* Fix collections/graph stack overflow

* Test fixes 

 [ci postgres] [ci mysql]

* Show engine in Task trouble shooting logs (#13608)

* Show engine in Task trouble shooting logs

in an effort to get better feedback when people raise issues about
long running sync processes, include the database engine in the logs.

Also wrapped `th` in a `tr` inside the `thead` to remove react warning

* Const intead of let on the db_id_to_engine

* Swap db id for db name

* Move comment out of jsx

* fixup! Merge branch 'release-x.37.x' into fix-collections-graph-stack-overflow

Co-authored-by: dpsutton <dan@dpsutton.com>

Avoid ansi output from log4j2 so that the environment variable NO_COLOR
behaves as expected.

Even though this Appender wasn't referenced anywhere, it still
initialized. Commented out this appender so it can be used as an
example, but it won't create a folder called ${logfile.path} now

Resolves #13422

We should not be logging to a file by default, this was an oversight in
the migration to log4j2

Resolves #13422

* Upgrade to log4j 2.13.3

Upgrades us to newer Log4j

Resolves #12719

* fixing reflection warnings

* reflection warning on test

* make tests work

took a while to figure out how to configure log4j properly
programatically - it's not as easy as log4j 1

add logging adapters for commons-logging, slf4j, liquibase, etc so that
all logging routes through log4j2 now

* appease the linters

* okay cljr-clean-ns makes this line too long, I get it

* add log4j2.xml config for Cypress tests

* update log4j configuration file documentation

update which flag is used by the backend tests

[ci all]

* update cheshire because clj-http updated

* Wire up the metabase-appender to Log4j2

This was brutal to figure out - log4j usually silently fails, so it took
finding the right combination of configuration settings and system
properties to get it to log enough to find out why the appender was
never invoked.

Changes the app db to lowercase all emails, and queries to default to the lowercased version of emails. This fixes the problem of users logging in with emails where one or more characters could be capitalized. Resolves #3047

* add new illustration and update /login layout

* extract common layout and apply

* remove old css

* add 1 and 3x images

* adjust image position based on screen size

* adjust image bottom

* optimize bridge pngs

Co-authored-by: Paul Rosenzweig <paul.a.rosenzweig@gmail.com>

* Test fix 

 [ci mysql] [ci mariadb] [ci postgres]

* object-defaults should always remove created_at/updated_at
[ci mysql] [ci mariadb] [ci postgres]

Co-authored-by: Cam Saul <github@camsaul.com>

* Remove Table.rows entirely

* Lots of test updates/modernization

* Test fixes 

Co-authored-by: Cam Saul <github@camsaul.com>
Co-authored-by: Maz Ameli <maz@metabase.com>

* Support substituting SQL snippets in queries

* Rename SQL snippets to native query snippets

* Add migration and model for native query snippet table

* Add API endpoints for native query snippets CRUD

Tests are incomplete. Consequently the API is not well tested.

* Cosmetic: Docstring typo and indentation

* Remove pre-delete handler that prevents deletion

It breaks in cases where deletion is reasonable, like during tests. No
DELETE API endpoint will be added, so no it shouldn't be necessary to
add any additional safeguards.

* Fix `:database_id` field reference

And remove unused import.

* Test snippets listing via API

* Add more API test stubs

* Refactor snippets listing endpoint for improved readability

* Align NativeQuerySnippet model behavior with Segment model's

* Clean ns

* Fix and test individual snippet read via API

* Fix snippet name schema

* Test creation of snippets via API

* Test update API endpoint for native query snippets

* Handle query snippet model in H2 dump/load commands

* Port load-from-h2 tests to clojure.test

* Deduplicate models list used to dump/load H2 dbs

The failing load-from-h2 test that checks that all models are accounted
for alerted me that I need to handle the new `NativeQuerySnippet` model.
However, there was no such test for dump-to-h2. Instead of adding such a
test, I opted to simply use load-from-h2's list in dump-to-h2 to
eliminate the problem, rather than testing for it.

* Dump/load collections before models that depend on it

* Linter appeasement

* Use snippet ID and content in query parameter record

* Implement snippet lookup for template tags with ID

* Basic FE support for snippet template tags

* Use shorter "snippet" prefix instead of "native-query-snippet"

* Use simpler `tru` instead of `str` with `deferred-tru`

* Allow whitespace between snippet prefix and name

Also accept any sequence of characters after prefix.

* Don't allow snippet names to start with whitespace

Whitespace is allowed between the `snippet:` template tag prefix and the
snippet name. That whitespace is trimmed out when extracting the snippet
name.

