fix: remove the 750ms LiveView latency simulator in dev - #3986
Merged
Merged
Conversation
chasers
added this pull request to stack #3983
September 14, 2026 19:29
chasers
force-pushed
the
fix/stack-dev-latency-sim
branch
from
September 17, 2026 19:30
73e2038 to
6bc6718
Compare
chasers
force-pushed
the
fix/load-more-review
branch
2 times, most recently
from
September 17, 2026 19:50
3f42a01 to
670dd85
Compare
chasers
force-pushed
the
fix/stack-dev-latency-sim
branch
from
September 17, 2026 19:50
6bc6718 to
df58f52
Compare
amokan
approved these changes
Sep 17, 2026
The button drove its spinner from `phx-click-loading`, which LiveView applies
only while the event round trip is in flight. `load_events` hands the query to
`SearchQueryExecutor` and returns immediately, so the spinner stopped in
milliseconds while a BigQuery page query still took seconds.
`EventPagination` now carries a `loading_intent` and a fourth button state,
`:loading`. `load_events` marks the intent, and the page result and page error
handlers clear it. The button renders its spinner from that state. A hidden
button stays hidden, and `disabled={@State != :ready}` already covers
`:loading`, so a second click cannot queue behind the first.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
A dropped page request was silent. The `with` fell through to
`{:noreply, socket}`, and the catch-all clause did the same. A request that the
guard or the cursor validation rejected looked like a page with no rows.
Both paths now log the intent, the cursor values and the guard state.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
Every pagination test was Postgres-only, and `apply_timestamp_filter_rules/1` and `cursor_condition/3` both take a different clause for BigQuery. A report that the top button loads nothing came from a BigQuery source, and nothing covered that path. The new test stubs `bigquery_jobs_query`, returns a full page plus a sentinel for the ranged query and a page of older rows for any query carrying `TIMESTAMP_MICROS`, then clicks the top button and asserts both that the page query ran and that the list grew by a page. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YQ8Y8zCvSKUFNPvkwH2WS4
A page request drops the query's own timestamp range so it can page past it, and kept only a one-sided partition bound. The scan therefore ran to the end of the table: on BigQuery that is every partition the source has ever had. A page now moves by the width of the range in view, bounded below at 60 seconds. The query carries both a `TIMESTAMP_MICROS` bound and a partition-date bound on each side, for `:timestamp`, `:pseudo` and Postgres alike. The LQL range grows by that same window on every click, rather than only when the rows that came back happened to fall outside it, and an empty page shifts the cursor by the window instead of clearing it. A quiet stretch no longer stalls the button. `extend_timestamp_range/2` and `event_timestamp/2` are gone with that change. The buttons name the window they move by: "Load more (-10 minutes)" and "Load more (+10 minutes)". Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YQ8Y8zCvSKUFNPvkwH2WS4
Dialyzer fails the typing check on `event_timestamp_datetime/1`. The only caller passes integer microseconds, so the `DateTime` and `NaiveDateTime` clauses can never match. Convert the integer directly. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
A chart click already scrolled the page to the newest results, but running a search did not, so submitting a query left the viewport wherever it was. `start_search` now sets the same `scroll_to_bottom_on_result?` flag that `datetime_update` uses, so the scroll happens once that search's first event page lands. The search button and the Enter key both route through one event: `lql_editor_wrapper_hook.js` binds `monaco.KeyCode.Enter` and the submit listener to the same `submitSearch/0`, which pushes `start_search`. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YQ8Y8zCvSKUFNPvkwH2WS4
A chart click scrolled the page to the newest results, but a search did not. Scrolling was also split between the server and a `data-tailing` attribute that the hook read on every update. A fresh load with a querystring never scrolled. The follow-up diffs of a search (aggregates, pagination buttons) each ran `restoreScrollAnchor/0` and undid the scroll. - Every initial event page pushes `scroll-to-bottom`. The initial page is always the newest events in range. - Tail appends push `scroll-to-bottom` too, so `scrollToLatest/0` and `data-tailing` come out. - The hook holds a pending scroll until a frame runs it. It skips the anchor restore until then. The server owns scroll intent. The hook owns viewport stability, which needs element geometry that the server cannot see. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
A query with no `t:` filter now gets one on the first page request. The range is the one the chart already draws. The aggregate query charts `default_period_tick_count/1` periods back from now, so `c:group_by(t::minute)` means the last 120 minutes. The page window also uses that implied range, not a fixed 60 seconds. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
Paging moves by a fixed window now, so a page that comes back short means the window was quiet, not that the source has nothing more. The visibility rules still came from the old sentinel model and read a short page as the end of the line. `page_cursors/3` takes the previous cursor from the rows it returned instead of from the 101st sentinel row, so a quiet window no longer clears the cursor and takes the top button away. `next_exhausted?` leaves `EventPagination` along with `complete_initial/2`, `complete_page/3` and `complete_tail/2`, which existed only to maintain it, and `next_page_available?/4` leaves the LiveView: a range that ends in the future is fine and should not hide the button that would move it further. A button now hides in exactly two cases: the view is tailing, or it has no cursor at all. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YQ8Y8zCvSKUFNPvkwH2WS4
A short page now means the window was quiet, not that the source has no older events. The e2e test still expected the button to disappear. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
- Store one page window per search in `EventPagination`. The label, the query, the cursor shift and the range growth all read it. The window no longer doubles on each click. - Set the page window from a tail result too. `resume_tailing/1` resets the pagination, and the result that follows takes the tail path, so a soft pause, play and pause left the window nil and dropped the next page request. The window belongs to the search in view, not to the first page. - Build the chart range in the search timezone. An implied range is no longer UTC written as local time. - Keep a one-sided `t:` bound when a page request makes the range explicit. - Cap the cursor of an empty "next" page at the request time. - Clamp a "next" range extension at now. Each empty click used to push the range max another window into the future and re-run the aggregate. - Drop a page result or page error that no request waits for. - Clear the page spinner when a search fails to parse. - Disable the other button while a page request runs. - Remove the unused sentinel row: the extra fetched row, `fetch_limit/0` and the `has_more?` field that nothing reads. - Remove dead clauses, a duplicate alias and inline comments. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
The dev root layout called `liveSocket.enableLatencySim(750)` on every page load. LiveView then held back every client message by 750ms, so every click and form change in dev waited 750ms. The server was never slow. This is the same change as #3901. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL
chasers
removed this pull request from stack #3983
September 17, 2026 21:09
chasers
force-pushed
the
fix/stack-dev-latency-sim
branch
from
September 17, 2026 21:09
df58f52 to
10c699d
Compare
This was referenced Sep 17, 2026
This was referenced Sep 17, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack 9/9 — "Load more" pagination fixes.
liveSocket.enableLatencySim(750)on each page load.sessionStorage. RunliveSocket.disableLatencySim()once in the browser console, or open a new tab.🤖 Generated with Claude Code
https://claude.ai/code/session_01YBPF1knvgfghTSdxuYryGL