UFAL/Subject keywords autocomplete stores an unsplit ';'-separated list as a single value

[milanmajchrak]

Original issue: ufal/clarin-dspace#1387

Problem

Some items store an entire keyword list in a single dc.subject metadata value (one long ;-delimited string, ~18 keywords) instead of one value per keyword. Confirmed on the DH Data collection (example item 615ee655-974e-467b-a36e-cc8a9a505430, hdl 20.500.12800/1-6204, accessioned 2026-07-14 — a fresh UI submission; already fixed manually).

Root cause (verified in code + production)

• The DH Data collection (20.500.12800/3) maps to the clariahSubmissions process (item-submission.xml:22). In clariahSubmissionsPageThree, dc.subject is not the tag input — it is <input-type autocomplete-custom="solr-subject_ac">autocomplete</input-type> (submission-forms.xml:1748; its hint at 1749-1757 promises separator splitting and documents hierarchical People::John Doe values). The Teaching materials collection (11234/5012 → teachingSubmissionsPageThree, :2018+) has the identical field.
• DsDynamicAutocompleteComponent (src/app/shared/form/builder/ds-dynamic-form-ui/models/autocomplete/ds-dynamic-autocomplete.component.ts) has no splitting at all:
  • onBlur (135-138) dispatches the entire raw input string as ONE value.
  • onSelectItem/updateModel (126-129 / 144-154) dispatch the picked suggestion verbatim — and Solr *_ac suggestions are built from stored values verbatim, so existing unsplit values breed new ones (feedback loop).
  • Bonus defect: ngOnInit (102-120) does not call super.ngOnInit(), so chips/hasAuthority stay undefined → pressing Enter/comma runs the inherited addTagsToChips into a silent TypeError and the typed value is lost.
• Users accustomed to clarin-dspace v5 (one box, ;-separated, server-side split via REPEATABLE_SPLIT_REGEX = ",|;") paste the whole list → one metadata value. The missing semicolon inside the example value ("Digital Humanities First Intermediate Period") confirms a hand-assembled paste.
• No server-side splitting exists in v7 (the only submission-path split(";") is DescribeStep.java:331 — local.sponsor complex handling).
• Secondary defects in the tag input (used by the traditional form's dc.subject, :284): onKeyUp (dynamic-tag.component.ts:186-193) handles only keyCodes 13/188 — ; is keyCode 186 and does not trigger the split despite the code comment and the form hint; the typeahead offers the raw text as a selectable entry (92-103) and onSelectItem (164-174) adds it as ONE chip without splitting; onBlur with an open popup discards the value.
• Other repeatable autocomplete fields have the same exposure: dc.publisher (:145/1633/1936), dc.contributor.author (:193), dc.language.iso (:254/269/1717/1732/2003), dc.relation* (specialFields).

Reproduction (local docker stack)

1. Either map a collection to clariahSubmissions (a fresh DB has no handle 20.500.12800/3 — create a collection and map its handle; note the docker compose uses a prebuilt image, so config changes need a mount/custom build), or zero-config: use dc.publisher / dc.language.iso on the default form — same component.
2. Paste kw1; kw2; kw3 into the field, click elsewhere (blur), save.
3. One dc.subject row with the literal value kw1; kw2; kw3.
4. Tag-input path (traditional dc.subject): paste a list → click the typeahead suggestion → one chip with the whole list; Enter with the popup open / blur can silently lose the value.

Suggested fix

Frontend (this repo):

1. DsDynamicAutocompleteComponent: split the input on ; (, — pending customer decision) in all three dispatch paths (blur, Enter, suggestion select — a suggestion itself can be an unsplit string from *_ac). Trim tokens, drop empties. Never split on : (hierarchical People::John Doe).
   Mandatory guard: DsDynamicSponsorAutocompleteComponent inherits onBlur, and its composed value legitimately contains ; (AUTOCOMPLETE_COMPLEX_PREFIX) — the split must be scoped (per-field config / whitelist / subclass); sponsor and complex-prefix values must be excluded, or local.sponsor gets corrupted.
2. Cleanup in the same PR: fix the inherited onKeyUp/undefined-chips TypeError (override or guard); add keyCode 186 to the tag component's onKeyUp; split in the tag onSelectItem path; drop the stale twobox comment (submission-forms.xml:1745).
3. Tests: pasting a; b; c + blur yields three values; Enter and ; work without TypeError; a suggestion containing ; is split; the sponsor composed value is NOT split; :: values are NOT split; non-repeatable fields unchanged.

Backend (companion issue dataquest-dev/DSpace#1373): server-side normalization in DescribeStep.doPatchProcessing (the original issue says "split on ingest"), QA detector, and the one-off curation of existing data (split preserving place order, copy text_lang, mandatory human review).