Skip to content

A full-featured CRM built with React, shadcn/ui, and Supabase.

Replace the search_index view with a full-text global_search RPC

François Zaninotto74c55db07298
diff --git a/doc/src/content/docs/users/global-search.mdx b/doc/src/content/docs/users/global-search.mdx
index 9fb56f07..5bca1b56 100644
--- a/doc/src/content/docs/users/global-search.mdx
+++ b/doc/src/content/docs/users/global-search.mdx
@@ -15,23 +15,26 @@ On mobile, tap the magnifier in the header. It is available on every screen.
## What is searched
-| Record | Searched fields |
-| ------------------------ | ------------------------------------------------------------------------ |
-| Companies | name, sector, description, website, phone, zip code, city, state |
-| Contacts | first and last name, job title, background, company name, emails, phones |
-| Deals | name, category, description, company name |
-| Tasks | text, type, contact name |
-| Notes (contact and deal) | note text, attachment file names, parent contact or deal name |
+| Record | Searched fields |
+| ------------------------ | ---------------------------------------------------------------- |
+| Companies | name, sector, description, website, phone, zip code, city, state |
+| Contacts | first and last name, job title, background, emails, phones |
+| Deals | name, category, description |
+| Tasks | text, type |
+| Notes (contact and deal) | note text, attachment file names |
Archived deals and completed tasks are left out, along with notes attached to an
archived deal.
## Searching with several words
-Every word must appear somewhere in the same record. Searching
-`sourdough Anderson` returns the notes about sourdough written on Thomas
-Anderson, and not the ones written on anybody else — the parent contact's name
-counts as part of the note.
+Every word must appear in the same record, as the start of a word: `tho and`
+finds Thomas Anderson, but `homas` finds nothing. Accents and case are ignored,
+so `helene` finds Hélène.
+
+Only the record's own fields count: searching `sourdough Anderson` does not
+find a note about sourdough written on Thomas Anderson, since his name belongs
+to the contact, not to the note.
## Good to know
diff --git a/src/components/atomic-crm/providers/commons/search.ts b/src/components/atomic-crm/providers/commons/search.ts
index 6033b388..3b839efe 100644
--- a/src/components/atomic-crm/providers/commons/search.ts
+++ b/src/components/atomic-crm/providers/commons/search.ts
@@ -55,17 +55,16 @@ const byDateDesc = (a: SearchResult, b: SearchResult) =>
(b.date ?? "").localeCompare(a.date ?? "");
/**
- * Emulates the `search_index` database view for the FakeRest provider, by
- * running FakeRest's own `q` search on each searchable resource and mapping
- * the matches to the view's shape.
+ * Emulates the `global_search` database function for the FakeRest provider,
+ * by running FakeRest's own `q` search on each searchable resource and mapping
+ * the matches to the function's result shape.
*
- * FIXME: this is an approximation of the view, not a reimplementation.
+ * FIXME: this is an approximation of the function, not a reimplementation.
* - one query per resource instead of one;
- * - FakeRest matches the whole term as a single substring, where Postgres ANDs
- * one `ilike` per word;
- * - the result SET differs, not only the field values: this path takes the top
- * 50 per resource and caps at 250, while Supabase takes one global top 50, so
- * the demo build can surface a record production would not return.
+ * - FakeRest matches the whole term as a single substring, where Postgres
+ * matches each word as a word prefix, ignoring accents;
+ * - FakeRest searches its own `q` fields, which differ from the fields
+ * indexed by `global_search`.
