On the topic view route we query for reviewables of each post in the stream,
using a query that filters on two unindexed columns. This results in a Parallel Seq Scan
over all rows, which can take quite some time (~20ms was seen) on forums with lots of flags
After index is added PostgreSQL planner opts for a simple Index Scan and runs in sub 1ms.
Before:
```
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Finalize GroupAggregate (cost=11401.08..11404.87 rows=20 width=28) (actual time=19.209..19.209 rows=1 loops=1)
Group Key: r.target_id
-> Gather Merge (cost=11401.08..11404.41 rows=26 width=28) (actual time=19.202..20.419 rows=1 loops=1)
Workers Planned: 2
Workers Launched: 2
-> Partial GroupAggregate (cost=10401.06..10401.38 rows=13 width=28) (actual time=16.958..16.958 rows=0 loops=3)
Group Key: r.target_id
-> Sort (cost=10401.06..10401.09 rows=13 width=16) (actual time=16.956..16.956 rows=0 loops=3)
Sort Key: r.target_id
Sort Method: quicksort Memory: 25kB
Worker 0: Sort Method: quicksort Memory: 25kB
Worker 1: Sort Method: quicksort Memory: 25kB
-> Nested Loop (cost=0.42..10400.82 rows=13 width=16) (actual time=15.894..16.938 rows=0 loops=3)
-> Parallel Seq Scan on reviewables r (cost=0.00..10302.47 rows=8 width=12) (actual time=15.882..16.927 rows=0 loops=3)
Filter: (((target_type)::text = 'Post'::text) AND (target_id = ANY ('{7565483,7565563,7565566,7565567,7565568,7565569,7565579,7565580,7565583,7565586,7565588,7565589,7565601,7565602,7565603,7565613,7565620,7565623,7565624,7565626}'::integer[])))
Rows Removed by Filter: 49183
-> Index Scan using index_reviewable_scores_on_reviewable_id on reviewable_scores s (cost=0.42..12.27 rows=2 width=8) (actual time=0.029..0.030 rows=1 loops=1)
Index Cond: (reviewable_id = r.id)
Planning Time: 0.318 ms
Execution Time: 20.470 ms
```
After:
```
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
GroupAggregate (cost=0.84..342.54 rows=20 width=28) (actual time=0.038..0.038 rows=1 loops=1)
Group Key: r.target_id
-> Nested Loop (cost=0.84..341.95 rows=31 width=16) (actual time=0.020..0.033 rows=1 loops=1)
-> Index Scan using index_reviewables_on_target_id on reviewables r (cost=0.42..96.07 rows=20 width=12) (actual time=0.013..0.026 rows=1 loops=1)
Index Cond: (target_id = ANY ('{7565483,7565563,7565566,7565567,7565568,7565569,7565579,7565580,7565583,7565586,7565588,7565589,7565601,7565602,7565603,7565613,7565620,7565623,7565624,7565626}'::integer[]))
-> Index Scan using index_reviewable_scores_on_reviewable_id on reviewable_scores s (cost=0.42..12.27 rows=2 width=8) (actual time=0.005..0.005 rows=1 loops=1)
Index Cond: (reviewable_id = r.id)
Planning Time: 0.253 ms
Execution Time: 0.067 ms
```
Legacy Firefox Android has some quirks around vibration where it:
- asks for permission
- doesn't persist the permission
This makes the default like vibration popup a permission on Firefox
Android <= 68.
This isn't the case (yet?) on their new Firefox which is rolling out
worldwide right now.
I'd say we merge this now and revert in 3 months or so when
https://arewefenixyet.com/ shows a full rollout.
Behavior was changed in #9966, which made the URL be relative.
If the user landed in a topic, for example, the browser was given a
service worker URL under that specific topic URL, which was a 404.
Fixes broken PWA install and broken push notifications
Follow up to 3b0fa9b using the new icon URL introduced in e618efe
Will automatically use icons of the primary color of the active theme
for the user during install.
* FEATURE: Support for App Shortcuts Menu
This adds a list of shortcuts to a installed Discourse instance.
It can be accessed by right clicks or long press on the app icon.
See https://github.com/MicrosoftEdge/MSEdgeExplainers/blob/master/Shortcuts/explainer.md
List of possible follow ups include:
- Making it admin customizable
- Making it user customizable
- Using SVG icons from the site icon sprite
- Picking an accent color for icons
* FIX: Add type to shortcut menu icons
The PR https://github.com/rails/rails/pull/36612 changes the raised exception
if the error message includes the target database name.
Since the error message contains the hostname, this could be triggered when
the hostname contains the database name.
Some providers don't implement the Expect: 100-continue support,
which results in a mismatch in the object signature.
With this settings, users can disable the header and use such providers.
DO does not implement tagging support for S3 objects. Removing our default
empty tag fixes compatibility.
The expire_missing_assets rake task can't be used with that service still,
but this patch allows normal operation.
Moves the new draft creation concurrency handling to PostgreSQL
so the database doesn't error out when the draft is being created
by multiple backends.
Also removes `retry_not_unique` parameter from Draft#set` which is
not called anywhere.
Also fixes a draft update not bumping the `updated_at` column.
Now that the spec is finished use the unprefixed API, which was also moved
from window to navigator.
Still uses feature detection so it fail gracefully when not available in
the user agent.
* DEV: Update Bundler
Latest RubyGems 3.1.1 vendors bundler 2.1.0 *again*. And our base
image build system even updates it to 2.1.1.
After that it is unable to run a simple `bundle install` because of
version mismatch.
Updating bundler to the one that comes with our enforced Ruby version
solves this.
* DEV: Update bundler in CI too
This PR aims to make poll results easily exportable to staff in a CSV format, so they can be analyzed in external software.
