i imagined it to be. Really should've taken Dan's question to heart and actually tested that bit...

* Fix some Eastwood failures

* Fix a lot of Eastwood errors now that it runs against test namespaces [ci skip]

* Run Eastwood against test namespaces [WIP]

* Bump Eastwood version to fix some errors in Eastwood itself

* Fix another lint warning

* Fix test failure

* More test fixes 

* Fix all the warnings!

* Test fix 

* Bump Eastwood GH action timeout to 20 minutes (!)

Implement JDBC based Presto driver

Adding new Presto JDBC driver using the PrestoDB JDBC driver from `https://github.com/prestodb/presto`

Marking the old Presto driver as being `superseded-by` the new one

Pulling out common Presto code into new presto-common driver (modeled after the relationship between, ex: `googleanalytics` and `google)`

Putting common QP/HoneySQL logic into the new (abstract) :presto-common driver

Updating :presto driver to extend from the new common driver and only adding HTTP/REST API related methods

Adding implementation of Presto JDBC driver, named :presto-jdbc, extending from :presto-common and :sql-jdbc

Using com.facebook.presto/presto-jdbc as underlying JDBC driver dependency (since this is explicitly for Presto clusters, as opposed to Trino)

Adapting code from the existing Presto driver where appropriate

Adding new dependency-satisfied? implementation for :env-var, to allow for a plugin to require an env var be set, and making the Presto JDBC driver depend on that (specifically: `mb-enable-presto-jdbc-driver`)

Adding CircleCI configuration to run against the newer Presto (0.254) Docker image

Adding explicit ordering in a few tests where it was missing

Fixing presto-type->base-type (timestamps were being synced as :type/Time because the regex pattern was wrong)

Add tx/format-name test implementation for :presto-jdbc to lowercase table name

Make modified test Oracle friendly

Fixing bug parsing the `[:zone :time]` case within `metabase.util.date-2.parse/parse-with-formatter`; the offset is nil so it can't be passed directly in this case, so use the `standard-offset` fn (which was moved from `date-2` to `common` to get a standard offset for that zone

Fixing more test failures by adding explicit ordering

Changing sync to check whether the driver supports foreign keys before attempting to sync those (since drivers might throw an exception if attempting to check)

Moving some common test dataset functionality between :presto and :presto-jdbc to a new test.data ns for :presto-common

Adding HoneySQL form for :count-where, since we have to explicitly give a higher precision Decimal in order for Presto to not reduce the precision in results

Put limit within subquery for `expression-using-aggregation-test` (since ordering from subquery is not guaranteed in Presto)

Adding impls for ->prepared-substitution to handle substitutions for native query params

Adding HoneySQL impls for `mod` (to do as "mod(x,y)" and `timestamp` (since it's a function with no parens to invoke) functions

Adding various `sql.qp/date` impls that use the `AT TIME ZONE` operator to account for report tz, and make bucketing tests happy

# Backend changes

Introducing new `superseded-by` property to plugin manifest YAML, which will indicate the driver that is to eventually replace this one (and will drive UI/UX behavior).  If a driver declares this property, then it's considered to be deprecated in favor of the specified one.

Adding top level `test_modules` directory (with the same structure as modules) for the sole purpose of module/plugin testing of YAML files, which will not be included with the driver build

Updating `driver-plugin-manifest` to look for the new `test_modules` directory in addition to `modules`, when loading the driver manifest

# Frontend changes

Calculate `supersededBy` and supersedes maps from the "superseded-by" property for each engine

Change the options for the engine field to use a function to dynamically show the legacy driver if allowed by rules (either the new driver is selected, or the legacy driver was already selected for an existing DB, or the driver is not superseded by anything)

Add new `DriverWarning` component to show these warnings based on supersede status

Co-authored-by: Anton Kulyk <kuliks.anton@gmail.com>

* offset is a reserved word in mariadb as of 10.6

- https://mariadb.com/kb/en/reserved-words/
- https://github.com/MariaDB/server/commit/299b935320

I think it is related to https://jira.mariadb.org/browse/MDEV-23908
"Implement SELECT ... OFFSET ... FETCH ..."

* Handle new names for utf8 in mariadb

https://mariadb.com/kb/en/unicode/

- utf8: Until MariaDB 10.5, this was a UTF-8 encoding using one to
three bytes per character. Basic Latin letters, numbers and
punctuation use one byte. European and Middle East letters mostly fit
into 2 bytes. Korean, Chinese, and Japanese ideographs use 3-bytes. No
supplementary characters are stored. From MariaDB 10.6, utf8 is an
alias for utf8mb3, but this can changed to ut8mb4 by changing the
default value of the old_mode system variable.

- utf8mb3: UTF-8 encoding using one to three bytes per
character. Basic Latin letters, numbers and punctuation use one
byte. European and Middle East letters mostly fit into 2
bytes. Korean, Chinese, and Japanese ideographs use 3-bytes. No
supplementary characters are stored. Until MariaDB 10.5, this was an
alias for utf8. From MariaDB 10.6, utf8 is by default an alias for
utf8mb3, but this can changed to ut8mb4 by changing the default value
of the old_mode system variable.