*/
export async function getSearchResults(
dataProvider: DataProvider,
diff --git a/src/components/atomic-crm/providers/fakerest/dataProvider.ts b/src/components/atomic-crm/providers/fakerest/dataProvider.ts
index 898b133f..3c355c28 100644
--- a/src/components/atomic-crm/providers/fakerest/dataProvider.ts
+++ b/src/components/atomic-crm/providers/fakerest/dataProvider.ts
@@ -17,6 +17,7 @@ import type {
Sale,
SalesFormData,
SearchResourceName,
+ SearchResult,
SignUpData,
Task,
} from "../../types";
@@ -141,11 +142,6 @@ const preserveAttachmentMimeType = <
})),
});
-const parseResourceFilter = (
- value: unknown,
-): SearchResourceName[] | undefined =>
- Array.isArray(value) ? (value as SearchResourceName[]) : undefined;
-
export const createDataProvider = ({
db = generateData(),
latency = 300,
@@ -177,17 +173,6 @@ export const createDataProvider = ({
const dataProviderWithCustomMethod: CrmDataProvider = {
...baseDataProvider,
async getList(resource: string, params: any) {
- if (resource === "search_index") {
- const { filter = {}, pagination } = params;
- const all = await getSearchResults(
- withSupabaseFilterAdapter(baseDataProvider),
- filter.q ?? "",
- parseResourceFilter(filter.resource_eq_any),
- );
- const { page, perPage } = pagination;
- const start = (page - 1) * perPage;
- return { data: all.slice(start, start + perPage), total: all.length };
- }
if (resource === "activity_log") {
const { filter = {}, pagination } = params;
const all = await getActivityLog(
@@ -312,6 +297,18 @@ export const createDataProvider = ({
return true;
},
+ globalSearch: async (
+ query: string,
+ resources: readonly SearchResourceName[],
+ maxResults: number,
+ ): Promise<SearchResult[]> => {
+ const results = await getSearchResults(
+ withSupabaseFilterAdapter(baseDataProvider),
+ query,
+ [...resources],
+ );
+ return results.slice(0, maxResults);
+ },
mergeContacts: async (sourceId: Identifier, targetId: Identifier) => {
return mergeContacts(sourceId, targetId, baseDataProvider);
},
diff --git a/src/components/atomic-crm/providers/supabase/dataProvider.ts b/src/components/atomic-crm/providers/supabase/dataProvider.ts
index 73f376f8..45d5bb23 100644
--- a/src/components/atomic-crm/providers/supabase/dataProvider.ts
+++ b/src/components/atomic-crm/providers/supabase/dataProvider.ts
@@ -13,6 +13,8 @@ import type {
RAFile,
Sale,
SalesFormData,
+ SearchResourceName,
+ SearchResult,
SignUpData,
} from "../../types";
import type { ConfigurationContextValue } from "../../root/ConfigurationContext";
@@ -209,6 +211,22 @@ const getDataProviderWithCustomMethods = () => {
async isInitialized() {
return getIsInitialized();
},
+ async globalSearch(
+ query: string,
+ resources: readonly SearchResourceName[],
+ maxResults: number,
+ ): Promise<SearchResult[]> {
+ const { data, error } = await getSupabaseClient().rpc("global_search", {
+ query,
+ resources,
+ max_results: maxResults,
+ });
+ if (error) {
+ console.error("global_search.error", error);
+ throw new Error("Failed to search");
+ }
+ return data ?? [];
+ },
async mergeContacts(sourceId: Identifier, targetId: Identifier) {
const { data, error } = await getSupabaseClient().functions.invoke(
"merge_contacts",
@@ -356,27 +374,6 @@ const lifeCycleCallbacks: ResourceCallbacks[] = [
return applyFullTextSearch(["name", "category", "description"])(params);
},
},
- {
- resource: "search_index",
- beforeGetList: async (params) => {
- if (!params.filter?.q) {
- return params;
- }
- const { q, ...filter } = params.filter;
- // A single `ilike` column: ra-data-postgrest splits the term on spaces
- // and emits one `ilike` per word. qs.stringify uses the indices format,
- // so the query string carries `content[0]=ilike.*a*&content[1]=ilike.*b*`;
- // PostgREST strips the `[n]` subscripts and ANDs the repeated column, so
- // "Thomas TF1" matches only rows whose content contains both words.
