shinyTree 1.0.0

Fixes from a review of the whole package

• input$tree now holds the tree as soon as it loads, and is updated when the server sends a new tree with renderTree() or updateTree(): it used to stay NULL until the user clicked, and kept a replaced tree – and its selection of nodes that no longer exist – until then. A tree is reported once, and only if it differs from what input$tree holds, so an observer of input$tree that sends the tree back still does not loop (#36).
• A tree re-rendered after an error in its render function – a req() or validate() that fails and then passes – no longer stops updating input$tree. jsTree’s teardown removed the input binding’s event handlers.
• Trees whose id contains . (or another character with a meaning in CSS selectors) now work. Their options, checkboxes and search were ignored and input$tree was never set, and with types the page threw an error that stopped every tree on it from reporting.
• updateTree(), expandTree() and collapseTree() on a tree that has not been shown yet – in a tab not yet opened – are applied once it is, rather than throwing an error in the browser.
• Deleting the last node with the context menu now clears input$tree; it kept the deleted nodes.
• A node attribute such as sttype_label or sticonColor is no longer taken for sttype or sticon: attributes were matched by prefix. The same applied to the text, name, icon and state fields in treeToJSON().
• Numbers in node data are sent at full precision; they were rounded to four digits, so a numeric id such as 123456.789 came back changed.
• sttype, sticon and sttooltip given as a number, a factor (from a data frame column, say), NA or more than one value no longer stop the tree from rendering.
• sticon values in current Font Awesome form – "fa-solid fa-folder", "fas fa-file", "fa-file fa-lg" – other icon sets, and image file names without a path are now used as given. Only a bare name such as "file" is expanded, to "fa fa-file".
• R’s own attributes on a node – a data frame’s class and row.names, say – are no longer sent as node data.
• A node without a name is now an error naming the problem, rather than a node labelled {}.
• JSON from jsonlite::toJSON() can be passed to renderTree() and updateTree(); it reached the browser already parsed and failed there.
• renderTreeAsync() is now built on renderTree(): it repopulates a tree in a modal that is reopened (#79), works with bindCache(), and an error in a rejected promise is shown in place of the tree.
• options(shinyTree.defaultParser = "tree") set before the package is loaded, e.g. in .Rprofile, is no longer overwritten, and the package can be loaded again in the same session, as devtools::load_all() does. The "tree" parser warns about siblings with the same name, which a data.tree cannot keep apart.
• treeToJSON() keeps the checked state, and gives an empty tree rather than one blank node for a root without children.
• dfToTree() builds a tree whose branches end at different depths, as a folder listing gives, where it stopped with an error; it gives an empty list for a data frame with no rows, and names any hierarchy column the data frame lacks. treeToDf() handles an empty tree and empty branches.
• shinyTree() accepts search = NULL and animation = NULL, writes ids into its scripts safely, and puts the search script after the search box rather than inside it. Search also reacts to pasting, not only typing.
• Examples 16-async, 20-api and 23-file-icons work again, 21-options has been rewritten to show what the options do since #36 and can be started with runApp(), and 14-files lists every file.
• Documentation: every exported function documents what it returns; new sections and examples for renderTree(), updateTree(), renderTreeAsync(), get_checked(), dfToTree() and treeToDf(); format = "classid", the shinyTree.defaultParser option, the contextmenu turning checkboxes off and the sticon forms are documented; and the ontologyToTree() example runs on the bundled ontology.
• Requires R >= 3.3.0 and shiny >= 1.6.0, which the package already needed. stringr, htmlwidgets and methods are no longer used, and the unused copies of jQuery and require.js are no longer bundled.
• Tests: an example app that fails to start, or a browser test app that errors, now fails the tests instead of skipping them; every example is started and checked; and fixed waits are replaced by waiting for what the test expects.

