main -> dev - #7194
Merged
Merged
Conversation
…6654) Jump-box searches for family names that coincide with a stored family, e.g. Sp(2,5)=SL(2,5) or PSp(2,5)=PSL(2,5), flashed "has not yet been added" although the groups are in the database under the other name. Add a FAMILY_ALIASES table used by group_jump to translate the gps_special_names lookup for the exact coincidences Sp/PSp/GSp/ASp/ PSigmaSp/ASigmaSp in dimension 2 and Spin in dimension 3; the valid_params fallback still sees the family the user typed so error messages stay accurate. Also refresh stale family names in valid_params (CSp->GSp, GO->Orth, PGO->PO, CSO->GSO, CO->GOrth, add ASp) to match current gps_families, as already done in web_groups.py. Verified with a test-client script (Sp(2,5) and Sp(2, 5) -> 120.5, PSp(2,5) -> 60.5, GSp(2,7) -> 2016.a, Spin(3,4) -> 60.5, stored families such as Sp(4,3) unchanged) and new checks in test_browse_page.py::test_lookup; pytest -k "lookup or famly" passes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Knowls clicked inside a table cell insert a full-width row whose content (min 700px via knowl_click_handler in lmfdb.js) stretched narrow tables, pushing the other columns apart on pages like /rcs. Add inline-size containment to td > .knowl-output so the inserted row is ignored when column widths are computed: the knowl box now simply spans the table's current width, and over-wide content scrolls horizontally inside the box instead of stretching the table. Verified with a local server and browser on /rcs (table width and column positions identical before/during/after opening source and reliability knowls, previously 526px -> 692px with columns jumping ~100px), on NumberField search results (th knowl in results header) and on the EllipticCurve/Q/11/a/2 invariants table (KaTeX-heavy td knowl); paragraph knowls are unaffected. lmfdb/tests/test_root.py (renders /style.css) passes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Only quadratic p-adic fields had a readable nickname (Q_p(sqrt d)).
Nickname every unramified extension as Q_p(zeta_{p^f-1}) and every tame
extension as Q_p(zeta_{p^f-1}, root[e](zeta^k p)); the f=1 radicand is
written p*r to match the existing sqrt(p*u) convention, with k=0 and
zeta^k=-1 simplified. The twist exponent k is read from the stored
Eisenstein/unramified data (no discrete log) and normalized to a canonical
Frobenius-orbit representative so Q_p-isomorphic presentations coincide;
wild fields keep their label. Nicknames now show on field homepages, the
field-data knowl, subfield/completion displays, and the families Base column.
Verified with sage/pari over all 7731 tame degree>2 fields: 0 mismatches
vs an independent discrete-log computation, per-(p,e,f) counts equal the
number of Frobenius orbits, and each nickname field is Q_p-isomorphic to
the stored field. pytest local_fields: 9 passed (new test_nicknames);
pyflakes clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…onstruction log files
Signed-off-by: ahmadalguydi <ahmadalgaidy@hotmail.com>
The jump-box tests only exercised four of the seven FAMILY_ALIASES entries, and none of them reached valid_params, so the refreshed family names there were untested. - test_family_alias_lookup: a stored example for each FAMILY_ALIASES entry, including the previously untested ASp(2,2), PSigmaSp(2,5) and ASigmaSp(2,2), plus an assertion that the case list and FAMILY_ALIASES stay in sync. - test_absent_family_lookup: valid but absent groups in both parity branches of valid_params, covering the newly accepted GSp/ASp and the renamed Orth*/PO*/GSO*/GOrth* identifiers, and two wrong-parity names that must still be reported as invalid. Targets verified against gps_special_names (ASL(2,2)=ASigmaL(2,2)=24.12, PSigmaL(2,5)=60.5) and every identifier in the two valid_params lists checked against the distinct family values in gps_families. Verified with sage -python -m pytest lmfdb/groups/abstract/test_browse_page.py -k 'lookup or family or famly': 5 passed, 68 deselected; pyflakes clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Review of PR #20 pointed out that the added `overflow-x: auto` on `td > .knowl-output` turns the box into a scroll container, which exposes the scrollable overflow of the absolutely positioned `.knowl-footer` (`width: 100%` starting after the parent's 10px left padding, plus its own 10px left padding). Measured in the browser, that is 20px of scroll range on every table knowl, so even a short knowl whose content fits gained a horizontal scrollbar (box height 123px -> 138px). The declaration was also redundant: `.knowl-content` already has `max-width: 100%; overflow: auto` and is the right inner scroller for over-wide content. Drop `overflow-x` (so the box keeps the inherited `overflow: hidden`) and reword the comment accordingly. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
PrimeBound.__call__ verified that the queried values were finite sets of
primes, but never checked them against the bound it was constructed with.
