Changelog

A plain-language summary of what changed in each release, for anyone deciding whether to upgrade. Anything under "Unreleased" has been made but not yet published as a version.

6.30.1

Opening a JSON file from disk, by drag and drop, or from a URL no longer rounds very large whole numbers (past about 9 quadrillion). Before, an ID like 12345678901234567890 changed as soon as the file opened, and saving wrote the changed number back.
Python (.py) files keep special characters intact. A string Python writes with an escape, such as '\xa0' for a non-breaking space, used to open as the literal letters and save that way.
Very large decimal numbers such as 1e16 now stay decimal numbers when you save, instead of turning into slightly different whole numbers.
Comparing two files is more reliable with tables. A list that contains a table and was reordered no longer breaks the comparison view. Swapping the two sides no longer changes the number of differences. Tables that differ only by a missing value, a short row, or a repeated column name are no longer reported as identical.
Editing a table as text (the ✎ button) no longer changes things you didn't touch. Pasting back an unchanged copy now really changes nothing, adding or moving rows no longer turns values like "001" into numbers, and a column of names copied from Excel that contain commas pastes correctly.
CSV files keep rows whose only value is empty, and values containing a tab, when you save and reopen them.
Opening a Python dict whose keys would merge into one in the viewer (such as 1 and "1") now shows a warning, instead of one entry silently disappearing.
Opening a Python data file (.py) from Python or the command line now shows its contents as data, instead of the whole file as one long text value.
When a file format can't store a value exactly as you typed it (for example the text "001" in a CSV file, which reads back as the number 1), the edit now shows the value the saved file will really contain, with a short note. Numbers too large to store at all are refused.
While editing or comparing, double-clicking a column heading of a table made of plain lists no longer opens a rename box that silently does nothing.
The bundled YAML interpreter now opens values containing an apostrophe or quote (such as note: it's fine), and files it saves always open again — including lists of lists, text with line breaks, and an emptied document.
The bundled INI interpreter no longer changes values you didn't edit when saving, such as a zip code 02134 becoming 2134. Values INI can't store (like text with a line break) now give an error instead of a broken file.
Exporting to another format through an interpreter now warns when that format can't hold your data exactly, and names what changes.
Editing a value whose key is empty ("") at the top of a document no longer replaces the whole document with that value when you save.
Settings loaded from a file or passed from Python now get the same limits as the Settings panel, so a value like 0 rows can no longer leave boxes empty or overlapping.
A key literally named __proto__ is no longer lost when you edit, compare, or open Python or CSV data that contains one.