New features and other fixes

• New ontologyToTree() reads an ontology from an OBO (.obo) or OWL (.owl, RDF/XML) file and returns what renderTree(lazy = TRUE) needs, so even a large ontology such as the Gene Ontology’s 48,000 terms is shown a level at a time. The hierarchy is is_a; a term with several parents appears under each; obsolete terms are greyed out under an obsolete node; each node shows its id and definition on hover; search finds terms by name or id anywhere in the ontology. New example 28-ontology browses the ontology from the issue, or an uploaded file (#33).
• A lazy tree that is rebuilt now reports the new tree to the server, so input$tree no longer keeps the old tree’s selection.
• New renderTree(lazy = TRUE) loads a tree’s nodes from the server as they are opened, for trees too large to send whole. The render expression returns function(path) (or function(path, node), to tell same-named siblings apart by their st* attributes) giving a node’s children; children with stchildren = TRUE are fetched when opened. With shinyTree(search = TRUE) the whole tree is searched on the server, by walking it or with a search function, and the branches leading to the matches are fetched and opened. state = TRUE reopens remembered branches after a reload. Branches needed at once are fetched in a single request. Requests are checked: only branches the server has sent can be asked for, with the server’s own node data, and errors follow shiny.sanitize.errors. updateTree() and expandTree() stop with an error on a lazy tree. See the new 27-lazy example (#11).
• Unticking a child of a ticked parent with tie_selection = FALSE and state = TRUE is now remembered: the state was saved before the parent was unticked, so a reload ticked everything again.
• renderTree() given a function now stops with a pointer to lazy = TRUE, rather than leaving the tree empty with a JSON error in the browser.
• A tree now repopulates when its element is removed from the page and added back – inside a modalDialog that is dismissed and reopened, for example. Contents are pushed to the client rather than returned as an output value, so Shiny had nothing to replay and would not re-run the render function. The client now reports binding an empty tree and the contents are resent (#79).
• Fixed types being ignored for trees nested more than one Shiny module deep. Only the first hyphen of the output id was replaced when looking up the types spec on the client, so the lookup missed and the tree failed to render (#129).
• Search now hides non-matching nodes with the proton theme. jsTree marked them correctly but proton’s stylesheet omits the rule that hides them, which the other themes ship, so every node stayed on screen. shinyTree now declares it in its own stylesheet (#113).
• New stcssclass node attribute gives a node one or more CSS classes, so an app’s own CSS can style it. stclass, despite its name, has only ever been data and is unchanged (#58).
• Node data that R refuses as an attribute – a class field holding a list, as trees built from XML often have – no longer stops input$tree from being read. It comes back under an st prefix, e.g. stclass (#58).
• New shinyTree(state = TRUE) remembers which nodes were open, selected and checked in the browser, and restores them when the page is reloaded. Each tree is stored under its own key ("shinyTree-<id>" by default, or pass a string). Nodes are matched by their path from the root, and only the first tree after a reload is restored. When the remembered state disagrees with the stopened/stselected/stchecked attributes the tree was rendered with, the remembered state wins; state_priority = "code" keeps the app’s instead (#19).
• A tree is no longer sent to the browser twice on page load when its render takes more than about 100ms. The check that repopulates a tree whose element is added back (#79) also fired for brand-new trees, so the server resent the tree and jsTree rebuilt it. It now only fires for a tree that was already sent.
• New sttooltip node attribute: text shown when the pointer rests on a node’s label. ?renderTree now documents all of the st* node attributes (#66).
• Fixed two broken stylesheet links to the Font Awesome CDN on every page with a tree, which each failed with a 403 in the browser console. Font Awesome itself was unaffected and still loads with its integrity hash.
• A tree now works in R Markdown documents with runtime: shiny. Its scripts and stylesheets were written into the page’s <head>, which R Markdown drops, so jsTree never loaded and the tree stayed empty. They are now declared as HTML dependencies, which also means each is included once per page however many trees it has (#18).