So the registered ec_curvedata checker
("conductor", PrimeBound(300000000), "elliptic curves with prime conductor at most 300 million")
certified any prime conductor as complete, however large:
sage: results_complete("ec_curvedata", {"conductor": 1000000007}, db)
(True, 'elliptic curves with prime conductor at most 300 million', None)
Delegate the range check to Bound.__call__, as the sibling CPrimeBound
already does, and add regression tests for a prime past the bound.
The bound is checked before primality so that a large range is rejected
without iterating over it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Check the bound in PrimeBound
…-7038 fix: show exponent one for cusp orbit sizes
Follow-on to the prose pass, kept separate so it can be dropped on its own. Every name here has all of its references inside a single file (or, for favourite_list, one view and its template), so each rename is the definition and its uses together: draw_spectrum.py centre_ratio, y_centre -> center_ratio, y_center WebEllipticCurve.py EC_R_plot's colour parameter -> color web_newform.py factorisation loop variable -> factorization galois_reps.py renormalise_coefficients -> renormalize_coefficients search_wrapper.py query_cancelled_error -> query_canceled_error hecke_algebras favourite_list -> favorite_list (view + template) test_st.py test_favourites -> test_favorites query_cancelled_error is worth a look: it handles psycopg's QueryCanceledError, which spells it with one l, so the handler and the exception now agree. Not renamed: signtocolour, which lmfdb/utils/__init__.py exports in __all__ and five modules import. Renaming it is an API change rather than a spelling fix, so it and the locals built on it (signcolour, thiscolour, colourplus, colourminus, test_signtocolour) are left for a separate decision. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The test suite runs against a shared server, devmirror by default, whose statement_timeout is 0. psycodict only sets a timeout for connections made as the "webserver" role, and the tests connect as "lmfdb", so nothing bounds a query the tests issue. A slow one runs until something outside the job stops it. That is not hypothetical. On 2026-08-06 the Dirichlet character search in lmfdb/characters/test_characters.py held five separate CI jobs for between 26 minutes and 2h33m; the run that started at 08:50 UTC finished at 11:24, and four others that had started as much as two hours later finished within a minute of it, all reporting a cancelled query. The same search takes about seven seconds against an idle server. Set PGOPTIONS in the tests environment. libpq applies it to every connection it opens, including the ones psycodict opens when it resets a broken connection, so the timeout cannot be lost on a reconnect and the test harness needs to know nothing about it. A query stopped this way raises the QueryCanceledError the search pages already catch, so a slow search now renders the ordinary timeout page in 120 seconds rather than hanging. 120s is chosen as a hang catcher rather than a performance budget: well above anything the suite issues against a healthy server, well below the hangs it exists to stop. Exporting PGOPTIONS overrides it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
DirichletSearchTest.test_search is the test that has been hanging and then failing CI. Its conductor and order ranges match every modulus admitting a character of conductor 25 to 50: 537,164 orbits, all of which must be read before the sort by modulus can pick the first page of 51. EXPLAIN ANALYZE on devmirror puts that at 434,307 buffer reads, about 3.4GB, for 6.7 seconds against an idle server. That cost is inherent to the query rather than a regression, and this test exists to check that the search parses conductor ranges, order ranges, the primitive flag and parity, not to measure the server. Use the check_args_with_timeout helper, which accepts either the expected label or the timeout page, for the five range searches. The first search fixes both conductor and order, which matches an index prefix and lets the index supply the sort too, so it returns from a handful of index entries. Leave that one strictly asserted so the test cannot pass vacuously if the database is unreachable. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
GitHub's default job timeout is 360 minutes. A job blocked on the database therefore holds a runner for most of a working day before anyone hears about it: one on 2026-08-06 ran 5h22m, and several others sat for hours behind the same query while their eleven sibling jobs finished normally in 3 to 20 minutes. A healthy job finishes well inside half an hour, the slowest green ones observed being around 25 minutes including building the conda environment on a cache miss, so 60 minutes leaves ample headroom while turning a lost morning into a prompt red build. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The searches this test issued used primitive=Yes|No and parity=Odd|Even. common_parse reads the primitivity flag from is_primitive via parse_bool, which accepts "yes"/"no" and not "Yes"/"No", and compares parity against lowercase "even"/"odd". None of those four parameters were recognised, so all five searches ran the same unfiltered conductor and order query. They passed because 25.d, 50.d and 56.n are all on that one query's first page. Confirmed against the live site: /Character/Dirichlet/?conductor=16&order=4&primitive=no still returns the primitive orbits 16.e and 16.f. That was already the case before this branch, but making the searches tolerate the timeout page made it worse: five identical expensive scans, none of which had to match a label to pass. Use the parameter names the parser reads, and check each filter in both directions against conductor 16 and order 4, which has a primitive even orbit (16.e), a primitive odd one (16.f), and imprimitive even and odd ones at modulus 32. A filter that stops being parsed widens the search and trips a negative assertion. Labels carry their surrounding tags because 416 is also a multiple of 16, so a bare '16.e' matches the '416.e' further down the page. Both columns are pinned to single values in these searches, which matches an index prefix and leaves the index supplying the sort, so they replace the expensive scan rather than repeating it. One range search remains, and it is the only one allowed to settle for the timeout page. The test issues six requests, the same as before. Verified against devmirror: passes, and the four label sets are exactly as asserted. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The parsing-bug history and the index-prefix reasoning belong in the commit message and the PR, not in the file. Keep only what a reader of the test needs: why each filter is asserted both ways, why labels carry their tags, and why the last search tolerates a timeout. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Stop CI jobs hanging for hours on a slow database query
GitHub Actions was disabled during a recent outage; this empty commit kicks the checks back into running.
