Skip to content

workarounds.md

Latest commit

 

History

History
76 lines (57 loc) · 8.27 KB

workarounds.md

File metadata and controls

76 lines (57 loc) · 8.27 KB
Raw

Workarounds

Code that exists only because something upstream (JavaFX, a library, a tool) does not do what it should or could. Each entry says where the workaround lives, what upstream does, which upstream issue tracks it, and how to tell that it can go. In code, every workaround carries a Workaround W<n> marker naming its entry here, so git grep "Workaround W3" finds all of it; scripts/consistency.sh checks that markers and entries match.

What to report upstream — these bugs, others without a workaround, and feature requests — is collected in upstream.md.

When upstream fixes one: remove the code at every marker, delete the entry (the number is not reused), and note it in CHANGELOG.md if behavior changes.

Removed

  • W1 — TextArea does not size itself to its content (JDK-8091714, JDK-8181766): MessageText measured its text. Gone 2026-09-22 with the TextArea bodies; a bubble body is a RichTextArea with useContentHeight now (W5 is the narrower remainder).
  • W2 — TextArea's inner ScrollPane swallowed wheel events: BubbleView redirected them to the list. Gone 2026-09-22 with the TextArea bodies: a content-high RichTextArea passes wheel events on (WheelOverBodyUiTest passes without the filter).

W3 — RichTextArea draws styled segments with resolved styles, not the area's text fill

  • Where: chatpane.css, rules .chat-pane .chat-pane-transcript .message-text and .chat-pane .message-bubble .message-text (-fx-fill).
  • Upstream: a segment added with RichParagraph.Builder.addWithStyleNames is not a Text carrying those names: the area resolves the names through the stylesheet into styles and draws a plain Text with them (seen in the node tree: the shown Text has no style class). The area's own -fx-text-fill (Modena: -fx-text-inner-color) reaches no segment — a standalone check (2026-09-22) showed a plain segment and a style-named one both black under .rich-text-area { -fx-text-fill: red } — so without an explicit -fx-fill the text stays black, unreadable in a dark theme. Consequence for tests: find shown segments by their text, not by style class. No upstream issue found; jfx.incubator.richtext (26.0.2) is an incubator module, tracked in general by JDK-8351982.
  • Removable when: transcript and bubble text read light in the demo's dark theme (just demo --layout=irc --theme=dark) with the -fx-fill rules removed.

W4 — OpenFastTrace does not scan CSS

  • Where: the [impl->dsn~chatpane-stylesheet~…] tag sits on ChatPane.getUserAgentStylesheet instead of in chatpane.css.
  • Upstream: OpenFastTrace's tag importer has no .css file type. No upstream issue filed.
  • Removable when: OpenFastTrace imports tags from .css files; then the tag moves into the stylesheet's header comment.

W5 — RichTextArea cannot shrink to its text and still wrap

  • Where: BubbleText (computePrefWidth and the measuring Group of Texts the cell puts into the bubble).
  • Upstream: useContentWidth makes the preferred width the content width but stops wrapping — a long message becomes one cut-off line (probe 2026-09-22: preferred 694 px, shown 336 px, one line); without it, a short message gets a full-width bubble. The body therefore measures its widest line itself, with Text nodes that carry the spans' style names inside the bubble, so the stylesheet gives them the fonts the area's text gets. No upstream issue found; related: JDK-8310593 (useContentWidth/Height for scrollable controls), JDK-8344643 (RichTextArea).
  • Removable when: a RichTextArea offers "preferred width = widest line, wrap below the max width"; then BubbleText sets it and drops its measuring. Check: BubbleTextUiTest.bodyIsAsTallAsItsTextAndShrinksToShortText.

W6 — RichTextArea fails when a CSS pass sets its wrapping again

  • Where: TranscriptView and BubbleText set wrapText, useContentHeight, displayCaret and highlightCurrentParagraph in code; chatpane.css no longer sets them.
  • Upstream: with -fx-wrap-text coming from the pane's user-agent stylesheet, adding a stylesheet to the scene (the demo's dark theme) re-applies CSS, the property changes in the middle of that pass, and the area's VFlow.handleWrapText lays out right away: IllegalArgumentException: Children: duplicate children added: parent = Pane[styleClass=content] from VFlow.prepareCell (Carl, 2026-09-22; TranscriptStylingUiTest reproduced it). A standalone check with the wrapping in an author stylesheet did not fail, so the trigger needs the property to change during the pass. No upstream issue found; tracked in general by JDK-8351982.
  • Removable when: the settings move back into chatpane.css (so applications can style them) and TranscriptStylingUiTest passes, with no exception on the FX thread.

W7 — RichTextArea resolves style names once per cell and never again

  • Where: StyleProbe in ChatPaneSkin (styled from every lookup the text uses, in chatpane.css) and ConversationView.restyle() — the transcript calls RichTextAreaSkin.refreshLayout(), bubbles rebuild their cells.
  • Upstream: the area turns a segment's style names into styles when it builds the text cell (VFlow.resolveStyles measures a probe Text) and keeps the cell. A later CSS change — another theme, a stylesheet added — does not reach built cells (standalone check: a segment named tinted stayed red after the stylesheet switched .tinted to blue), and cells built before the pane's CSS was in place keep the defaults (Carl, 2026-09-22: the transcript "shows initially without styling" until the renderer is switched, which builds a new model). No upstream issue found.
  • Removable when: the area re-resolves styles on CSS changes itself; check: TranscriptStylingUiTest passes with the probe's callback doing nothing.

W8 — RichTextArea knows its content height only after its own layout

  • Where: BubbleText.computePrefHeight.
  • Upstream: with useContentHeight, the area's preferred height comes from the arrangement its VFlow builds in layoutChildren; before that first layout it reports Params.LAYOUT_MIN_HEIGHT plus insets (26 px in the probe), and once laid out it sets its preferred height and asks for another layout — a pulse later. The bubble list measures a fresh cell before its body ever laid out, and a growing answer builds its cell anew with every word: the newest bubble shrank to one line for a frame and grew back (66 px), so the whole list jumped by the difference with every word (Carl, 2026-10-05: "the scroll jumps crazy"). The body therefore lays itself out at its preferred width within the bubble before it reports its height — the width is -1 there, the area has no content bias. No upstream issue found; related: JDK-8310593 (useContentWidth/Height for scrollable controls).
  • Removable when: a fresh RichTextArea with useContentHeight reports its content height before its first layout; then the override goes. Check: FollowNewestUiTest.bubblesDoNotShrinkWhileTheNewestGrows.

W9 — RichTextArea scrolls to a caret that an edit elsewhere moved

  • Where: TranscriptView.removed.
  • Upstream: the selection model holds its caret as a model marker; an edit above the caret shifts the marker, the area sees a selection change, and VFlow.handleSelectionChange calls scrollCaretToVisible() — as if the user had moved the caret. Following the newest message parks the hidden caret at the end of the transcript, so deleting any message the user looks at scrolled the view to the end (Carl, 2026-10-05: "in irc and modern it still jumps to the end on delete"). Without a selection, the view therefore clears the caret before it removes; the next follow puts it back. No upstream issue found.
  • Removable when: a caret moved only by a model edit no longer scrolls the area; then removed stops clearing it. Check: FollowNewestUiTest.transcriptStaysWhenAMessageIsDeleted with the clearSelection() call removed.