gitea_admin/zurich_kotak
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

form: make OCR autofill actually reach the fields

Browse Source 
Uploading an RC read the document correctly, said so, and filled
nothing.

The endpoint answers keyed by the ocr_config's extraction_fields[].key
— `reg_no`, `make_model`, `mfg_year`, `engine_cc`, `fuel` — because
that is what the vision prompt was asked to produce. Its own package
doc says so. The form's fields are `motor_reg_no_6`, `motor_cc_3`,
`motor_fuel_3`. The console wrote the response straight into values, so
every extracted value landed on a key no field renders: the read
succeeded, the "we read it for you" panel appeared, and the boxes
stayed empty.

This is the third place the same mismatch has bitten. Prefill hit it
(the pipeline names globals, the form uses suffixed activity keys) and
the validation errors hit it (the workflow reports the machine key).
Different surfaces, one cause: nothing in the stack translates between
a global field name and a form's per-activity id, so every consumer has
to do it and each one forgot.

The bridge was already on the field. ocr_config.field_mappings maps
extraction_key -> target_field (the global), and the form field is that
global plus a suffix. Verified against the live Collect Documents
schema: all five RC keys now resolve to real fields.

Also: which value wins. Seeding the form from the lead record (6a6801d)
made every mirrored field non-empty, so the old "only fill what is
blank" rule would have blocked OCR from writing anything at all. Values
copied off the lead are now tracked, and a document may overwrite them —
it is the better source for what it states. Anything typed by a person
is never overwritten, and stops being overwritable the moment it is
typed.
This commit is contained in:
Yashas 2026-09-08 11:06:41 +05:30
parent 6a6801d2bd
commit fc5eace791
1 changed files with 71 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
src/components/ActivityForm.jsx
View File

@ -1,4 +1,4 @@
import { useEffect, useState } from 'react'
import { useEffect, useRef, useState } from 'react'
import { useZino } from '../api/provider.jsx'
import { baseFieldId, fieldApplies } from '../api/config.js'
import { describeError, describeValidation } from '../api/errors.js'
@ -41,6 +41,10 @@ export default function ActivityForm({ activityUid, instanceId, lead, onDone, on
// `error`, because this is the form talking about itself rather than the
// server refusing something.
const [missing, setMissing] = useState([])
// Field ids filled from the lead record rather than typed. OCR may overwrite
// these — the document is more authoritative than a copy of the record — but
// must never overwrite something a person typed.
const seededRef = useRef(new Set())
const [busy, setBusy] = useState(false)
useEffect(() => {
@ -88,22 +92,35 @@ export default function ActivityForm({ activityUid, instanceId, lead, onDone, on
// ids are never seeded — a file reference is not a value to copy, and an
// id_gen is the platform's to issue. Only on an existing lead: an INIT
// form has no record behind it.
const seededFromLead = new Set()
if (lead && instanceId) {
for (const f of s.fields) {
if (seed[f.id] !== undefined) continue
if (['file', 'ocr', 'id_gen'].includes(f.data_type)) continue
const v = lead[baseFieldId(f.id ?? '')]
if (v !== undefined && v !== null && v !== '' && typeof v !== 'object') seed[f.id] = v
if (v !== undefined && v !== null && v !== '' && typeof v !== 'object') {
seed[f.id] = v
seededFromLead.add(f.id)
}
}
}
seededRef.current = seededFromLead
if (Object.keys(seed).length) setValues(seed)
})
.catch((e) => { if (!dead) setError(e) })
return () => { dead = true }
// `lead` is READ here but deliberately not a dependency. It is the parent's
// polled record object, so its identity changes on every refresh; listing it
// would re-run this effect, re-fetch the schema and call setValues(seed) —
// discarding whatever the operator had typed, every few seconds, mid-form.
// The seed is a one-time starting point, not a subscription.
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [client, activityUid, instanceId])
function set(id, v) {
// Typed by hand: from here on it outranks any document.
seededRef.current.delete(id)
setValues((p) => ({ ...p, [id]: v }))
// Clear the complaint as soon as the field is filled. Leaving a red field
// marked after it has been corrected teaches people to ignore the marking.
@ -127,6 +144,57 @@ export default function ActivityForm({ activityUid, instanceId, lead, onDone, on
const hidesMandatory = schema ? schema.fields.some((f) => f.mandatory && !fieldApplies(f.id, lead)) : false
const fields = applicable.length && !hidesMandatory ? applicable : (schema?.fields ?? [])
/**
* Turn an /ocr-extract response into form values.
*
* THE ENDPOINT DOES NOT SPEAK THE FORM'S LANGUAGE. It answers keyed by the
* ocr_config's `extraction_fields[].key` — `reg_no`, `engine_cc`, `fuel` —
* because that is what the vision prompt was asked to produce. The form's
* fields are `motor_reg_no_6`, `motor_cc_3`, `motor_fuel_3`. Written straight
* in, as they were, every extracted value landed on a key no field renders:
* the read succeeded, the panel said so, and nothing filled.
*
* The bridge is already on the field. `ocr_config.field_mappings` maps
* extraction_key -> target_field (the GLOBAL name), and the form field is that
* global plus a per-form suffix. So: key -> target -> the field whose base id
* matches. Same suffix rule as prefill and the same trap, in a third place.
*/
function applyExtraction(ocrField, extracted) {
const maps = ocrField?.properties?.ocr_config?.field_mappings
?? ocrField?.ocr_config?.field_mappings ?? []
const all = schema?.fields ?? []
// extraction_key -> form field id
const target = {}
for (const m of maps) {
if (!m?.extraction_key || !m?.target_field) continue
const hit = all.find((f) => f.id === m.target_field)
?? all.find((f) => baseFieldId(f.id) === m.target_field)
if (hit) target[m.extraction_key] = hit.id
}
setValues((prev) => {
const next = { ...prev }
for (const [key, val] of Object.entries(extracted)) {
if (val === null || val === undefined || String(val) === '') continue
// An unmapped key may still name a field directly on forms whose
// extraction keys ARE the field names.
const id = target[key]
?? all.find((f) => f.id === key)?.id
?? all.find((f) => baseFieldId(f.id) === key)?.id
if (!id) continue
// Never overwrite something a person typed. A value we copied off the
// lead is fair game — the document is the better source for it.
const isBlank = next[id] === undefined || next[id] === ''
if (isBlank || seededRef.current.has(id)) {
next[id] = val
seededRef.current.delete(id)
}
}
return next
})
}
function isEmpty(v) {
return v === undefined || v === null || v === '' || (Array.isArray(v) && v.length === 0)
}
@ -204,19 +272,7 @@ export default function ActivityForm({ activityUid, instanceId, lead, onDone, on
<FileField
field={f} value={v} instanceId={instanceId} activityUid={activityUid}
onChange={(refs) => set(f.id, refs)}
onExtract={(fields) => {
// The document answered questions the form was about to ask.
// Only fill what is still blank — never overwrite something a
// person typed with something a model read.
setValues((prev) => {
const next = { ...prev }
for (const [k, val] of Object.entries(fields)) {
if (val === null || val === undefined || String(val) === '') continue
if (next[k] === undefined || next[k] === '') next[k] = val
}
return next
})
}}
onExtract={(fields) => applyExtraction(f, fields)}
/>
) : f.data_type === 'id_gen' ? (
<div className="af__auto">