Commits for braney

v502_preview2 to v502_base (2026-08-04 to 2026-08-10) v502

98ac3822afa0712f1897d17e5b7934c4aa742f57 Tue Aug 4 17:11:00 2026 -0700

• hgConfCatalog: report only the gates that changed this release, and settle the gate/knob calls, refs #37925 Groundwork for running the sunset report from the weekly build's wrap-up. --sunset prints the whole standing list every run, which reads as a worklist the first week and as wallpaper by the third. New --sunset-new reports only the gates that went overdue or stalled since hgConfGateBacklog.txt was last accepted, plus the ones that were cleaned up, and exits 1 when there is something new, the same contract --reconcile already has. --update-baseline accepts the current state, so the file's diff becomes the release-to-release history of the backlog. Gates already in the backlog are marked in --sunset too, so an unmarked row there is one this release created. Settled all eleven gate-or-knob calls that were filed debatable, each by reading the call site. forceTwoBit, freeType, sleepOn429 and trustTrackDb become knobs: what their off position describes is where a machine keeps its sequence, whether it has the URW fonts, whether it wants to hold a process after a 429, and whether its trackDb can be trusted. Three of those were on the overdue list, so the report was telling somebody to delete flags that should stay. autoBlatBigPsl becomes a gate: its default is the file-scope autoBigPsl, initialised FALSE under the comment "DEFAULT VALUE change to TRUE in future", and a flag whose own source says to flip it later is not a permanent switch. alwaysItemRgb and blatNewPageBanner stay gates, and hgHubConnect.validateHub stays a gate with its note corrected: it gates the Hub Development tab, not validation on attach. Overdue is 18 down to 15. Added a fourth reconcile question: does each flag's hand-written default still match the tree's. That field is the one part of a gate's lifecycle not read out of git, and gate_lifecycle decides a gate has shipped from it alone, so a flag flipped TRUE in code without the row updated sat in the stalled list under a heading telling somebody to turn on a flag that was already on. Zero mismatches today. A flag read with both TRUE and FALSE as its compiled-in default is reported under --verbose rather than counted as a problem, since the fix belongs to whoever owns the code: multiRegionButtonTop is read TRUE at hgTracks.c:9126 and FALSE at config.c:990, so on a default install the button is in the top bar while the checkbox the same flag guards is hidden. The age cache is overridable with --cache or HGCONF_AGE_CACHE, and hgConfCatalog.py takes --refresh, so an automated run can rebuild it without writing into a git checkout. Refreshed the committed cache, which dates blatNewPageBanner and showLiftRequest: both were added after it was last built, which is why they read as undatable. Documented forceTwoBit and freeType in ex.hg.conf, since calling them mirror knobs and leaving them undiscoverable is the same mistake in the other direction.
  • src/hg/utils/hgConfCatalog/harvestHgConf.py - lines changed 36, context: html, text, full: html, text
  • src/hg/utils/hgConfCatalog/hgConfAges.json - lines changed 57, context: html, text, full: html, text
  • src/hg/utils/hgConfCatalog/hgConfCatalog.py - lines changed 447, context: html, text, full: html, text
  • src/hg/utils/hgConfCatalog/hgConfGateBacklog.txt - lines changed 41, context: html, text, full: html, text
  • src/product/ex.hg.conf - lines changed 12, context: html, text, full: html, text

