Repository navigation
Conversation
Long variant lists, such as the 24 connector configurations, folded behind "show type options" and then scrolled as one long column. Lists with three or more variants now show one variant at a time, picked from a row of chips like the examples rail. From eleven variants the "more" menu gets a filter. The other variants stay in the page with hidden="until-found", so find-in-page and links to their fields switch the chip. The chip row moves out of the examples carousel into `chip-row.ts`. Both now use it, each with its own selectors, and the examples tests pass unchanged. The old fold rendered its `hidden` attribute HTML-encoded, so find-in-page never reached a folded list; it is gone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Docs preview (local build)Handbook preview: https://docs-v3-preview.elastic.dev/elastic/docs-builder/pull/4371/ |
Clicking a chip showed the variant's name again below the chips, such
as `{} bedrock_config`, and then a "show properties" toggle. Picking a
chip is already the choice to read that variant. Its fields now show
straight away, and the heading keeps only what the chip does not say:
the discriminator value, or the link to the type's own page. Lists of
two variants keep their names and toggles.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each variant before the picked one still took 8px: hidden="until-found" keeps an element's box, and with it the variant's padding. The further down the list a chip was, the lower its variant started. Hidden variants now take no padding. Near the bottom of the page, a shorter variant also made the browser clamp the scroll, which moved the chips under the pointer. The variant list now keeps the height that holds the scroll position. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Requesting changes due to a runtime UI regression: variant-switch scroll compensation can leave a stale min-height on the variants container, causing persistent extra whitespace after bottom-of-page switches. Please fix that behavior, then this looks ready for another pass.
What is this? | From workflow: PR Review
Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not.
| const maxScroll = document.documentElement.scrollHeight - window.innerHeight | ||
| const shortfall = scrolled - maxScroll | ||
| if (shortfall <= 0) return | ||
| list.style.minHeight = `${list.offsetHeight + shortfall}px` |
There was a problem hiding this comment.
[MEDIUM] Scroll compensation min-height is never released
keepScrollPosition increases list.style.minHeight when the page is near the bottom, but there is no path that clears that override on later switches. After one bottom-of-page switch from a taller variant to a shorter one, the union list can remain artificially tall and leave persistent blank space for the rest of the session.
A concrete way to reproduce is: scroll near the bottom, switch to a shorter variant (which sets minHeight), then continue switching variants or navigating within the same page — the stale minHeight remains applied.
Please clear stale compensation when it is not needed (for example, reset list.style.minHeight before recomputing and/or whenever shortfall <= 0).
There was a problem hiding this comment.
Fixed in 3d27df7. keepScrollPosition now clears min-height on every switch and sets it again only when that switch would make the browser clamp the scroll. In Chromium, near the page bottom, the switch to a shorter variant keeps 181px, and the next switch releases it. releases the height kept for the scroll position on the next switch in api-union-chips.test.ts covers it.
Chips showed each variant's schema name, so the 22 connector secrets all ended in `_secrets` and the detection rules in `RuleUpdateProps`. Chip labels now drop the words that every name in the list shares at its start or end: `bedrock`, `crowdstrike`; `Eql`, `SavedQuery`. Words split at `_`, `-`, `.` and capitals, and every label keeps a word of its own. The full name stays as the chip's tooltip and in the "more" menu, so the filter still matches it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
Requesting changes: there is still a correctness regression in union variant rendering where chip mode can produce a selectable but empty array-variant panel for X[]/X pairs in 3+ variant unions.
What is this? | From workflow: PR Review
Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not.
| // the label keeps only what the chip does not say (the discriminator value, or the link to the type's page). | ||
| var collapsible = variant.IsCollapsible && !Model.UseChips; | ||
| var folded = variant.UseHidden && !Model.UseChips; | ||
| var showName = !Model.UseChips || variant.PageUrl is { Length: > 0 }; |
There was a problem hiding this comment.
[MEDIUM] Chip mode can render array variants as an empty panel
In chip mode, showName is false unless PageUrl exists, so array variants in an X[]/X pair can lose their only visible content.
Concrete path:
ApiPropertyTreeBuilder.Variants.cssuppresses array-pair properties (showPropertiesfalse) and description (description = null) for theX[]side.- Here, chip mode also suppresses the label name when
PageUrlis absent. - Array variants also have
DiscriminatorLabel = null.
