Highlight active search constraints on results pages

[roed-math]

Closes #2770.

On search results pages, the inputs that are actually constraining the results being displayed are now tinted (text boxes and selects) or ringed (checkboxes), so it is visible at a glance why the results are filtered.

How it works

Marking happens server-side in the shared search_boxes.py machinery, so every section gets it. SearchBox._attrs(info) emits exactly one class attribute per input, merging the box's own classes with advanced, search_constraint (this input constrains the results when filled in) and search_active (it is filled in right now). Class names are passed with a first-class classes= argument rather than hand-written into extra, since two class attributes on a tag is invalid HTML and browsers silently discard one of them.

search_active describes the search that produced the rows on screen, and is deliberately server-owned: nothing rewrites it as the user edits the form, so the highlight never claims a constraint that has not been submitted. The existing fresh/stale "Search again" button remains the signal that the form has been edited.

What counts as a constraint

is_constraint is a bool, or a function of info for boxes whose role depends on the rest of the search.

Not constraints: result count, sort and column controllers, dynamic-statistics variable boxes, diagram axis/colour/count boxes, hidden inputs, and the trace and Euler table display controls (n, n_primality, view_modp) in classical modular forms and L-functions. That last group includes the n_primality=primes that set_Trn inserts on every trace search, which would otherwise look filled in when the user never touched it.

Contextual, via a callable:

• an_modulo counts only when there are trace constraints for it to turn into congruences, not when it merely drives view_modp=reductions.
• The hypergeometric prime p counts only when one of the Ap, Bp, Apperp, Bpperp boxes is filled, since it only picks which columns those compare against.
• The abelian variety geometric-decomposition checkbox counts only when one of the inputs it retargets is used.
• A select beside a text box (TextBoxWithSelect) counts only once that text box is filled. This is safe because SearchParser.__call__ returns immediately on an empty field, so a select handed to a @search_parser as mode= cannot affect the query alone. Six selects are read directly during parsing and do constrain by themselves, so they opt out with is_constraint=True.

Colours

Two colour-scheme keys, so both stay overridable per scheme: search_active_background (#FFF9C4) for the tint and search_active_ring (#F9A825) for checkboxes, which are rendered natively and cannot be tinted. The ring uses box-shadow rather than outline so the browser's keyboard focus indicator survives. The tint uses background-color rather than the background shorthand so it cannot reset a background image. The exact shades are a maintainer taste call and easy to tune.

Two bugs found and fixed along the way

Both were pre-existing and unrelated to the highlighting, but each was directly in the path of deciding whether an input constrains the results.

• Classical modular forms level type. The dropdown offers "prime squared" and "powerful", but common_parse recognised neither, so both fell to the invalid-type branch. That branch flashes an error and returns a redirect, and both callers use the query common_parse builds while discarding its return value, so the search then ran with no condition on the level at all. mf_newforms and mf_newspaces already have level_is_prime_square and level_is_powerful, so the two values only needed adding to the list. The invalid branch now raises ValueError after the flash, the way the rest of that module reports bad input.
• Abelian variety geometric decomposition. CheckBox renders value="yes" but the parser tested == "on", so ticking the box in the form did nothing: the search used the ordinary dim*, number_fields and galois_groups columns instead of the geometric ones. It now accepts both values, keeping hand-written on links working.

Tests

New SearchBoxTest in lmfdb/tests/test_utils.py covers the rendering rules. Integration assertions in test_hgm.py, test_cmf.py, test_cmf2.py, test_lfunctions.py, test_modular_curves.py and test_av.py check the rendered pages, plus direct common_parse query-construction tests for the geometric decomposition checkbox. Tests compare class tokens as sets through a shared helper that asserts there is at most one class attribute, so nothing depends on ordering. No new test files, so the CI count is unchanged.

---

Ported from roed-math#23, where the full write-up and comment history live.

🤖 Generated with Claude Code

[jenpaulhus]

I like this addition! Just a note that on genus 2 curve searches, the two include/exclude boxes never get highlighted. That might be fine (as the box immediately above each of them does change colors) but wanted to point it out in case that was not intended.

[jwj61]

I also like it. I looked at it with a blue color scheme. Have others looked with others?

[roed314]

@havarddj had some opinions on the color scheme, so we should wait to merge this until we get some input from him.