74dc80fd3bd2a70371fa75218347ddc45a556dde Wed Aug 5 07:28:59 2026 -0700

• Docent: render a tour at print resolution, and make zoom wait for its redraw refs #37892 scale: k (DOCENT_SCALE=k, make hires [SCALE=3]) renders the same tour with k times the pixels, for figures that have to print -- a screen still is about 120 dpi across a journal column. Nothing is upscaled; each layer that draws is asked for more: * deviceScaleFactor: k with the viewport left at its 1x CSS size, so the page lays out exactly as at 1x -- same line breaks, same jQuery-dialog width, same tooltip placement -- and rasterizes with k times the pixels, * pix x k so the server draws a wider browser image, with textSize stepped up its ladder to match (3x lands on 24) so hgTracks makes the same layout decisions in it: same tick spacing, same room for labels, same packing of features into rows, * zoom: 1/k on the image table, handing that wider image the 1x amount of layout space, so one image pixel falls on one device pixel. A scaled run is stills-only (no mp4), and DOCENT_STILLS names a different parent so a print render lands beside the screen stills instead of over them. Everything hgTracks reports about the image -- map-box coords, mouseOver spans, insideX, a px: drag endpoint -- is in the pixels the SERVER drew, which is not the displayed size once the image is scaled. Those now go through the image's natural-to-displayed ratio (1 at 1x). Without it a named mouseover: lands k times off and pins a different feature. zoom: in|out waited only for #imgTbl, which the buttons never remove (they redraw in place via ajax), so the following step could read the previous window's map boxes and report an item "not found" that was simply not in view yet. It now waits for the window itself to change. Pre-existing, hidden by the dwell that FAST removes -- and every hires run is FAST. An item-not-found error also names the current window and the items that ARE in that row, which is what told the two cases apart. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
  • src/hg/utils/docent/README.md - lines changed 46, context: html, text, full: html, text
  • src/hg/utils/docent/docent.js - lines changed 143, context: html, text, full: html, text
  • src/hg/utils/docent/docent.mk - lines changed 20, context: html, text, full: html, text

1ffe5f2fad53035d54f7b37c4da55b1767ec4d62 Wed Aug 5 10:15:26 2026 -0700

• hgConfCatalog: keep the file:line citations true and register new settings without asking a developer, refs #37925 Code review caught autoBlatBigPsl's row citing hgBlat.c:2629, an unrelated function, while the note in that same row already named the real read at 2972. Sweeping every citation, with the CFG_ and #define indirection resolved so a macro-spelled read is not a false positive, found 66 of the 269 line-cited rows pointing at the wrong line. They cluster by file (+343 across hgBlat.c, +11 across hgTracks.c, +2 across hdb.c), so this was the tree moving under the catalog rather than anybody mistyping. All 66 now match a site the harvester finds on its own. Rows whose citation opens a multi-line cfgOptionEnvDefault call were already correct and are untouched, as are the four whose only literal lives in a #define. reconcile() grows a fifth check so this cannot drift again, and --fix-citations repairs what it finds: line drift prints a count and the command, while a read that has left its cited file lists the rows, because choosing among call sites needs a person. --auto-register writes a row for any setting the tree reads that the catalog lacks. The developer who adds an hg.conf read should not have to know this catalog exists, and asking them to had not worked: hubSpaceLockTimeout arrived as one supporting line inside an unrelated bugfix and nothing noticed until --reconcile was run by hand a day later. It copies only what the call itself states, plus the introducing commit and ticket read out of git, and refuses every judgement. role= in particular is left unset on purpose: calling a release gate a knob would hide it from the sunset report for good, and calling a knob a gate would feed fiction into the overdue arithmetic. Rows land in a new "Awaiting review" section, which --reconcile counts as needing attention until somebody classifies them, so the silence is bought for the developer and not for us. nightlyRegister.sh runs both passes nightly and commits. It works in the weekly build's own tree, so it acts only on master (cherryPickCommits.csh and tagBeta.csh both check out release branches in there), never leaves the tree dirty for ensure_clean_git() to abort on, stands aside while /tmp/autoBuild.lock is held, and throws its own commit away rather than leave one unpushed if the push is rejected. hubSpaceLockTimeout is classified here as well: a limit, 300 seconds by default, bounding the flock poll in lockHubDir.
  • src/hg/utils/hgConfCatalog/hgConfCatalog.py - lines changed 452, context: html, text, full: html, text
  • src/hg/utils/hgConfCatalog/nightlyRegister.sh - lines changed 173, context: html, text, full: html, text