• New expandTree() and collapseTree() open or close every node of a tree from the server, for “Expand all” / “Collapse all” buttons. The tree is not resent, so selection, checkboxes and search are kept, and input$tree reports the new open state (#42).
• get_selected() and get_checked() gain leaves_only. With leaves_only = TRUE they return only nodes that have no children, however deep each branch goes, so selected parents – and, with three_state, the parents jsTree checks automatically – are left out (#98).
• Large trees are much faster to render and to read back. Building a tree’s JSON grew the node list one element at a time and serialised it node by node, and reading input$tree with get_selected()/get_checked() did the same with the matches, so both were quadratic in the size of the tree. For a 30,000-node tree, rendering went from 14s to 2s and get_checked() from 15s to under 1s. The JSON sent to the browser and the values returned are unchanged (#83).
• renderTree() can now be wrapped in shiny::bindCache(). It was refused outright, and since a tree is sent as a message rather than returned as the output value, caching that value would have left the tree empty on every cache hit. The tree’s JSON is cached instead and sent again on a hit. A tree that is removed from the page and added back (#79) now gets the tree last sent rather than a fresh evaluation of the render expression, matching how Shiny replays other outputs (#109).
• Shiny no longer warns about a shared input/output id for every tree. A shinyTree is an input and an output under the same id by design, so the browser console warning – and, with shiny::devmode(), the on-page error panel – was a false alarm in every app. A shared id that involves anything other than a tree is still reported (#110).
• The 04-selected example now shows get_selected() and get_checked() side by side, with toggles for checkbox, tie_selection, whole_node and three_state and an explanation of how each combination behaves. 24-checkbox-options shows both too, and ?shinyTree documents the difference between checked and selected nodes (#136).
• With tie_selection = FALSE, input$tree now reflects three_state cascades. It was read before jsTree applied the cascade, so checking a parent reported only the parent rather than its children too, and unchecking a child left its parent reported as checked.
• Fixed get_checked() returning nothing when checkbox = TRUE and tie_selection = FALSE (#121).
• Fixed get_checked() returning nothing for ticked boxes under the default tie_selection = TRUE. jsTree treats checked and selected as the same thing in that mode, so a selected node is now reported as checked, matching jsTree’s own is_checked()/get_checked() (#121).
• updateTree() no longer reports its own refresh back to the server. Calling it from an observer watching input$tree could otherwise loop indefinitely when options(shinyTree.refresh) or options(shinyTree.setState) was enabled. Refreshes from other sources are still reported, and the two options are now documented (#36).
• Node attribute values are no longer trimmed on their way back to R, so padded keys and fixed-width codes stored in st* attributes survive a round trip. Node names are still trimmed, which is deliberate (#3).
• Fixed get_selected()/get_checked() returning nothing for trees rendered from treeToJSON(). A node state without a checked key made the input handler fail with “argument is of length zero”, so input$tree was never populated (#103).
• Fixed duplicate-named sibling nodes collapsing into a single entry, so get_selected()/get_checked() now report the correct node and its own attributes rather than only the last sibling (#104).

shinyTree 0.3.1

• Checkbox functionality.
• Bug fixes.

shinyTree 0.3.0

• New function set_node_attrs() to set distinct node or leaf attributes (useful for creating file tree icons).
• New functions dfToTree() and treeToDf() convert between a data frame and a tree.

shinyTree 0.2.7

• Added context menu support.
• Added support for promises/future to render a tree asynchronously.
• Added additional examples for file browser and metadata retrieval.
• Added support to display various icons using types and specifying tree items
• Added support for conversion from/to data.tree

shinyTree 0.2.6

• Many improvements to expose more jsTree options.
• Tree can now be correctly updated (see new examples).

shinyTree 0.2.0

• BREAKING CHANGE: Model the input$list-style lists after the list structure provided to renderTree, meaning that they are now a simple list with the same st- prefixed attributes. See inst/examples/05-structure.
• BREAKING CHANGE: removed selected attribute from shinyTree() in favor of get_selected(). See inst/examples/04-selected/.

shinyTree 0.1.0

• Initial Release