Use American spelling throughout
group_jump matches the input against the regexes in gps_families, which spell each family the way it is stored, but neither name printed on a group page always agrees with that spelling. The "Groups of Lie type" row prints the family's tex_name, and for the orthogonal and unitary families that uses other letters (Orth prints as O, GOrth as GO, Unitary as U, GUnitary as GU) and a +/- exponent for the Plus and Minus families; the code snippets print magma_cmd, which differs again (Orth is Magma's GO, GOrth is Magma's CO). So 103680.a shows O(5,3) and GO(5,3) in its Lie type row and GO(5,3) and CO(5,3) in its snippets, and none of those four could be pasted into the jump box. Add FAMILY_NAME_ALIASES, rewriting the family part of a Fam(n,q) name onto the stored spelling before the gps_families regexes are tried. Magma's GO, GOPlus, GOMinus and GU are left out: we print GO and GU for the conformal groups Magma calls CO and CU, so honoring Magma's reading would return a different group from the one shown under that name. No alias is an existing family name, so every input accepted today still matches as before, and the error messages keep the user's own spelling. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The redirect tests could not catch a wrong target on their own: families such as Orth and GOrth agree for many small parameters, so a typo like "CO": "Orth" still redirected to the expected group. Pull the rewrite out of group_jump into normalize_family_jump, so it can be tested as a pure function, and add three tests that fix the meaning of the table rather than just exercising it: an independently written copy of the expected mapping, a check that each +/- alias agrees with the alias it is the exponent form of, and a check of the Magma half against the magma_cmd column the code snippets are generated from, which also asks for an alias whenever a family's Magma name differs from its own. Strengthen the unclaimed-alias test to probe the gps_families input regexes at both parities and two field sizes, since those regexes are what decides which inputs are accepted. Move the redirect examples to parameters where an isometry group and its similitude group have different labels. Deliberately mutating "CO", "GU" and "GO+" to Magma's reading now fails the tests. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Accept the family names printed on group pages in the abstract groups jump box
Add nicknames for unramified and tame p-adic fields
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add an optional database-independent precheck hook to CompletenessChecker, running after $or/$and decomposition and before any table lookup or null-count query. Register an NFBound precheck that recognizes a root discriminant range lying above the Galois root discriminant range, so the intrinsically impossible query no longer waits on the expensive null-data probe that CI's statement timeout cancels. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The devmirror shards share one database; cap concurrent matrix jobs at six so functional CI does not double as an uncontrolled load test. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Raise the CI database statement timeout to 10 minutes
psycodict predicates involving None carry SQL-null semantics ({"$ne": None}
means IS NOT NULL), which to_rset misreads as empty numeric sets: None maps to
the whole real line, so its complement is empty. The precheck could therefore
declare a nonempty search (e.g. grd IS NOT NULL, or "non-null or at most 30")
intrinsically impossible without touching the database.
Conservatively bypass the numeric precheck whenever None appears anywhere in
the rd/grd expression tree, using a sentinel so an omitted constraint stays
distinguishable from an explicit null predicate. The numeric contradiction
fast path is unchanged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…urrency Avoid needless completeness queries and cap CI database concurrency
Add Sage and Oscar download links for abstract groups
Fixes by autopep8 action
Completeness for NF
Add "verified" annotations for the 97 genus 2 curves where Sha is rigorously known
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.