chromosomix 0.20.0

What changed in chromosomix 0.20.0, released on 4 November 2026.

Added

• The viewers have a page of their own (§18.10). The time to a picture — one map and one track in chromosomix, juicebox.js and HiGlass, from disk and from the bucket — moves out of the timings into notebooks/viewers.qmd, published at https://chromosomix.app/viewers/ with viewers.tsv beside it, and a release keeps viewers-<tag>.html beside its tables. The timings are the validation’s commands alone. About → Notebooks lists five notebooks, the viewers among them, and the pages link to each other as before.

• An environment for the laptop (§18.10). notebooks/environment-laptop.yml makes one, named open2c, for micromamba, mamba or conda — Quarto, Jupyter, the packages the page imports and the four its deep human map runs, at the reference environment’s versions — and micromamba run -n open2c quarto render notebooks/laptop.qmd renders the page in it. The page now looks for every package it needs, and for Node.js, before it downloads or runs anything, rather than stopping hours in at the deep map, and names under The machine the Python it ran in and the version of each reference package, with the pin where it differs.

• An embed opens its side panels when its page asks (§21.2). ?embed=1&panels=open opens the Annotations and Aesthetics panels, panels=annotations or panels=aesthetics one of them; without it they arrive folded, as before. Written nowhere, like any fold in a frame.

• The notebook widgets can be resized by the reader (§21.5). A grip under the Jupyter widget’s and the R widget’s frame changes its height, dragged or with the arrow keys, and a double click puts back the height the notebook gave. In Python the reader’s height is widget.height; in Shiny it comes back as input$<id>_height.

• The documentation turns the page with the arrow keys, and a live view can be resized (§32.4). Left and right go to the previous and the next page, except in a field, the search or a block that scrolls. Pressed, a live view opens with the app’s two panels unfolded in a frame wide enough for them, and the grip under its corner resizes it, by dragging or with the arrow keys. The page on embedding shows a frame any page can make resizable, with a few lines of script.

• A pull request’s preview carries its documentation (§32.5). Beside the preview’s app at chromosomix.app/pr/<number>/, the documentation of the same head is published at chromosomix.app/pr/<number>/docs/, replaced on every push and removed when the pull request closes. Its live views open the preview’s own app, its banner names the pull request, and the pull request’s comment gives both addresses. Documentation that does not build leaves the app’s preview published, and the comment says so.

• The Contact maps card opens a map in a cell of its own (§16.9). Under the lower triangle’s rectangle a second one, Load a map into a cell of its own, takes a cooler — or several, or a row of the Files column — and opens each in the deck beside the map, the first empty cell or a new one at the end. It is there whenever a map is open, and stays once a second file fills the lower triangle and that rectangle goes; its hint says what the triangle’s cannot, a cooler at any bin size. The card’s only drop with a map open used to be the triangle’s, so a new file reached the deck from the card by no road at all.

• Live examples in the documentation (§32.4). Ten pictures — the introduction’s, moving around, a contact map and its O/E, two maps compared, a ratio, one map read two ways, the 45° view, the genome browser and an embed in a page — carry Try it live, which swaps the picture for the app itself, embedded, on the same view of the demo data, across the page’s width from the sidebar to the window’s edge, with a line to open it in a tab of its own or go back to the picture. The contents list on the right steps aside while the frame is under it. Nothing of the app is loaded until it is pressed, and each copy of the site embeds its own release’s build.