6.30.0

"Save as HTML…" while editing now saves a copy that opens normally, with your edits in it. Before, the copy opened stuck in the Edit bar with buttons that did nothing; a copy saved from the Jupyter widget, or with Settings open, also came back with parts of the toolbar hidden or the panel stuck open. It now also keeps which boxes were collapsed or opened on the same items even when your edit renamed keys or removed rows.
The export options (JSON, Python, CSV, Excel, HTML) now always save what's in the edit pane, even if you export a moment after typing. Before, they could still save the version from before your last keystrokes. "Save as HTML…" is now greyed out while your edit has an error, since the copy it made would open empty.
Clicking Paste… or From URL… while editing no longer throws your edit away. Paste… now pastes into the edit you're reviewing, and From URL… only replaces the document once something has actually loaded. Paste… also works again from a comparison of two documents. Leaving an edit while a search was running no longer brings the old edit-review values back when you clear the search.
Large documents redraw faster: parts of the view that are off screen are no longer drawn until you scroll to them, which roughly halved the time a load, a collapse or a link jump took on a document with 1,600 dicts.
When a view gets big enough that every click redraws tens of thousands of page elements, a warning in the top-right corner now says so and offers a fix: limiting each dict to its first 20 nested dicts/tables, or opening the setting that controls the warning. Settings → Nested containers → warnElements sets the threshold (20,000 by default, 0 turns it off).
A dict with many nested dicts or tables now shows only the first 20 of them, behind a clearly marked "k of n nested dicts/tables hidden" banner right under its header; the banner's buttons show 100 more at a time or all of them. Following a link, a search match or a click in the source pane to something hidden reveals just that one. Both numbers are settings (Settings → Nested containers), and a checkbox there turns the whole thing off if you'd rather always see everything. Saved views remember what you revealed.
The widget's "Set as default" button works again — it had silently stopped applying anything once the "clickable badges" checkbox setting existed.
Hovering a link (a value that names another object in the document) now shows a preview of that object right there, with two buttons: pin it in place beside the link, on top of everything else, or detach it into a side panel connected to the link by dashed lines. Both windows have a close button, and clicking the link itself still jumps to the object as before. A link that points at one row of a table previews just that row. Open link windows are kept when you save and reload a view.
A link can now show the object it points to right where the link is, as if the object were written there: click the chain button in the link's preview, and click it again on the box to get the link back. When a document has links, an "Expand" button at the top does this for every link at once, and for the links inside those, 10 levels deep (the number next to it changes that, or removes the limit); loops of links stop just before repeating. Editing a value inside such a box edits the original object, and every other place that shows it updates too. Your file still contains the link, not a copy.
Dates, times, decimals, UUIDs, enum values and your own classes in the data you open are now shown as Python writes them, for example datetime.date(2020, 1, 1), instead of stopping open() with an error. You can edit them in place, and after saving, .value holds real Python objects again. If an edit can't be turned back into an object, the viewer tells you and .value keeps its previous value.
A table's Copy menu (⎘) can now save the table as a JSON, Markdown or native-format file (for example .xml when your viewer has an XML interpreter built in). Before, only CSV and Excel could be downloaded, and the other formats could only be copied to the clipboard.
In a diff or while editing, the i guide to the six diff views now lets you pick one. Hover over a view to preview it behind the guide, and click to switch to it. Moving away without clicking brings back the view you had, including any collapsed or minimized tables, scroll position and search.
The [?] popover and structile --version now say when a build carries changes newer than its version number, as "6.29.3 + unreleased changes" (a build taken straight from the repository between releases). The popover's date row is now labelled "Released": it is the date of the version shown.
After choosing "Show as hex" for a binary file, you can still drop its interpreter onto the page and it reads that file. If the interpreter can't read it, the viewer says so and keeps showing the hex view.

6.29.3

Fixed: clicking a link in an expanded table used to scroll that table back to its first row. It now still jumps to the linked item, and the table stays where you had scrolled it. The same fix applies when you click inside a hover preview to pin it open.

6.29.2

Fixed: a file opened from Python with structile.open("data.xml") (or any other file path) was labelled as having no file on disk, just like pasted data. Its name now shows the folded corner, and hovering it gives the file's real name. Data passed in directly, not from a file, is still labelled as having no file.

6.29.1

Fixed: after you edited a value in the graph of an XML file (or any other interpreter-read format), clicking a value no longer highlighted it in the source, and moving the cursor in the source no longer selected it in the graph. This worked for JSON but broke for XML after the first edit. It now keeps working after every edit, in the viewer, in Edit review, in Split-view diff and in the VS Code editor.

6.29.0

A custom build can now bundle an interpreter for a binary file format: mark the manifest entry "binary": true (or give it "format": "bin"), and dropping a file of that extension opens it straight away — no companion .js, no hex-view prompt. The interpreter always receives the file's exact bytes, even when they happen to look like text. This also covers the file picker, a data file dropped together with its interpreter, and comparing two such files in diff view.
"Save viewer with interpreter…" from a binary file now produces a viewer that recognises that file's own extension (for example .abf) instead of a generic .bin that never matched anything.
"Save as HTML…" now works for binary data, both when read through an interpreter and in the hex view: the saved page carries the bytes and reopens showing the same data (still read-only, as in the live session). When a saved copy can't be made at all, the menu row is greyed out with the reason instead of failing after the click.
The [?] popover's list of bundled interpreters marks the ones that read binary files.
A binary file no longer has to carry the bundled interpreter's extension to be recognised: whatever the file is called, every bundled binary interpreter is tried on its bytes before the viewer asks you for one — including files whose bytes happen to look like text. The same goes for an interpreter you drop yourself: if it rejects the file read as text, it gets the raw bytes as a second try.
VS Code extension: an interpreter can now declare "format": "bin" to read a binary file format, and structile.extensionFormats accepts "bin" for a sibling or configured interpreter. Such files open read-only; "Structile: Show Interpreter Resolution" reports which candidate read them, as for any other format.