* Fix Card async-metadata running query even when metadata can be determined programatically

* Update another test

* Adjust affected repros by waiting for the metadata to load (#17068)

* Wait for the `result_metadata` to load

* Reduce the number of bins for the faster execution

* Update repro for #12568

Co-authored-by: Nemanja Glumac <31325167+nemanjaglumac@users.noreply.github.com>

I thought (and assured dpsutton, lol) that the default behavior of the pagination spliced in was just not paginating, if there wasn't a limit passed in. Unfortunately, it appears that I tested this on the endpoint without pagination spliced in... so the default behavior isn't that way, it paginates if you don't pass in stuff.

So rip it out of the DB endpoint (only) for now and put it back in when we paginate

Authored-by: Howon Lee <hlee.howon@gmail.com>

* Smartscalars

* smartscalar html styling

* Docstring, ns, prevent top-level trs/tru

* Scalar rendering as text in slack

* Fix tests for pulse

* Tests for slack scalars

We defriendlified but the migration running in the first place is not happening for some users. Investigating but it's better just to be safe on this front

* Breadth first search on graph for chain filters

- abandons at max-depth and looks in breadth first fashion rather than
descending as far as possible on each possible path

* Add loop detection

not actually fooled by loops since we abandon after a certain path
length. But will remove some false starts before they have to be
killed by the path length check.

* Add seen optimization of nodes

We reach nodes in the shortest path possible, so keep a list of nodes
visited and don't revisit them: we got to them as fast as possible,
and once there, there's no new nodes that we can see from that vantage
point.

This allows a serious speedup and should hopefully fully resolve the
cpu issues seen by some.

* alignment

* Fix Oracle/Vertica comparison against empty string

* Remove extra println

* Add test for ::empty-string-is-null

* Fix SQL saved question reference to Saved Questions with parameters

* Test fixes 

* Don't run test for drivers that don't support FKs

* Test fix 

* Test fix 

When metrics get macro-expanded in xrays, they don't have joins in them. I stick joins in them. Joins! Joins! Joins!

But the underlying problem is that the metric macros have no idea what this "join" business is. Unhappy times!

* Only save successful-query-execution if it wasnt a cached result

the only thing this function does is update the rolling average query
execution time

* Failing tests for cache execution stats

These tests fail, but they shouldn't.

The save-execution information should not be called twice, and the
average execution duration should remain the same. What's happening is
that it is getting called twice just in the cache call, and only one
of those has the new key `:metrics/ignore-execution-time`. This
remains a mystery to me

* Don't call completion arity from step arity

the `(rf acc)` call violated the terms of transduction. If it does
need to be there, perhaps it could be used with reduced

wary of #12207 (which fixes #12165) and seeing if that crops up again.

- i had introduced :metrics/ignore-execution-time but this was because
i thought that we were running a simple query (limit 1) to get col
metadata. Now that i know we were just being a bit careless with `(rf
acc)` this extra information is no longer needed, and we can just
react to cached data.

* hoist closing paren

* Move *ignore-cached-results* to middleware

Had tried to adjust the binding to work, but the binding ended before
all of the bound functions and such were submitted to work, so when
that code actually executed the binding was no longer in
place. Annoying and it should be in middleware anyways

* Remove now unused import (cache became keyword in middleware not binding)

I thought the collections appearing in the list page problem was solved, but this turned out to be a bug with the ttest. Upon actual inspection the lists remained.

To fix this without changing all sorts of crap that depended upon collections api contract (we futzed with the output but the futzing of the output is transparent to the frontend because of the entity thing they got going on), needed a way to elicit null response from collections endpoint. shoved one in. then renamed it because the first name, "dummy", makes it all sound kind of jank, so 'no-model' for explicitly asking for no model without handing in an empty set, which turns into the set of all models.

This is an 80% unjankening. Full unjankening would entail whacking this special case.

* Fix can't unarchive question without data access

* Test API checks perms when unarchiving a card

Fix typo in DS name for app DB

Add test assertion

* Extract largish sort-order function

also, running tests locally could run into timezone issues. only
interested in comparisons (<) so just make sure that its 24 hours in
the past so no issues with time zones

* tests are pretty literal but just satisfying the test checker

* bump coverage from 71 to hopefully 75%

* Don't compute only on non-zero scores

I was removing a helper function when refactoring the scoring. That
had a filter on :score?. I'm not sure why i translated this to only
interpret the denominator of scoring to only include scores that were
positive. This is obviously wrong.

Also, previously, all databases would be returned in the archived
query which seems obviously wrong. Databases cannot be archived and so
should never appear when searching for archived items.

* Add some tests

- had been asserting that ee-score = os-score if not in an official
collection. This is actually hard to do under the current scheme, and
actually not important. The important bit is that the relative
ordering is the same. It had been adding 0 score 2 weight previously,
and removing all scores with a score of 0. This created a bad way to
score, as something with 1/2, 0/3, 0/4, 0/10 (net score 1/2=0.5) would
appear higher than something with 1/2, 1/3, 1/4, 3/10 (net score
6/19=0.315) and it really shouldn't. The former really should
be (1/19=0.05). And since we are adding in a denominator of
2 (official collection weight) all of the scores are actually smaller,
but that's ok, as they are all proportionately smaller.