- return {
- ...params,
- filter: {
- ...filter,
- "content@ilike": q,
- },
- };
- },
- },
];
export const getDataProvider = () => {
diff --git a/src/components/atomic-crm/search/useGlobalSearch.ts b/src/components/atomic-crm/search/useGlobalSearch.ts
index c36024cb..1767fcbd 100644
--- a/src/components/atomic-crm/search/useGlobalSearch.ts
+++ b/src/components/atomic-crm/search/useGlobalSearch.ts
@@ -1,22 +1,14 @@
-import { useGetList } from "ra-core";
+import { useQuery } from "@tanstack/react-query";
+import { useDataProvider } from "ra-core";
import { useEffect, useState } from "react";
-import type { SearchResourceName, SearchResult } from "../types";
+import type { CrmDataProvider } from "../providers/types";
+import type { SearchResourceName } from "../types";
export const MIN_SEARCH_LENGTH = 2;
const DEBOUNCE_MS = 300;
const MAX_RESULTS = 50;
-const DISPLAYED_COLUMNS = [
- "id",
- "resource",
- "record_id",
- "title",
- "subtitle",
- "contact_id",
- "deal_id",
-] as const;
-
export const useDebouncedValue = <T>(value: T, delay: number): T => {
const [debouncedValue, setDebouncedValue] = useState(value);
@@ -29,7 +21,7 @@ export const useDebouncedValue = <T>(value: T, delay: number): T => {
};
/**
- * Query the `search_index` view for a debounced search term.
+ * Run the global search for a debounced search term.
*
* Stays idle until the term reaches MIN_SEARCH_LENGTH, so opening the dialog
* does not fire a request. `resources` restricts the query server-side, so the
@@ -39,22 +31,16 @@ export const useGlobalSearch = (
query: string,
resources: readonly SearchResourceName[],
) => {
+ const dataProvider = useDataProvider<CrmDataProvider>();
const debouncedQuery = useDebouncedValue(query.trim(), DEBOUNCE_MS);
const isEnabled = debouncedQuery.length >= MIN_SEARCH_LENGTH;
- const { data, isPending, error } = useGetList<SearchResult>(
- "search_index",
- {
- filter: {
- q: debouncedQuery,
- "resource@in": `(${resources.join(",")})`,
- },
- pagination: { page: 1, perPage: MAX_RESULTS },
- sort: { field: "date", order: "DESC" },
- meta: { columns: [...DISPLAYED_COLUMNS] },
- },
- { enabled: isEnabled },
- );
+ const { data, isPending, error } = useQuery({
+ queryKey: ["globalSearch", debouncedQuery, resources],
+ queryFn: () =>
+ dataProvider.globalSearch(debouncedQuery, resources, MAX_RESULTS),
+ enabled: isEnabled,
+ });
return {
results: isEnabled ? (data ?? []) : [],
diff --git a/supabase/migrations/20260910140459_global_search.sql b/supabase/migrations/20260910140459_global_search.sql
deleted file mode 100644
index e0e9e016..00000000
--- a/supabase/migrations/20260910140459_global_search.sql
+++ /dev/null
@@ -1,99 +0,0 @@
-alter table "public"."tasks" add column "created_at" timestamp with time zone not null default now();
-
--- Existing rows have no creation time to recover, so fall back to the best
--- available proxy: when the task was completed, a due date that has already
--- passed, or the contact's first_seen. `least(..., now())` keeps the result in
--- the past -- search_index sorts on this column, and a future date would put
--- every open task above every note, contact and deal. A bare `now()` would be
--- no better: it would tie every open task at the top of that same sort.