6.28.1

Filenames in the new name tag are no longer clipped along the bottom: letters like g, j, p, q and y, and the underscore, showed up with their tails cut off. The tag itself is unchanged in size and position.

6.28.0

The filename in the top-left corner is now a small tag shaped like a document, with the file's type shown beside the name instead of tacked onto the end of it. Diff mode's two filenames look the same way, so a file is labelled identically wherever you see it — a tab-separated file used to be called "csv" in one place and "tsv" in the other.
The viewer no longer makes up filenames. Opening config.yml used to show it as config.yaml — a file that doesn't exist. Hovering the name now shows what the file is really called, how the data was read, and what extension saving would use.
Data that came from a paste, a URL or a notebook shows a square tag rather than a folded one, so it's clear at a glance when there's no file on disk behind what you're looking at.

6.27.0

CSV files now open on their own. Drop a .csv, .tsv or .tab file (or open one from Python or VS Code) and it renders as a table straight away — no interpreter script to download, pair or write. Comma- and tab-separated files both work, and which one you have is detected from the file's own content, so a spreadsheet exported "as text" opens correctly too. Saving writes back the same dialect it read.
Every table has a new ✎ button next to its ⎘ Copy button: paste a CSV or Excel table into the box, press Process, and the changes arrive as edits you can review row by row, undo, or cancel before saving. The box opens pre-filled with the current table, so you can also just edit it in place. Copying a table out and pasting it straight back changes nothing.
"Export CSV…" and "Export Excel (tab-separated)…" are now always available in the Save menu, for data loaded from any format — they used to appear only if you happened to have opened a CSV file earlier in the same session. When the document isn't a table, the row explains why and points at the table's own ⎘ menu, which can now also download one table as a CSV or Excel file.
The bundled CSV interpreter (generic_csv.js) is gone, because nothing needs it any more. If you are on an older version and click "Use Structile's generic CSV interpreter" for a dropped .csv, that download no longer exists — upgrading is the fix, and afterwards the file just opens.
The VS Code extension now opens .csv/.tsv/.tab files with the viewer too.
The downloadable starter templates for shipping your own format — the Python plugin, the VS Code companion extension, and the custom/branded distribution — have been refreshed. The plugin and companion-extension downloads had fallen behind and contained an older example interpreter missing selection sync, plus a link that only worked inside a private repository. Re-download them if you started from a copy earlier.
The guide to building your own branded viewer now shows how to bake several interpreters into one build, which was always supported but only ever demonstrated with a single one.

6.26.3

Clicking a file name in the diff header no longer pops open a file dialog. It now simply opens the "Assembling a diff" panel for that side, where the new file can be dropped in or chosen with a button — so an accidental click is easy to ignore, and dragging a file in is possible again.

6.26.2

Diffs now weave right-only rows, columns, and list items into their logical positions while keeping the left order stable, preserving both source orders whenever possible. Tables with repeated key values also match exact and uniquely closest rows instead of discarding the chosen key; genuinely tied cases remain visibly added/removed rather than being guessed.

6.26.1

Table Copy is now the leftmost table control. JSON tables no longer repeat JSON as a separate Native option, and copied Markdown tables have neatly padded columns and matching separator widths for immediate readability.

6.26.0

Every table now has a Copy menu in its top-right corner. Copy the complete table—not only the rows currently visible—as CSV, Excel-friendly tab-separated text, JSON, a formatted Markdown table, or the current file type's native representation.

6.25.0

Assembling a comparison is no longer picky about order or aim. While Structile is waiting for the rest of a diff, anything you drop anywhere on the page joins it — the interpreter no longer has to land exactly on the panel's drop zone. Dropping the interpreter first and the two data files afterwards works too, and an interpreter you already used this session (or that came with your build) is tried automatically, so two files of a familiar format often compare with nothing else to drop.
While a comparison waits for its interpreter, the panel now shows the raw text of each file you dropped, so you can see what you handed over instead of staring at a request for a file.