3ea2ef2ddf64b7e366af1e8c0e500ea1fe88a4e6 Wed Aug 5 10:55:25 2026 -0700

• gwEditPage: read and write genomewiki page text from the command line gwUploadFile next door handles images, but it needs python2 and mwclient, neither of which is installed for python3 on hgwdev, and nothing in the tree could edit page text at all. This talks to the MediaWiki API with the standard library alone. The trap it exists to document: two wikis run behind genomewiki.ucsc.edu, the public one at / and the internal Genecats one at /genecats, and they keep separate user databases. An account that works on one may not exist on the other, so the failure reads like a mistyped password when it is really the wrong wiki. Credentials keep gwUploadFile's ~/.gwLogin convention, one line and mode 600, with ~/.gwLogin.genecats for the internal wiki so the first line of ~/.gwLogin stays exactly what gwUploadFile expects to find. A bot password from Special:BotPasswords works in place of an account password and is preferable: scoped to the grants chosen, and revocable on its own. get prints wikitext and put replaces it, requiring a summary and refusing to write an empty page. --dry-run shows the diff and writes nothing, which is worth the habit: it caught two wiki-markup mistakes on its first real use, a blank line between list items ending the list, and text appended past the Category tags landing outside the section it belonged to. Reading needs no account on either wiki, so it does not ask for one.
  • src/hg/utils/automation/gwEditPage - lines changed 189, context: html, text, full: html, text

9ea569dd7b6b5f189bd5bec3cebc37ab95cb88fb Wed Aug 5 10:55:33 2026 -0700

• autoBuild.sh: note that something else writes to the build tree now, refs #37925 nightlyRegister.sh runs from the build account's crontab and commits and pushes hgConfCatalog.py out of $BUILDHOME/kent, so "hgConfCatalog: register ..." commits appearing on master with nobody behind them are expected rather than a stray edit somebody left in the tree. The note sits at ensure_clean_git() rather than in the header because that is the function whose failure would be puzzling: it aborts the whole build on any uncommitted change in that tree. The writer is built not to trip it, acting only when HEAD is master (cherryPickCommits.csh and tagBeta.csh both check out release branches in there), restoring the file on any exit that did not commit, and standing aside entirely while the lockfile is held. The note says what to do in the one case where it does trip: the script died between writing and committing, git checkout is the whole fix, and the next run redoes the work. Also written up on the CGI_Build_Process wiki page, under "The build tree has a second writer".
  • src/utils/qa/weeklybld/autoBuild.sh - lines changed 15, context: html, text, full: html, text

9e183778db2595f86065c84518a68f4490c7b375 Wed Aug 5 11:03:35 2026 -0700

• nightlyRegister.sh: say one line every night while this job is still new, refs #37925 A job that only speaks up when something is wrong is a job nobody notices has died. This one was written to be quiet on a clean night, which is right for the long run and wrong for its first weeks, when the failure nobody would catch is the cron never running at all. So it now emits one line per run, whatever happened, carrying enough state to be worth reading rather than being a bare pulse: 2026-08-05 03:30 nothing to register, 0 awaiting classification, tree at a74f3a3 The same line is appended to $WEEKLYBLD/logs/hgConfRegister.history, the build's own untracked log directory. The mail is what keeps the job in mind; the history outlives the mail, so a gap in the dates shows when it stopped. Set HGCONF_NIGHTLY_HEARTBEAT=no in the crontab once the mail is boring and the history keeps going without it. A failure records its own line, so a run that died is distinguishable from one that never happened. The skip paths beat as well, which matters more than the ordinary nights: until now "a build is running" and "the tree is on v501_branch" both exited in total silence, which is indistinguishable from a dead cron during precisely the week nobody wants to be wondering. Counts come from the --reconcile that already runs after the writes, so the awaiting-classification figure includes anything registered the same night rather than lagging a day behind.
  • src/hg/utils/hgConfCatalog/nightlyRegister.sh - lines changed 51, context: html, text, full: html, text

140e29ae6894cc8d55bb231e73e7a16f7e51a146 Wed Aug 5 14:20:32 2026 -0700

• Docent: keep a print render's tooltips and track heights in proportion Two things stayed 1x in a 3x still. hgTracks takes the tooltip's font-size from the browser text size, which a scaled run has already multiplied by k for the image, and then the device pixel ratio scales that same text a second time -- so the popups came out k times too big, swamping the figure, and the last tooltip pinned fell off the crop. Pin the tooltip font-size back to the 1x value. The other is height. Neither pix nor textSize reaches a track whose height is a fixed pixel count, so a 128px bigLolly row that was 15% of an 850px image was 5% of a 2550px one -- ClinVar's lollipop row came out a sliver with unreadable y-axis labels next to a bed track that had grown with the font. After each view change, ask for k times the height of every row the page actually drew. Going by the drawn names is what reaches a lifted view, whose tracks are hub tracks under names trackDb never saw. Each track's own maxHeightPixels still clamps the request, so a track that should stay short does. refs #37892 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
  • src/hg/utils/docent/README.md - lines changed 14, context: html, text, full: html, text
  • src/hg/utils/docent/docent.js - lines changed 52, context: html, text, full: html, text