-update "public"."tasks" t
-set "created_at" = least(
- coalesce(
- t."done_date",
- case when t."due_date" < now() then t."due_date" end,
- c."first_seen"
- ),
- now()
-)
-from "public"."contacts" c
-where c."id" = t."contact_id";
-
-create or replace view "public"."search_index" with (security_invoker = on) as SELECT ('company.'::text || c.id) AS id,
- 'companies'::text AS resource,
- c.id AS record_id,
- c.name AS title,
- c.sector AS subtitle,
- concat_ws(' '::text, c.name, c.sector, c.description, c.website, c.phone_number, c.zipcode, c.city, c.state_abbr) AS content,
- NULL::bigint AS contact_id,
- NULL::bigint AS deal_id,
- c.created_at AS date
- FROM public.companies c
-UNION ALL
- SELECT ('contact.'::text || co.id) AS id,
- 'contacts'::text AS resource,
- co.id AS record_id,
- concat_ws(' '::text, co.first_name, co.last_name) AS title,
- cc.name AS subtitle,
- concat_ws(' '::text, co.first_name, co.last_name, co.title, co.background, cc.name, (jsonb_path_query_array(co.email_jsonb, '$[*]."email"'::jsonpath))::text, (jsonb_path_query_array(co.phone_jsonb, '$[*]."number"'::jsonpath))::text) AS content,
- co.id AS contact_id,
- NULL::bigint AS deal_id,
- co.first_seen AS date
- FROM (public.contacts co
- LEFT JOIN public.companies cc ON ((cc.id = co.company_id)))
-UNION ALL
- SELECT ('deal.'::text || d.id) AS id,
- 'deals'::text AS resource,
- d.id AS record_id,
- d.name AS title,
- dc.name AS subtitle,
- concat_ws(' '::text, d.name, d.category, d.description, dc.name) AS content,
- NULL::bigint AS contact_id,
- d.id AS deal_id,
- d.created_at AS date
- FROM (public.deals d
- LEFT JOIN public.companies dc ON ((dc.id = d.company_id)))
- WHERE (d.archived_at IS NULL)
-UNION ALL
- SELECT ('task.'::text || t.id) AS id,
- 'tasks'::text AS resource,
- t.id AS record_id,
- "left"(t.text, 200) AS title,
- concat_ws(' '::text, tco.first_name, tco.last_name) AS subtitle,
- concat_ws(' '::text, t.text, t.type, tco.first_name, tco.last_name) AS content,
- t.contact_id,
- NULL::bigint AS deal_id,
- t.created_at AS date
- FROM (public.tasks t
- LEFT JOIN public.contacts tco ON ((tco.id = t.contact_id)))
- WHERE (t.done_date IS NULL)
-UNION ALL
- SELECT ('contactNote.'::text || cn.id) AS id,
- 'contact_notes'::text AS resource,
- cn.id AS record_id,
- "left"(cn.text, 200) AS title,
- concat_ws(' '::text, nco.first_name, nco.last_name) AS subtitle,
- concat_ws(' '::text, cn.text, nco.first_name, nco.last_name, ( SELECT string_agg((a.a ->> 'title'::text), ' '::text) AS string_agg
- FROM unnest(cn.attachments) a(a))) AS content,
- cn.contact_id,
- NULL::bigint AS deal_id,
- cn.date
- FROM (public.contact_notes cn
- LEFT JOIN public.contacts nco ON ((nco.id = cn.contact_id)))
-UNION ALL
- SELECT ('dealNote.'::text || dn.id) AS id,
- 'deal_notes'::text AS resource,
- dn.id AS record_id,
- "left"(dn.text, 200) AS title,
- nd.name AS subtitle,
- concat_ws(' '::text, dn.text, nd.name, ( SELECT string_agg((a.a ->> 'title'::text), ' '::text) AS string_agg
- FROM unnest(dn.attachments) a(a))) AS content,
- NULL::bigint AS contact_id,
- dn.deal_id,
- dn.date
- FROM (public.deal_notes dn
- LEFT JOIN public.deals nd ON ((nd.id = dn.deal_id)))
- WHERE (nd.archived_at IS NULL);
-
-grant all on table public.search_index to anon;
-grant all on table public.search_index to authenticated;
-grant all on table public.search_index to service_role;
diff --git a/supabase/migrations/20261010054346_global_search.sql b/supabase/migrations/20261010054346_global_search.sql
new file mode 100644
index 00000000..a9dbb62b
--- /dev/null
+++ b/supabase/migrations/20261010054346_global_search.sql
@@ -0,0 +1,205 @@
+alter table "public"."tasks" add column "created_at" timestamp with time zone not null default now();
+
+-- Existing rows have no creation time to recover, so fall back to the best
+-- available proxy: when the task was completed, a due date that has already
+-- passed, or the contact's first_seen. `least(..., now())` keeps the result in
+-- the past -- global_search sorts on this column, and a future date would put
+-- every open task above every note, contact and deal. A bare `now()` would be
+-- no better: it would tie every open task at the top of that same sort.