6.24.0

You can now drop any file on Structile, not just the handful of data extensions it used to accept. Text files open as before — a file holding JSON or Python data just loads, and anything else shows its content and asks for an interpreter, whatever the extension is.
Files that aren't text (images, archives, documents, anything binary) are no longer garbled into the page. Structile says what the file appears to be and offers a built-in hex view of its bytes, or you can hand it an interpreter that reads them.
When you drop a file in a format Structile publishes an interpreter for (CSV, INI, TOML, YAML, XML, HTML), it now offers that interpreter directly: use it straight from the web with one click, download it to keep next to your data, or paste the address of your own.
TOML and YAML config files can be read out of the box now — both come with a new interpreter covering the shapes a normal configuration file uses. Each says so plainly when a file uses something it doesn't cover (multi-line strings, YAML anchors and the like) instead of quietly misreading it.

6.23.1

In a diff, the file names now sit on the same line as "Left:"/"Right:" and the file type, instead of slightly below them, when the new reload buttons are showing.

6.23.0

If you edit a file after opening it, Structile now notices and offers a ⟳ button next to the filename to reload it in place — no need to find the file in the picker again. In a diff there is one button per side, each tracking its own file. The button is always there once a document can be reloaded, and lights up only when its file has actually changed.
This needs a Chromium-based browser (Chrome, Edge) and works with files opened straight from disk. In Firefox and Safari everything works exactly as before, just without the reload button.

6.22.1

The dark theme is lighter and easier on the eyes now, using the same colors as VS Code's own default dark theme instead of the previous near-black background.

6.22.0

The CSV interpreter now keeps graph selections and source fields in sync in both directions, including after an edit. CSV cells can also be set to null, saved, and reopened as real null values.
The CSV sample, interpreter, and ready-to-open HTML demo are now included in the downloadable examples and study material.

6.21.0

The Save menu can now convert your data into other formats you have interpreters for, not just the one it was loaded from. Alongside the existing "Export JSON…" there's now "Export Python…", plus one "Export \…" entry for every other data format you've loaded an interpreter for this session (or that's built into your copy of the viewer) — for example, once you've loaded both a CSV and an XML interpreter, you can convert CSV data to XML and back again.

6.20.0

Removed the separate "Save view"/"Load view" buttons. Instead, "Save as HTML…" now remembers which panels you had open or closed and restores them when you reopen the saved file.

6.19.0

Interpreter scripts (the small .js files that teach the viewer a new data format) can now advertise their own documentation links, which show up in the "?" info popup right next to that interpreter's name — useful for an in-house format whose real docs live on an internal wiki.
Added a bundled CSV interpreter and a small sample CSV file with a ready-to-open demo, showing that new feature in action.

6.18.3

Fixed a crash when editing a value inside certain Python-style set data (rare, but reachable by opening a .json file that isn't strictly valid JSON). Clicking Edit used to freeze up and every later double-click said "Nothing to edit here" — editing now works normally, and the changed entry is now correctly colored red (old) and green (new) instead of showing two identical-looking, uncolored copies.

6.18.2

The FpML links have moved from the viewer's "?" popup to the repository page, which now links the FpML project and latest JavaScript interpreter. A new FpML sample and ready-to-open HTML demo show them working together.

6.18.1

Added: the "?" info popup now has an Interpreters line linking the interpreter guide and a site hosting an FpML interpreter you can load straight into the viewer by URL.

6.18.0

Added: an interpreter can now flag problems with a file without refusing to open it. If a file is mostly valid but has something wrong — a missing recommended field, a value out of range — the interpreter can point it out, still show you the data, and drop a note (with the rule and line) in the message log added last release. A genuinely wrong file type is still rejected as before.

6.17.0

Added: a message log. The little status messages and warnings that pop up and fade away are now kept in a list behind a new button next to the "?" in the top-right corner (it shows a count of ones you haven't looked at yet). Open it to re-read anything you missed, with timestamps; there's a Clear button, and the list resets when you reload the page.

6.16.0

Fixed: cross-reference links now work when the thing being pointed at sits in a list that's shown as a table (which is how most real FpML documents lay out their parties, streams, and so on). Previously those references showed as plain text; now they're clickable and jump straight to the matching row. Interpreters that declare an id field get this for free.

6.15.0

Added: a "Save viewer with interpreter…" option in the Save menu. After you open a file through an interpreter, this saves a fresh copy of the viewer with that interpreter built in and no data. Send that single HTML file to someone and they can just drag matching files straight onto it — no separate interpreter .js, no internet needed. Dropping a file the built-in interpreter doesn't recognise still works the same as before (it asks you to pick one).