f066bc4cf3adb12de8d6c65f41b47c2e463ed27e Thu Aug 6 10:07:24 2026 -0700

• hgConfCatalog: escape what the auto-writer splices, and count the right rows, refs #37925 Two things a code review caught in the nightly job, neither of which had gone wrong yet and both of which would have gone wrong quietly. --auto-register writes Python into this file and the file is imported again the next time anything runs, so free-form text in a generated row has to survive being a string literal. Most of it does; the git commit subject does not. A subject that quotes a setting name, which is an ordinary thing to write, would put a bare " in the row and break every import of the catalog tree-wide, and it would land on master unreviewed because running unattended is the whole point. Escape the backslashes and quotes and drop the control characters on the way in. The note is wrapped before it is escaped so the added backslashes, which nobody reads, do not push a line over. The heartbeat's "awaiting classification" number came from reconcile's aggregate problems: total, which also counts wrong hand-written defaults and citations whose call site left the file. Right now the tree has seven drifted citations and an empty holding pen, so tonight's line would have claimed seven settings were waiting on somebody when none were. Read the count off the "written down but not classified" line instead. A heartbeat exists to be believed later by somebody skimming the history file, and a wrong number there is worse than none.
  • src/hg/utils/hgConfCatalog/hgConfCatalog.py - lines changed 21, context: html, text, full: html, text
  • src/hg/utils/hgConfCatalog/nightlyRegister.sh - lines changed 8, context: html, text, full: html, text

babf25e13ff2e01e1f2a3c295f7b84420d3de421 Thu Aug 6 10:07:35 2026 -0700

• Docent: say which height the print scaling actually scales, refs #37892 The k*128 a scaled run asks for is the default heightPer times k, not each track's own height times k, so a row deliberately configured at 50px or 300px comes back the same size as everything else instead of in proportion. Every track a tour has used took the default, so this has not bitten, and fixing it properly means reading each row's heightPer out of the cart before asking for k times it. Write down the assumption and what the symptom looks like -- a row that is the wrong size at k and the right size at 1x -- so the next person to hit it knows where to look rather than rediscovering the constant.
  • src/hg/utils/docent/docent.js - lines changed 6, context: html, text, full: html, text

3b13914cf3079b1c2211f14d5295558c5e4d3cc7 Thu Aug 6 10:08:04 2026 -0700

• hgConfCatalog: refresh the file:line citations, refs #37925 Seven rows in hg/hgLogin/hgLogin.c drifted when the OIDC provider and login.emailLink work inserted lines above them: six by one line, pwdEyeIcon by five. Nothing about the settings changed, and hgLogin.c:81 just landed on a bare brace instead of the cfgOption call the row names. Written by hgConfCatalog.py --fix-citations, which only moves a line number onto the read the row already cites. The nightly job would have done this on its own; doing it here gets the count out of reconcile's problem total.
  • src/hg/utils/hgConfCatalog/hgConfCatalog.py - lines changed 7, context: html, text, full: html, text

ffccd9cceb44efbd4d07dab8eda90e95ccd72d77 Thu Aug 6 13:45:12 2026 -0700

• Docent: pinned-mouseover cursors, and a montage verb for multi-panel figures, refs #37892 A pinShot puts several tooltips on one still, but nothing in the picture says which feature each was raised from -- the reader infers it from an 8px anchor offset, and on a dense row that is a guess. recordTip now keeps the hover point alongside the tooltip's own offset and pinShot draws a static pointer there. The glyph and its box move out of CURSOR_INIT into constants the two share: a pinned pointer that did not match the animated one would read as a different cursor rather than as the same tour paused, and that is the kind of drift nobody notices until the figure is in a proof. pinShot takes a map form, {name: x, cursors: false}, for a figure that does not want them, and the pointers join the crop's bounding box so one on the bottom row cannot fall off. The other half is composition. A journal wants parts (A), (B) as a single file, so the panels had to be assembled by hand after the run. That puts the figure's layout outside the script: rename a shot and the montage quietly drops a panel instead of failing, which is the sort of thing found at submission. montage: {name: x, shots: [a, b]} does it in the tour, so the composite is a product of the same run as its parts. Panels stack in order and letter themselves; labels:, direction:, gap: and labelSize: override. Composition happens in a browser page at deviceScaleFactor 1 with every panel at its natural pixel size, so the result is pixel-for-pixel its inputs -- a make hires montage is print resolution because the panels were, not because anything was upscaled. Panels narrower than the widest are left-aligned and padded rather than stretched, and a named shot that was never taken warns and is skipped.
  • src/hg/utils/docent/README.md - lines changed 2, context: html, text, full: html, text
  • src/hg/utils/docent/docent.js - lines changed 103, context: html, text, full: html, text