• A documentation site, one copy per release (§32). https://doc.chromosomix.app is written for the people who use chromosomix: getting started, the files it opens and from where, using the browser, every analysis, the command line, the notebook widgets, hosting data and publishing a store of your own, privacy, and a glossary. Ways to use chromosomix draws the nine places it runs — the web app, installed or not, the desktop app, one HTML file, a fixed release, your own server, a web page, a notebook and a shell — and says how to start each. Its 75 pictures are the real app, shot by a script per page on the demo data and shown at the size the app drew them. Each task has a page whose method, validation result, command and every parameter are read from the app’s registry as the site is built, so they are that release’s; the reference pages for links, the command line, embedding and formats are the ones the app writes for language models, and the release notes are this file, one page a release. Every release gets its own copy, built from its tag, at /v<version>/; the latest is also at the root and main at /dev/, and a menu in every copy’s header lists them all. It is Starlight with the Rapide theme in the app’s own colours, grain and type. The source is documentation/ (npm run docs:dev); .github/workflows/docs.yml publishes it to the gh-pages branch of js2264/chromosomix-docs with a deploy key, and builds without publishing until that key is set; the pull-request suite builds it and checks every internal link.

• The app opens its documentation (§5.5, §32). An open book in the header, between the shortcuts and the source, opens https://doc.chromosomix.app in a tab of its own; so do Documentation in the palette’s Help, a line in About before the source, and the last lines of chromosomix --help. Each opens the site’s root — the latest release’s copy — since the live site, built from main, cannot name a copy of its own. The README carries a badge for it.

• Test coverage, over the whole app. npm run test:coverage runs the full suite with V8 coverage over every file under src/ and electron/ — a file no test loads counts at zero, so the figure is the app’s and not the suite’s — and prints it one row per area of the source, with the files each leaves unloaded; coverage/index.html is the file-by-file view. The Tests workflow reports the fixture-free suite’s coverage on every push, and the full suite’s where metacooler’s fixtures let it run, in each job’s summary, with the report kept as an artifact for two weeks. After every green full suite on main, the report is published at https://chromosomix.app/coverage/, and the README’s Coverage badge shows its line coverage (docs/deploy.md).

• The laptop ends on a deep human map, start to finish, against the packages it reproduces (§18.10). notebooks/laptop.qmd now times six analyses one after the other over 4DN’s IMR-90 in situ Hi-C (4DNFIJTOIGOI, 1.08 billion contacts): loop calling and diamond insulation (10 bins) at 10 kb, compartments at 250 kb phased by GC content, APA over every loop called, raw, P(s) at 10 kb over every diagonal, and a virtual 4C at 10 kb from the bin of MYC’s promoter, in cis — the first five over the whole genome. chromosomix runs them as six commands; the reference is one Python script running cooltools and coolpuppy in one process (notebooks/scripts/cooltools_pipeline.py). Each side’s start-up, each step and the whole are timed — CPU, wall clock and peak memory, the memory weighed four times a second on a Mac as on Linux — and what the two wrote is compared step by step. Every step is a line of pipeline.tsv, and every weighing of each side’s memory a line of pipeline-memory.tsv, both carried in the page and unpacked beside it at laptop/; the page and notebooks/scripts/timing.plot.R’s panel c draw the memory against the wall clock, each phase shaded, its peak and the end of each side marked. The laptop needs cooler, cooltools, coolpuppy and bioframe for this part, and hg38 (3.2 GB) for its GC content.

• The parity runs in Firefox and WebKit too (§18.10). notebooks/parity.qmd drives its ten tasks in each of the three engines Playwright drives headless — Chromium, Firefox and WebKit — and runs the line each browser’s console wrote on the files that browser’s sessions read, so the pileup is over the loops that browser called. Every file is compared browser by browser: summary.tsv has a browser column, each case lists its verdict in each browser and says whether the three wrote the same line, and PARITY_BROWSERS names fewer. The Notebooks workflow installs the two for the parity’s job.

Changed

• The notebooks render all at once (§18.10). .github/workflows/notebooks.yml starts the parity, the validation with the timings after it, the viewers and the laptop together, each on a runner of its own, so a run takes as long as its longest job, where the validation and the timings, the viewers’ pictures among them, had come to four hours after the parity’s quarter of an hour, and outran the six hours a job may take once the deep human map joined them. No job waits for the parity to pass any more; the four published pages still go up together or not at all. The timings stay on the validation’s runner, since they run its commands on its files.