6.14.0

Added: you can point the viewer at a companion interpreter hosted on the web with a ?interpreter=<url> in the page address, instead of having a copy of the .js file to drop in. The viewer fetches it on load and pairs it with whatever XML (or other structured file) you open next. The hosting site has to allow cross-origin requests; if the fetch fails you get a clear warning rather than nothing happening.

6.13.0

Added: an interpreter can now set up cross-reference links for you. If it declares its reference delimiters and id field, then dropping an id/href-style document with that interpreter makes the references clickable straight away — each jumps to the item with the matching id — without opening Settings and typing a link prefix/suffix yourself. A new bundled ref_xml.js interpreter and sample_ref_xml.xml show it working. Your own saved link prefix/suffix are not overwritten on disk.

6.12.0

Added: a new bundled interpreter for the ISDA CDS Standard Model / Markit "Interest Rate Curve XML" format. Pair it with an <interestRateCurve> document (drop or paste both together) to view and edit the deposit and swap curve points as an ordinary structured tree, and save changes back to the same XML. A sample curve file is included.

6.11.0

Changed: you can now drop a file anywhere on the standalone viewer to load it, not just onto the top bar. Dropping onto a blank area of the page used to do nothing (or, worse, make the browser navigate away and lose unsaved edits).
Added: while you drag a file over the window, a full-screen hint appears saying what dropping will do — one file loads; dropping two or more data files at once compares them (the same as the "Diff…" button, whether or not something is already open). When a document is loaded, the "Diff…" button lights up during the drag, and dropping the file onto it compares it against whatever is already open. Before anything is loaded the hint doesn't mention "Diff…", since the button isn't available yet.

6.10.0

Fixed: in a Jupyter widget showing data that wasn't loaded from a file (e.g. a plain Python object passed straight in), "Save As…" used to silently do nothing when clicked — there was no file location to save a copy to. It's now shown grayed out in that case, so it's clear upfront that it isn't available, instead of appearing to work.

6.9.0

Added: while reviewing an edit (the Edit button's before/after view), Save As…, Export JSON…, and Save as HTML… now work right away instead of being grayed out until you'd made a change and waited a moment — and once you have changed something, they save your in-progress edit, not the original. Export JSON still needs your edit to be valid first, since there's nothing to export as JSON otherwise.
Changed: saving mid-review used to refuse outright if the text you'd typed didn't parse yet ("Fix the source error before saving"). It now saves your in-progress text anyway as a safety checkpoint, with a warning that it wasn't applied to the graph yet — so a typo or a mid-edit pause can no longer block you from saving your work.

6.8.1

Added: the read-only-editing warnings added in 6.8.0 now always appear in the same place — the top-right corner of the graph area, with a small countdown ring showing how long it'll stay up — instead of the read-only-source-pane hint showing at the bottom of the page while the "can't edit here, switch to Split view" warning showed in the corner.
Added: the read-only-source-pane hint now also covers pasting, cutting, and deleting a word, not just typing — and works whether paste/cut was triggered from the keyboard or a right-click menu.

6.8.0

Added: trying to type directly into a read-only source pane (the single-value source panel, or the original/left side while reviewing a self-edit) now shows a note explaining why nothing happened and how to actually make the change — click Edit, or type in the pane on the right. Pasting is left alone, since it wasn't the gesture this was trying to explain.
Added: Save now works even if you haven't changed anything yet — useful for saving a copy of something you just pasted, fetched from a URL, or are viewing as sample data, without needing to make an edit first. This only applies to the standalone page; saving inside an embedded host (like the VS Code extension) still requires a real change.
Added: the "can't edit here, switch to Split view" warning (shown when double-clicking a value in Unified diff view) now appears in the top-right corner of the graph itself, with a small countdown ring showing how long it'll stay up, instead of the general notification banner at the bottom of the page.
Changed: the About popover's description no longer says it packs "nested JSON" — it reads/writes several formats, so this was undersold; now says "nested data".

6.7.6

Fixed: collapsing/expanding a section in the graph for an XML/HTML file didn't fold the matching part of the source panel, and double-clicking a collapsible section's opening tag in the source didn't collapse it in the graph either. Both directions already worked for JSON/Python files; now XML/HTML matches.