+update "public"."tasks" t
+set "created_at" = least(
+ coalesce(
+ t."done_date",
+ case when t."due_date" < now() then t."due_date" end,
+ c."first_seen"
+ ),
+ now()
+)
+from "public"."contacts" c
+where c."id" = t."contact_id";
+
+create extension if not exists "unaccent" with schema "extensions";
+
+-- unaccent() is only STABLE (its dictionary could change), which generated
+-- columns reject. The dictionary is fixed here, so the wrapper is immutable.
+create or replace function public.search_unaccent(value text) returns text
+ language sql immutable strict parallel safe
+ set search_path to ''
+ as $$ select extensions.unaccent('extensions.unaccent'::regdictionary, value) $$;
+
+-- Documents and queries are normalized the same way: unaccented, lowercased,
+-- split on anything that is not a letter or a digit. So "hélè[email protected]"
+-- indexes as helene, dupont, acme, fr and any of those prefixes finds it.
+create or replace function public.search_document(variadic parts text[]) returns tsvector
+ language sql immutable parallel safe
+ set search_path to ''
+ as $$
+ select to_tsvector(
+ 'simple'::regconfig,
+ regexp_replace(public.search_unaccent(array_to_string(parts, ' ')), '[^[:alnum:]]+', ' ', 'g')
+ )
+ $$;
+
+-- Every word of the query must match, each one as a prefix: "tho dup" finds
+-- "Thomas Dupont". Returns null (matching nothing) for a blank query.
+create or replace function public.search_query(query text) returns tsquery
+ language sql immutable strict parallel safe
+ set search_path to ''
+ as $$
+ select to_tsquery('simple'::regconfig, string_agg(word || ':*', ' & '))
+ from regexp_split_to_table(lower(public.search_unaccent(query)), '[^[:alnum:]]+') as word
+ where word <> ''
+ $$;
+
+create or replace function public.search_attachment_titles(attachments jsonb[]) returns text
+ language sql immutable parallel safe
+ set search_path to ''
+ as $$ select string_agg(attachment ->> 'title', ' ') from unnest(attachments) as attachment $$;
+
+alter table public.companies add column fts tsvector generated always as (
+ public.search_document(name, sector, description, website::text, phone_number, zipcode, city, state_abbr)
+) stored;
+
+alter table public.contacts add column fts tsvector generated always as (
+ public.search_document(
+ first_name,
+ last_name,
+ title,
+ background,
+ jsonb_path_query_array(email_jsonb, '$[*]."email"')::text,
+ jsonb_path_query_array(phone_jsonb, '$[*]."number"')::text
+ )
+) stored;
+
+alter table public.deals add column fts tsvector generated always as (
+ public.search_document(name, category, description)
+) stored;
+
+alter table public.tasks add column fts tsvector generated always as (
+ public.search_document(text, type)
+) stored;
+
+alter table public.contact_notes add column fts tsvector generated always as (
+ public.search_document(text, public.search_attachment_titles(attachments))
+) stored;
+
+alter table public.deal_notes add column fts tsvector generated always as (
+ public.search_document(text, public.search_attachment_titles(attachments))
+) stored;
+
+-- Partial indexes mirror what global_search() returns: open tasks, live deals.