• A scaled metaplot’s x axis says where the features start and end (§18.8). It was labelled with column numbers — 0, 100, 200, 300 — which told a reader nothing about the plot. It is now ticked at Start and End, the edges of the rescaled body, with round steps that fit the plot’s width between and outside them: 25%, 50%, 75% (or tenths) through the body, and -5 kb … -1 kb before the start and +1 kb … +5 kb after the end along the flanks. A wider figure and a window wheeled into one part of the axis take finer steps, and a PNG or PDF carries the ticks on screen; the axis now reaches the flanks’ far ends; and the crosshair and the tooltip spell the position the same way (37.4%, -150 bp). The per-feature heatmap under it, which had five blank ticks, names -5 kb, Start, 50%, End and +5 kb where they fall. The numbers the curve’s TSV and the command line write are unchanged.

• The laptop notebook’s memory table gives a browser’s whole peak (§18.10). Its column for a browser was what a run added to what the page held when run was pressed, beside the command line’s whole peak, so the app read as lighter than the command line in 49 of its 69 runs, though its page held 160 MB to 1.4 GB before the run began. The table now gives the whole peak in every venue, the page’s in each browser, the app itself included, and beside it, for each browser, what the run added; the sentence under the tables gives the median of each.

• The timings no longer time Pairs QC (§18.10). Its seven commands — chromosomix pairsqc, and cooler cload, pairtools restrict on each strand, the counts drawn from them, the coverage against the cuts and pairtools stats as its reference — are written to commands.json and not run by notebooks/timing.qmd, as §§ 20–23’s are not: Pairs QC is still held to them in notebooks/validation.qmd § 27, and timed nowhere. notebooks/scripts/timing.plot.R draws no point for it.

Fixed

• A notebook’s download asks again after a 404 (§18.10). NCBI’s server for GEO answered 404 Not Found for the validation’s yeast map, a file it had served to the parity’s and the laptop’s runners minutes before and served to every request after, and the validation stopped on its first cell. notebooks/scripts/download.py now asks a 404 again three times, after 2, 4 and 8 seconds, as it asks a 500 or a 403 again, and still names a file that is truly not there within a quarter of a minute; any other 4xx is raised at once, as before.

• An embedded map takes the frame’s height (§21.3). Every picture sized against the window counted out a header, a tab strip, a breadcrumb row and a status bar that a frame does not have, so in an 850-pixel frame the map stopped 120 pixels short of the foot, over a band of empty page. It now grows into it.

• The documentation looks like the app (§32.4). Its header is the app’s header, measure for measure: the bar’s height and padding, the mark, the search well and the app’s buttons. The home page’s rows follow the sidebar’s order, and Open the browser sits beside What chromosomix is. The guides are a third shorter — 112,000 words down to 73,000 — and keep every control, default, limit and warning they named. And a live view’s right side, under the contents list, takes the pointer again: the list’s column, folded away or not, was still catching every click there.

• A deck cell’s heading no longer runs into the next cell (§16.9). Four maps across a column with two side panels open are cells of 154 pixels, and a heading — a grip, the name, the sequence, the star, the chain and the × — could not be less than 196: the name was cut to one letter and the controls ran on past the cell’s edge, the chain drawn over the next cell’s grip and name. Where the name cannot keep 6rem beside them, the controls now go together to a line under it, at the cell’s right, and the name takes the whole first line; in a narrower cell still they wrap again inside it. Every kind of cell does this — the first map’s, a structure’s, a tracks cell’s, an empty one. The star and the chain in a heading are square, where .btn’s padding had made them 36 pixels wide round a 13-pixel mark, and a collection’s rung picker is as wide as the rung it shows, as the sequence picker beside it already was — so a narrow cell’s controls fit under its name on one line.