* Add database ID to snippet template tags before expanding parameters

* Ensure that snippet db matches query db

* Avoid setting `:template-tags` for native queries without it

* Factor out snippet database ID validation

* Test snippet database ID validation

* Avoid validating nil snippets

* First pass at nested template tags in snippets

* Pass parent query parameters on to nested sub-queries

* Remove unused let-binding

* Added disabled test for reading query param from template tag

* Temporarily disable failing test

* Remove half baked nested snippet support

* Snippet model perms = native query perms

* Rename param k→ks for clarity

* Remove superfluous `some?` call

* Use utility functions to reduce cruft

* Test expansion of multiple snippets in (sub-)query

* Rather use testing utils from metabase.test

* Check for snippet changes only in provided fields

* Deduplicate field checks, with added test case output

* remove unused import from bad merge

Co-authored-by: Paul Rosenzweig <paul.a.rosenzweig@gmail.com>
Co-authored-by: Paul Rosenzweig <paulrosenzweig@users.noreply.github.com>

* Backported fixes

* Explicitly include data.xml as a dep

* Use OffsetDateTime instead of org.h2.api.TimestampWithTimeZone [ci mysql]

* Revert JUnit tweaks -- don't seem to be working correctly

Co-authored-by: Marcelo Gimenes de Oliveira <marcelo.gimenes.oliveira@gmail.com>

Related to issue https://github.com/metabase/metabase/issues/2808

Add :visibility option to defsetting which is used to determine which settings to load on the frontend. Port/rename existing public settings.

* update translations for included languages

* Add i18n CI step from master

* fix newlines in it.po

* remove \n from nl.po

* remove \n from pl.po

* add Slovak and update others

* remove newlines

* update locales.clj

* reset config.yml

* remove Google Fonts usage

* local Lato in resources/frontend_client/app/fonts

* create assets/fonts aliases, update publicPath, use absolute paths in css

* add svg to extensions

* update flow config to account for frontend_client assets

* add explicit .svg extension to no_results usage

* remove google fonts references from CSP headers

* Avoid using `undefined` array when column formatters not specified

This led to `TypeError: Cannot read property "forEach" from undefined`
errors when no formatters were specified.

* Improve logging of unexpected errors during format compilation

* Log unsupported formatting operators and return a dummy operator fn

* Linting fixes

* Manually ES5-ify code that is to be executed by the Nashorn engine

Nashorn in JDK 8 only supports ES5.1.

* Update Rebuilt shared JS from `yarn build-shared`

* Move formatting foctory fns into an operator-indexed object

* Linter appeasement

* Test that all defined operators are supported in table formatting

* Flow annotations

* Return `null` in stead of `undefined`

Formatting/style changes snuck in.

* `yarn build-shared`

* Use operators from table formatting and not column filters

* Compute Content-Security-Policy hashes for inline JS (#10504)

* Split out inline JS from index/init templates to separate files

* Read inline JS from resources at run time

* Calculate inline JS hashes for CSP header from content

* Move inline JS to resource sub-directory

* Update and memoize inline JS loading

* Revert debug code

* Deduplicate `resp/response` calls

* Fix paths to moved inline JS resources

* Force creation of test data DB so things don't get left in the cache

This fix was provided by @camsaul.

* Combine and `defonce` inline JS hashes

* s/inlinejs/inline-js/

* Ensure that Metabase JAR is readable by `metabase` container user (#10510)

If the umask of the user that builds the image is xx7, the JAR will not be
world readable in the resulting image.

* Remove `--add-modules=java.xml.bind` Java option (#10521)

Fixes #10244. @camsaul confirmed that it is no longer needed.

* Join on collection table's ID, not the joining table's `collection_id` (#10481)

* Join on collection table's ID, not the joining table's `collection_id`

This caused each record that joins on collection, to be duplicated for each
existing collection:

`LEFT JOIN collection collection ON card.collection_id = collection_id`

should be

`LEFT JOIN collection collection ON card.collection_id = collection.id`

* Test fixes from @cam on EE

* Insights: correctly handle infinities in results (#10502)

Filter out models with infinities

* Transfer log events as JSON objects and allow filtering on front-end (#10522)

* Return log events as JSON objects from API; render on front-end

Site UUID added to log events.

* Remove superfluous memoization of site UUID retrieval

Settings are already cached.

* Transfer timestamps in ISO-8601 format

* Move `metabase.metabot.instance/local-process-uuid` to `metabase.public-settings`

* Use `local-process-uuid` in stead of `site-uuid`

`site-uuid` is not unique across processes, which is what we want to
associate with log events.

* Remove unused import

* Allow log events to be filtered by process UUID on front-end

* Move process selector out of loading wrapper

Otherwise, if there are process UUIDs with no log events, the selector is
removed with the log event list.