25ff975568a24a89dc2dcd4836875a448beaec9f Thu Aug 6 15:23:11 2026 -0700

• Docent: hide a composite as a unit under hideKids, and warn on an over-long nav, refs #37892 `{cCREs: hideKids}` on hg38 sent 1701 cart variables in a 42,020-character GET. The walk behind hideKids went to true leaves, straight through the ENCODE4 Core Collection composite, and enumerated all 850 of its ENCFF subtracks. Apache's LimitRequestLine is 8190, so the server answered 414 and the next shot: photographed "Request-URI Too Long" -- and nothing failed: the page loaded, the still got written, the build exited 0. Only looking at the figure caught it. Hiding a composite already reaches its subtracks, so the expansion now stops at the first container that propagates its own visibility and keeps descending only through superTracks, which do not. hubApi never lists a superTrack container, so the flag tdbParse already sets when it synthesizes one answers the question without new data. That step is two variables now. This is a separate walk rather than a change to tdbLeaves, whose other caller resolves a container to a row on the page and does need true leaves -- a composite has no img_data_ of its own. nav() also warns when a URL passes 7800 characters. A 414 is not an exception, it is a page, so the run continues and the damage is a wrong figure; the one thing that helps is saying so at the time.
  • src/hg/utils/docent/README.md - lines changed 1, context: html, text, full: html, text
  • src/hg/utils/docent/docent.js - lines changed 43, context: html, text, full: html, text

6a22efd821c976be180e03c187be89d1155ad41a Sat Aug 8 09:04:15 2026 -0700

• hgConfCatalog: correct the blatNewPageBanner default and classify blatOldTracks, refs #37925 blatNewPageBanner was born TRUE at c683ecb63d7 and put back to FALSE at 81d7cff6cb6 while the new BLAT page is still being tested. The catalog kept saying TRUE, and gate_lifecycle reads that field alone to decide a gate has shipped, so the row would have put a gating flag in the shipped list. blatOldTracks came in from --auto-register and needed a person. It is a three-value string, not a flag, so no gate/knob role applies: it chooses what happens to the custom tracks left by a user's earlier BLAT searches. Filed under mirror knobs, whose description now admits the occasional non-boolean.
  • src/hg/utils/hgConfCatalog/hgConfCatalog.py - lines changed 34, context: html, text, full: html, text

61d7da07a793fe0395aadfe566c0d0c3e70762f3 Sat Aug 8 09:04:24 2026 -0700

• urlCommandCatalog: describe the hgLogin and BLAT share parameters, refs #37923 The social-login, email-link and BLAT share-link work added 24 names the registry had not seen. Ten of them are read with cgiUsualString or cgiOptionalString and so are genuine URL parameters: provider, code, state, error and error_description on the provider's redirect back to hgLogin, newEmail, exp and sig on a signed change-of-email link, and blatPslId and blatFaId on a stable BLAT share link. Those get rows. The other fourteen are read only from the cart: the oauth_pending_ family, oauth_state, and the hgLogin form fields. They go in the baseline. Four of the ten are leaks. blatPslId and blatFaId are not in hgBlat's excludeVars, which still lists only the older u and s share parameters, and error and error_description are not in hgLogin's. Marked as such; the fix belongs with whoever owns the leak list.
  • src/hg/utils/urlCommandCatalog/urlCommandCatalog.py - lines changed 71, context: html, text, full: html, text
  • src/hg/utils/urlCommandCatalog/urlNamesNotCataloged.txt - lines changed 14, context: html, text, full: html, text