• An embed’s bar fits a narrower frame (§21.3). The bar was one row that scrolled, so in a frame of 1,040 pixels the contacts box’s example was cut, at 910, the width of an R Markdown page, the row ended at con, at 720 the resolution and the box were out of sight, and at 480 the frame scrolled sideways. The two now take a line of their own under the file and the place, together and with their labels, and the file is named in full where it fits; narrower, the brand keeps its mark alone; and at a phone’s width the contacts box is put away and the resolution kept, its stops wrapping and its slider gone, so a file of five resolutions keeps all five on the frame.

• The pointer gives a derived cell’s reading and the count behind it (§7.5). On one map drawn as O/E, the line under the map read O/E 1.95× · balanced 1.95, the count being the ratio drawn again, over a cell whose balanced count was 0.000537; and the card beside the pointer gave balanced 1.95 alone. Both now give the derived value, the card on a line of its own, and under it the cell’s own count; the expected map and the O/E along the arms read the same way.

• A sequence’s name opens the sequence, and a gene typed whole is listed first (§5.4). With a gene model loaded, IV and Enter in a locus box opened the gene IVY1 rather than chromosome IV, since Enter takes the first gene listed and IVY1 begins with IV; and RAD5 was listed fifth, after RAD59, RAD57, RAD55 and RAD54, so Enter opened RAD59. The sequence a name means now heads the list wherever genes begin with it, read through every spelling the box accepts (chrIV, iv), and a name or accession typed whole comes before the names it begins. The map’s box, the whole genome’s and the 1D browser’s all follow it.

• A host’s error status is headed as the refusal it is (§10, §24.6). A 404 or a 403 to the HEAD a remote file is opened with was headed That URI does not parse, and its message sent a reader with a private bucket to Open a file → Private buckets, which no longer exists. A presigned link that had expired, which skips the HEAD, failed the range probe under That host does not serve byte ranges. Both are now headed The host would not give this file, with advice for a public file, a private bucket (connect a bucket…) and a presigned link, and an expired presigned link says that it is one; a host that answers a range with the whole file is still told it serves no ranges. The Google sign-in offer and the command line’s whole download, which read the same codes, are unchanged.

• The HiC-Pro conversion hint names --one-based (§3.9). HiC-Pro numbers its bins from 1, and cooler load -f coo reads a COO file’s bin ids from 0 unless it is told otherwise.

• A store from a newer build is called a store (§24.1), in its own format’s numbers: the refusal said workspace format 2 and the workspace version this build reads.

• The note for a hub on UCSC’s servers says what works (§24.10): a copy of the hub and its files on a host that sends the header, or the tracks dropped one by one. It sent the reader to a connected folder, which lists a hub.txt dimmed and opens no hub.

• The directionality index’s price refuses what its runner cannot score (§18.8). A bin needs a whole window each side, so a sequence must be longer than twice the window; the price refused only one no longer than the window, and a sequence in between was priced, read and then dropped without a score.

• The drawer says where a new 3D model goes (§25.3): a structure cell of its own, beside what the deck shows, rather than add one from the deck’s bar if none is open.

• SPECS, the notes under docs/ and the glossary say what the code does where they had drifted: the tutorial’s 58 steps (§30), which builds keep a connected folder (§24.2), the two 3D tasks’ ρ (§25.6), the desktop notes on compressed hosts, writing and what quitting leaves, the store generator’s anonymous listing, where the stored-data screen is, and the console’s key.

• The notebooks no longer fail on one failed download (§18.10). Both runs of the Notebooks workflow for 0.19.0 stopped a minute into the parity, on its first file: NCBI’s server for GEO answered 500 Internal Server Error, as it did that day to about one request in two, and now and then 403 Forbidden, to requests it answered in full a second later. Every file the four notebooks fetch now goes through notebooks/scripts/download.py, which asks again after a server’s error, a connection cut or timed out, or a body shorter than announced, with a wait that doubles to half a minute, resumes from the byte it reached where the server takes a range, and gives up after twenty failures in a row; the range reads and size checks of the timings’ viewers do the same.