It also makes the export data easily customizable by allowing users to leverage any data explorer query to generate the report. By default, we use a query that ships with data explorer, but user can change the ID in settings or use 0 to disable this feature.
One potential upgrade is using the recent work that allows arbitrary group to run data explorer and allow all the groups with access to the configured query to also export polls, but that can be added later.
Co-Authored-By: Joffrey JAFFEUX <j.jaffeux@gmail.com>
Trying to truncate encoded slugs will mean that we have to keep the URL
valid, which can be tricky as you have to be aware of multibyte
characters.
Since we already have upper bounds for the title, the slug won't grow
for more than title*6 in the worst case. The slug column in the topic
table can store that just fine.
Added a test to ensure that a generated slug is a valid URL too, so we
don't introduce regressions in the future.
* FIX: Do not encode the URL twice
Now that we encode slugs in the server we don't need this anymore.
Reverts fe5na33
* FIX: More places do deal with encoded slugs
* the param is a string now, not a hash
* FIX: Handle the nil slug on /categories
* DEV: Add seeded? method to identity default categories
* DEV: Use SiteSetting to keep track of seeded categories
When an admin changes the site setting slug_generation_method to
encoded, we weren't really encoding the slug, but just allowing non-ascii
characters in the slug (unicode).
That brings problems when a user posts a link to topic without the slug, as
our topic controller tries to redirect the user to the correct URL that contains
the slug with unicode characters. Having unicode in the Location header in a
response is a RFC violation and some browsers end up in a redirection loop.
Bug report: https://meta.discourse.org/t/-/125371?u=falco
This commit also checks if a site uses encoded slugs and clear all saved slugs
in the db so they can be regenerated using an onceoff job.
Advanced trigger is currently broken on:
ca
es
et
fr
he
it
pt_BR
And that is because the translation levels for the plugin are kinda low, so I would guess it's broken for half the languages.
Since we have only two tracks for a while now, a quick fix to me is inverting the selectors.
This patch works because the advanced key is "larger" than the new user one.
The current manifest validation in Chrome requires at least one
non-maskable icon to make it installable.
This commit adds the maskable entry as another entry, following up
on changes added in 3e590b8
Because:
- Chrome 78 and Firefox Fenix have support for it
- The icons will look better by using all the availiable space
- Admins can control the icon and add appropriate padding since we have a
dedicated asset for the manifest logo. Read more about it on
https://css-tricks.com/maskable-icons-android-adaptive-icons-for-your-pwa/
This commit:
- Adds a new key under the icons array in the webmanifest, named purpose
with maskable value.
* FEATURE: Incorporate PWA install prompt into Discourse UI
This is mainly done so Discourse forums stop nagging people to install
on the very first visits to a website.
We will prevent the native install "mini-info" bar from ever appearing,
capture the event that pops with it, and delay it until the user meets
our criteria, which currently is trust_level 1.
If the event happens and the user meets our criteria we show a Discourse
alert banner proposing the install to the user. Dismissal of the banner
is recorded so the user ins't bothered anymore on the same device.
Co-Authored-By: Gerhard Schlager <mail@gerhard-schlager.at>
Co-Authored-By: Joffrey JAFFEUX <j.jaffeux@gmail.com>
Introduces isIpadOS to our capabilities sniffer, which is currently
compatible with latest iPad beta OS. Older iPad versions will match our
isIOS detection.
Using this new feature, fixes the service worker being blocked in Google
Chrome, Firefox and other browsers while running on Mac OS, where they
work just fine. This enables Desktop PWA install in Mac OS.
Only restricting cache per age wasn't enough for instances with lots of
multimedia usage and high number of posts.
MaxEntries is also more effective on cleanup, and purgeOnQuotaError
advertise that Discourse cache can be purged if necessary.
https://developers.google.com/web/tools/workbox/guides/storage-quota
There is a bug that when Safari starts up, and reloads the tabs from
the previous session **and** there is a service worker registered for
the scope of the document, all cookies marked as `SameSite=Lax` won't be
sent in the request.
This puts Discourse in a **very** broken state, where:
- You appear as a anon user
- Subsequent xhr requests will come with logged in data
- Refreshing doesn't log you in (cookies are still not sent)
- Clicking on the address bar and hitting enter, will log you in (as it
will finally send those damn `SameSite=Lax` cookies.
Looks a lot like a corner case missed by the fix at
https://trac.webkit.org/changeset/241918/webkit
Adds `DISCOURSE_MESSAGE_BUS_REDIS_ENABLED` env var, that when set
to true, will allow Discourse to connect to a different redis
instance for MessageBus needs.
When enabled you can configure the same env vars user for redis,
but prefixed by `MESSAGE_BUS`, eg:
`DISCOURSE_MESSAGE_BUS_REDIS_HOST`
On busy sites, concurrent requests to insert into post_timings can
occur, which was dealt with using Ruby exceptions.
This moves the handling to PostgreSQL which makes it a bit faster,
and prevents a spam of ERROR in the database logs.
This will allow users installing a Discourse PWA to use their active
theme colors on the generated app. Thanks for @mgiuca for the tip.
Also makes the share_target config explicit to silence Chrome warnings
This adds a **very basic** support for share to Discourse.
Currently, this is only supported in Android + Chrome 71+.
After installing a Discourse site to the Home Screen, you will be
able to share from anywhere in the OS to the Discourse site.
Discourse will use the title and text from the share event.
- Remove share target because the spec is changing
- Allow any orientation again because natural is too restrictive
- Use correct file and mime types for the manifest
This makes Discourse compliant with latest Google PWA requirements,
so we get the App Install banner back.
Should bump our Lighthouse PWA Audit score to 11/11.