5d76e853fa84baa6c4a05880d3f3b985a5e51878 Sat Aug 8 13:22:15 2026 -0700

• docent: add session, loadSession and expect verbs, refs #37892 session: writes the whole cart at that step to sessions/<base>/<name>.txt, in the format hgSession's save-to-a-local-file button produces. No login is involved. With sessionUrlBase: set, the run also prints the load URL, so a tour hands back the state it built instead of only a picture of it. A file rather than a live hgsid link, because the hgsid cart keeps changing as the tour runs. loadSession: is the other direction: a settings file by URL, a share link, a named session, or a local file written by an earlier session:. The load always goes to the server named in target:, since named sessions are per-server and every later step navigates to target: by absolute URL. expect: is the only verb that can fail a run. It checks the rows drawn, the image height, the tooltip now up, and the page text, then throws, so make stops at that step rather than writing a wrong figure over a right one. A session taken on a lifted view is short-lived: it names the quickLift hub by its trash path, and re-pointing it at a durable copy does not work today. See #38046. The README says so rather than working around it.
  • src/hg/utils/docent/README.md - lines changed 126, context: html, text, full: html, text
  • src/hg/utils/docent/docent.js - lines changed 197, context: html, text, full: html, text
  • src/hg/utils/docent/docent.mk - lines changed 7, context: html, text, full: html, text

f963b73576b5c69915366893da7dfa6afe633456 Sat Aug 8 14:04:23 2026 -0700

• docent: add a tests directory and a browser-free derive mode, refs #37892 tests/ holds Docent scripts that assert with expect:, run by hand with `make test` rather than by the tree's test target, since each one drives a real server. Nine of them: the two-request composite split (#37953), hideKids on a view and on a superTrack, the cCREs expansion that once overran the request line, addCustomTrack, a 3x run, and the session/loadSession round trip. A script named *.xfail.docent.yaml is expected to fail, which is how the hideKids-aimed-at-the-composite trap is pinned rather than only written down, and how expect: itself is checked. DOCENT_DERIVE=1 prints what each track: step turns into and stops, with no browser and no navigation. That derivation is where most of Docent's own decisions are, and it was previously visible only in the log of a full run. `make derive` diffs it against baselines in tests/expected/ for the scripts whose derived set is small enough to be stable. The track: verb now calls trackRounds() for that derivation instead of doing it inline. No behaviour change intended; the tests above pass before and after. Two things the tests turned up, both recorded in tests/README.txt: turning on anything under a superTrack sends <superTrack>=show and undoes an earlier hide: all for its other members, and hideKids on a view has to enumerate leaves, so one such step sends 188 variables in a 6,986-character request.
  • src/hg/utils/docent/README.md - lines changed 13, context: html, text, full: html, text
  • src/hg/utils/docent/docent.js - lines changed 148, context: html, text, full: html, text
  • src/hg/utils/docent/tests/README.txt - lines changed 86, context: html, text, full: html, text
  • src/hg/utils/docent/tests/composite.docent.yaml - lines changed 25, context: html, text, full: html, text
  • src/hg/utils/docent/tests/customtrack.docent.yaml - lines changed 22, context: html, text, full: html, text
  • src/hg/utils/docent/tests/expected/composite.derive - lines changed 6, context: html, text, full: html, text
  • src/hg/utils/docent/tests/expected/supertrack.derive - lines changed 4, context: html, text, full: html, text
  • src/hg/utils/docent/tests/expected/urllen.derive - lines changed 3, context: html, text, full: html, text
  • src/hg/utils/docent/tests/expectfail.xfail.docent.yaml - lines changed 17, context: html, text, full: html, text
  • src/hg/utils/docent/tests/makefile - lines changed 102, context: html, text, full: html, text
  • src/hg/utils/docent/tests/scale.docent.yaml - lines changed 22, context: html, text, full: html, text
  • src/hg/utils/docent/tests/selftest.docent.yaml - lines changed 30, context: html, text, full: html, text
  • src/hg/utils/docent/tests/supertrack.docent.yaml - lines changed 19, context: html, text, full: html, text
  • src/hg/utils/docent/tests/urllen.docent.yaml - lines changed 20, context: html, text, full: html, text
  • src/hg/utils/docent/tests/views.docent.yaml - lines changed 39, context: html, text, full: html, text
  • src/hg/utils/docent/tests/views.xfail.docent.yaml - lines changed 21, context: html, text, full: html, text