That leaves a selectable chip whose panel is blank for unions with 3+ variants that include an X[]/X pair (plus at least one other variant to trigger chips).
Please keep at least one visible payload for these variants in chip mode (for example, keep the variant name visible when there is no discriminator/page link, or avoid suppressing array-pair properties in chip mode).
There was a problem hiding this comment.
Fixed in 3d27df7. Behind chips each variant shows alone, so the X[] side of a pair no longer depends on the plain variant. When the list has chips, the builder keeps its fields and description, and the panel keeps its [] X heading so the fields read as each item's. A list of two variants still puts them on the plain variant only. BuildPropertyList_ArrayAndPlainPairBehindChips_BothShowTheFields covers both cases. No chip list on the current Elasticsearch and Kibana pages has such a pair, so no page changes from this today.
The height a chip list kept to hold the scroll position near the page bottom was never released, so blank space could stay after later switches. Every switch now starts from the list's own height. An X[] / X pair listed one after the other puts the fields on the plain variant only. Behind chips each variant shows alone, so the X[] chip opened an empty panel. It now shows the fields and description too, under its "[] X" heading. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
![infra-vault-gh-plugin-prod[bot]](https://avatars.githubusercontent.com/u/6764390?s=60&v=4){kind=small}
| <button | ||
| type="button" | ||
| role="tab" | ||
| class="api-example-chip@(i == 0 ? " is-active" : "")" | ||
| data-chip="@variant.AnchorId" | ||
| aria-controls="@variant.AnchorId" | ||
| aria-selected="@(i == 0 ? "true" : "false")" |
There was a problem hiding this comment.
Severity: P3 (Low)
Variant chips use tab semantics but lack the arrow-key and roving-focus behavior expected of tabs, making keyboard navigation inconsistent for keyboard and screen-reader users.
Generated by Libra. React with 👍 or 👎 to give feedback on this comment.
A row like `knn.filter` reads "One of: QueryContainer[] or QueryContainer". The `QueryContainer` option was styled as a link but was plain text, and `QueryContainer[]` was not even styled. Both now link to the `QueryContainer` page in the same tab, as the type of a row already does. The `[]` stays outside the link, so the link names the type. The Markdown export links them too. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ingest pipeline pages were almost all processor listings: 519 of `ingest-put-pipeline`'s 519 rows, repeated on `ingest-simulate` and `simulate-ingest`. `ProcessorContainer` now gets a page under Types, Ingest, as QueryContainer and the aggregations do, and those rows link to it. `ingest-put-pipeline` drops to 12 rows. Which types get a page lived in three hardcoded places, keyed by display name, so any schema whose short name was `Aggregate` linked to the Elasticsearch page. `TypePages` now lists each page once with its schema id, navigation category and map spelling, and rows link by `$ref` id. The name-based link styling goes too: it styled `QueryContainer` as a link on its own page and wherever it was not one. A type line built by `Describe`, such as a response type, now carries a real link instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
akira28
changed the title
…chips # Conflicts: # tests/Elastic.ApiExplorer.Tests/UnionVariantTests.cs
The chips read `bedrock`, `email`, but the "+N more" menu still listed `bedrock_config`, `email_config`. Menu entries now use the same short labels, with the full name as their tooltip, so the menu reads like the row it extends. The filter still matches the short names. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
| var leansOnPlain = variant.IsArray && hasBothVariants && variantsToRender.Count < ApiUnionVariants.ChipThreshold; | ||
| var showProperties = hasProperties && !leansOnPlain; |
There was a problem hiding this comment.
Severity: P3 (Low)
In chip-mode X[]/X pairs, building the array variant first can register a shared nested shape and make the plain variant a “Same fields as” link, removing its descendant anchors and breaking existing field links.
Generated by Libra. React with 👍 or 👎 to give feedback on this comment.
Lists switched to chips at three variants, so two object variants still listed both field sets one after the other while three showed one at a time. A list now uses chips whenever it has two or more object variants; a list with a single object, such as `TotalHits` or `number`, keeps reading one after the other. An X[] / X pair is two object variants, so the rule that left the array variant's fields to the plain one goes. A primitive variant behind chips showed an empty panel, since its name was on the chip and it has no fields. It now keeps its type name. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
akira28
changed the title
|
|
||
| [Test] | ||
| public async Task Render_PrimitiveVariantBehindChips_KeepsItsTypeName() | ||
| { | ||
| var html = await Render(3, objects: 2); |
There was a problem hiding this comment.