6.7.5

Fixed: for an XML table with named columns (generic_xml.js/ tagged_xml.js), clicking a column header in the graph — or the matching bit of source text — didn't sync at all in either direction. Individual cell values already worked; only the column headers themselves were missed. Fixed.

6.7.4

Fixed: XML/HTML selection sync (added in 6.6.0, with a type="null" fix in 6.7.2) was still wrong for any file saved on Windows — clicking a value in the graph would highlight the wrong text in the source panel, worse the further down the file it was. Windows text files use a different line-ending convention than the sync code accounted for; now fixed for real, on either platform.

6.7.3

Changed: the raw text of a dropped XML/HTML/etc. file waiting on its interpreter now shows up in the actual source panel (if it's open), instead of a separate preview box — the same place it'll show once the interpreter is found, with nothing else changing when that happens.

6.7.2

Fixed: dropping an XML (or HTML/ini/toml/yaml/csv) file before its matching interpreter was available used to show nothing but a message asking you to pick one. Now the raw file contents show up right there too, so you can see what you're working with while you find or write the interpreter for it.
Fixed: with the two example interpreters that mark a null value as type="null"/kind="null" (generic_xml.js, keyed_xml.js), clicking that value in the graph didn't sync to the source panel at all — worse, it silently left whatever you'd clicked previously still highlighted, which looked like sync had jumped to the wrong spot. Fixed; a null value now highlights its own element.

6.7.1

Fixed: editing a value that belongs to a set (rather than a list/dict) through the graph — e.g. one of the tags in generic_xml.js's type="set" fields — silently did nothing: no error, but the value stayed unchanged everywhere (graph, source pane, saved file). Editing a set member now actually commits.

6.7.0

Fixed: the XML/HTML selection sync added in 6.6.0 didn't actually work for a normal XML file — almost every real .xml file starts with an <?xml version="1.0"?> line, and that alone was enough to silently turn sync off for the whole file. Now fixed; sync works the way 6.6.0 intended it to.

6.6.0

Clicking a value in the graph now highlights the matching text in the source pane for XML/HTML files too, not just JSON and Python — and vice versa: clicking in the source text highlights the matching graph value. This works whenever the interpreter you're using supports it (all of the bundled example interpreters do); if you're using a custom one that doesn't, sync just quietly doesn't happen for that file, the same as before. Collapsing/expanding a section in the source pane also stays in sync with the graph for these files now, including through the VS Code extension's real-editor view.
Two new example XML interpreters are bundled alongside the existing three: one where a whole record lives in one element's attributes (the common spreadsheet-export style), and one that needs no special attributes at all — it just reads a tag's own name as the key and its text as the value, the way most hand-written XML already looks.

6.5.1

Fixed: dropping an XML (or HTML/ini/toml/yaml/csv) file before its matching interpreter was available used to show nothing — just a message asking you to pick one. Now the raw file contents show up right away in a preview box, so you can see what you're working with while you find or write the interpreter for it.

6.5.0

The [?] About popover now shows a "Baked-in data" line whenever the HTML file you're looking at has data built into it — from "Save as HTML…" or from opening it via structile in Python — telling you its size in MB and how many individual values it contains.

6.4.1

Fixed: dropping a saved .view.json file before opening any data used to do nothing visible, with no explanation. Now, if there's no data loaded yet, it just opens the .view.json file itself so you can see what's in it; drop it again once your data is loaded and it applies as a view like normal.

6.4.0

The Save button now has a small ▾ menu next to it with two new options: "Export JSON…", which saves your data as plain JSON even if you opened it from XML, Python, or another format; and "Save as HTML…", which bundles your data (and its interpreter, if any) into a single HTML file you can send to someone else — they can open it directly, with nothing else to install.
Save now stays put in the same spot while you're reviewing an edit, instead of disappearing and being replaced by a separate Save button elsewhere. On the standalone page, it's now labeled "Save As…" to be clear that it always asks where to save, since there's no original file it can silently overwrite.

6.3.0

Interpreters can now identify themselves: the [?] About popover shows the name and version of whichever interpreter loaded your current data, and, in a custom/branded build, the name and version of every bundled interpreter too.