* Figured out how to add docs to `defonce`

* Fix ns declaration

* Linter appeasement

* Merge log events received from back-end with previously received ones

This allows responses from multiple back-end instances to be spliced
together.

* Use MB components in stead of plain, styled HTML tags

The select is only displayed if more than one process UUID was found.

* Linter appeasement

* Move `metabase.public-settings/local-process-uuid` to `metabase.config`

* Fix sorting on multiple log event fields

* Flatten `Select` children to allow nested arrays of children

* Generate options directly under parent Select

* Limit the number of log events that we render

* Docstring and style updates

* Update editor config with Clojure files max line length

* Ensure that `NumericWrapper` is imported before use

* Cleanup layout

* Bump HoneySQL version (#10559)

The new version (0.9.5) includes @camsaul's Turkish upper case fix:
jkk/honeysql#237

* Factor out test for access to localStorage (#10533)

* Factor out test for access to localStorage

This allows code that uses localStorage to test against
`window.hasLocalStorage`, in stead of handling exceptions generated when
access to localStorage is denied.

* Move check for localStorage to metabase/lib/dom

* Fix typo in warning message (#10556)

* revert text color to #74838f (#10560)

* revert text-medium color to #74838f

* use light instead of medium on preview button

* MBTL (#10105)

Add transforms and domain models

* Added page about maps and updated nav (#10495)

* Added page about maps and updated nav

* Clarified step 1

* Added more details throughout

* Quick URL change

* Added additional resource!

* fix typo: 'sdisplay' to 'display'

* Add message with link to docs for fields without any filter widgets (#10340)

* Correct proxy command line [ci skip] (#10430)

* Bump Toucan version to include Turkish lower case fix (#10581)

* Wrap `tru` and `trs` macro results in `str` (#10571)

* Add versions of `trs` and `tru` that wrap results in `str`

The pre-existing `trs` and `tru` were renamed to `lazy-trs` and `lazy-tru`
respectively, and new `trs` and `tru` funcs were added, wrapping the result
of its lazy counterpart in `str`.

This way the normal use case of `trs` and `tru` is improved by no longer
requiring that it be wrapped in `str`.

Cases where the translated result is passed to `str` anyway, the lazy
version can be used.

* `(str (trs ...))` → `(trs ...)`

And the same for `tru`.

* Allow `defsetting` descriptions to use lazy versions of `trs`/`tru`

* Use `lazy-trs`/`lazy-tru` where `str` is applied separately

* Remove unnecessary `vec` from `tru`/`trs` macros

* puppetlabs.i18n does not have a `lazy-tru`

* Missed some settings that now need to use `lazy-tru`

* Only use `lazy-tru` in top-level `def`s

If the localized strings are looked up on module import, before locales are
loaded, the lookup will fail and the original input string will be used.

* Fix bad usages of `tru` that should be `lazy-tru`

* s/lazy-tr/deferred-tr/g

* Avoid `tr*` macros being used in top-level (compile time) statements

Because localizations are not yet loaded at compile time.

* Add type hint for `str*` wrapping of `trs`/`tru`

* Fix assigned function's return type hint

* Expand docstrings for `i18n/tr[su]`

* Use `trs` and `tru` from metabase.util.i18n instead of puppetlabs.i18n.core

* Defer compile time translation lookups

These changes were pointed out by exceptions generated by `i18n/str*`.

* Remove unused `require`d function

* New compile time `trs` usages caught by `lein uberjar`

* Ensure that error message from db driver is a string

It is necessary to ensure that the error message is a string, because some
db driver errors come from statically declared error messages, using
`deferred-tru`. Like those in
`metabase.driver.common/connection-error-messages`.

Fixes #10347 and #10473.

* Remove string type hint from `humanize-connect-error-message`

* Don't disable preview for fields with non-trivial type  (#9687)

* Don't create API callback if no database is associated with card (#10573)

* Don't create API callback if no database is associated with card

Fixes #10538

* Return empty array to avoid mapping over null

[ci skip]

Add transforms and domain models

* Split out inline JS from index/init templates to separate files

* Read inline JS from resources at run time

* Calculate inline JS hashes for CSP header from content

* Move inline JS to resource sub-directory

* Update and memoize inline JS loading

* Revert debug code

* Deduplicate `resp/response` calls

* Fix paths to moved inline JS resources

* Force creation of test data DB so things don't get left in the cache

This fix was provided by @camsaul.

* Combine and `defonce` inline JS hashes

* s/inlinejs/inline-js/

Notebook Mode + View Mode + Joins + etc