Severity: P3 (Low)
The primitive-label regression test uses an array variant, so it would still pass if a non-array primitive lost its only visible type name.
Generated by Libra. React with 👍 or 👎 to give feedback on this comment.
Lists with a single object, such as `TotalHits` or `number`, still listed their variants one after the other, so a union read one way or the other depending on its members. Every list with two or more variants now uses chips; a primitive variant shows its type name in its panel. The "or" separator between listed variants goes with the last lists that used it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
akira28
changed the title
This branch was successfully deployed
1 participant
API Explorer variant lists with two or more variants now show one variant at a time, picked from a row of chips like the examples rail. From eleven variants the chips' "more" menu gets a filter. On the Elasticsearch and Kibana pages, 646 of 660 variant lists, on 237 pages, get chips; 13 of them get the filter, among them the 24 connector configurations.
Affects:
API reference,Site UIPrompt summary: The author asked to reuse the examples rail's chips, with their "more" menu and search, for long lists of union variants. Lists of 3–10 variants get chips, and lists of 11 or more get chips with a filter. After review, the author asked for chips from two variants, primitives included, so that every union reads the same way.
Why
A variant list with more than two variants folded behind "show type options" and then opened as one long column. A reader looking for one connector scrolled through 24 of them. The fold also rendered its
hiddenattribute HTML-encoded, ashidden="until-found". Browsers read that as plainhidden, so find-in-page never reached a folded list.What
Chips pick the variant
A list with two or more variants renders a chip per variant above the variants and shows the first one. Clicking a chip, or an entry in the "+N more" menu, shows that variant. A chip drops the words every name in its list shares, so the connector secrets read
bedrock,crowdstrikerather thanbedrock_secrets,crowdstrike_secrets; the "more" menu uses the same short labels, and the full name stays as the tooltip on both. Each menu entry also names the discriminator value, such astype: eql, so the filter matches it too. The picked variant shows its fields at once, with no "show properties" step. Its heading drops the name the chip already shows and keeps only the discriminator value or the link to the type's own page. This includes a list such asTotalHitsornumber: a primitive variant keeps its type name, which is all its panel has to show. Only a lone variant renders without chips, so the "or" separator between variants is gone. AnX[]/Xpair is two object variants, so each side carries its own fields.Hidden variants stay reachable
The other variants stay in the page with
hidden="until-found", and their anchors are unchanged. Find-in-page reveals a match and switches the chip. A link to a field inside a variant that is not picked switches every chip list around it, on operation and schema pages. This includes the case where the browser already revealed the variant before the script ran. The Markdown export still lists every variant.One chip row for examples and variants
The packing, the "more" menu, its filter and its keyboard handling move from
api-examples-carousel.tsintochip-row.ts. Each caller passes the selectors of its own markup. The examples rail's tests pass unchanged. The menu now also moves back against its button when the filter shrinks it, on both.The old fold is gone
The "show type options" markup, its two click handlers, its
beforematchbinding and its CSS are removed, along withShouldCollapseand the union list'sUseHiddenUntilFound.Where to look
Every form a union takes in the API Explorer, on this PR's preview. A request body may start collapsed; open it to see the row.
track_total_hitsreadsboolean | numberknn.filter: "One of: QueryContainer[] or QueryContainer"rescore: "A single Rescore or an array of them"hits.total:TotalHitsornumbermgetresponsedocs:GetResultorMultiGetErrorindices.query:string,QueryContainer,RoleTemplateQuerytype: azure,type: gcsEql,Query,SavedQuerybedrock,email, …index_configentriesof a trusted-apps exception itemmore_like_this.like: "Like or Like[]; each Like is one of:"incident: "Requires at least one of: correlation_id or externalId"error.caused_by: "Same fields as" an earlier listingAggregate, 73 variantsVerify
In a 1,322-page comparison against
main, every text change is the "show type options" toggle giving way to chips, or a removed "or" separator.🤖 Generated with Claude Code