+create index companies_fts_idx on public.companies using gin (fts);
+create index contacts_fts_idx on public.contacts using gin (fts);
+create index deals_fts_idx on public.deals using gin (fts) where archived_at is null;
+create index tasks_fts_idx on public.tasks using gin (fts) where done_date is null;
+create index contact_notes_fts_idx on public.contact_notes using gin (fts);
+create index deal_notes_fts_idx on public.deal_notes using gin (fts);
+
+-- A word found in most rows (a common word, a 2-letter prefix) makes the GIN
+-- path fetch and sort every match. Walking a date index backwards instead stops
+-- at the first max_results matches. The planner picks between the two per
+-- query, which needs fine-grained lexeme statistics on the large tables.
+create index contacts_first_seen_idx on public.contacts using btree (first_seen desc nulls last);
+create index tasks_open_created_at_idx on public.tasks using btree (created_at desc) where done_date is null;
+create index contact_notes_date_idx on public.contact_notes using btree (date desc nulls last);
+create index deal_notes_date_idx on public.deal_notes using btree (date desc nulls last);
+
+alter table public.contacts alter column fts set statistics 10000;
+alter table public.tasks alter column fts set statistics 10000;
+alter table public.contact_notes alter column fts set statistics 10000;
+alter table public.deal_notes alter column fts set statistics 10000;
+
+-- Each branch takes its own newest max_results matches, so the union never
+-- holds more than 6 * max_results rows, and the lookups that build display
+-- labels only run on those. The SQL is built per call with the tsquery inlined
+-- as a literal: with a parameter the planner could not tell a rare word (use
+-- the GIN index) from a common one (walk the date index).
+create or replace function public.global_search(
+ query text,
+ resources text[] default null,
+ max_results integer default 50
+) returns table (
+ id text,
+ resource text,
+ record_id bigint,
+ title text,
+ subtitle text,
+ contact_id bigint,
+ deal_id bigint,
+ date timestamp with time zone
+)
+ language plpgsql stable security invoker
+ set search_path to ''
+ as $$
+declare
+ tsq tsquery := public.search_query(query);
+ branch_templates constant jsonb := jsonb_build_object(
+ 'companies', $q$
+ select 'company.' || c.id, 'companies', c.id, c.name, c.sector, null::bigint, null::bigint, c.created_at
+ from public.companies c
+ where c.fts @@ %1$L::tsquery
+ order by c.created_at desc
+ limit %2$s $q$,
+ 'contacts', $q$
+ select 'contact.' || co.id, 'contacts', co.id, concat_ws(' ', co.first_name, co.last_name),
+ (select cc.name from public.companies cc where cc.id = co.company_id), co.id, null::bigint, co.first_seen
+ from public.contacts co
+ where co.fts @@ %1$L::tsquery
+ order by co.first_seen desc nulls last
+ limit %2$s $q$,
+ 'deals', $q$
+ select 'deal.' || d.id, 'deals', d.id, d.name,
+ (select dc.name from public.companies dc where dc.id = d.company_id), null::bigint, d.id, d.created_at
+ from public.deals d
+ where d.archived_at is null and d.fts @@ %1$L::tsquery
+ order by d.created_at desc
+ limit %2$s $q$,
+ 'tasks', $q$
+ select 'task.' || t.id, 'tasks', t.id, left(t.text, 200),
+ (select concat_ws(' ', tc.first_name, tc.last_name) from public.contacts tc where tc.id = t.contact_id),
+ t.contact_id, null::bigint, t.created_at
+ from public.tasks t
+ where t.done_date is null and t.fts @@ %1$L::tsquery
+ order by t.created_at desc
+ limit %2$s $q$,
+ 'contact_notes', $q$
+ select 'contactNote.' || cn.id, 'contact_notes', cn.id, left(cn.text, 200),
+ (select concat_ws(' ', nc.first_name, nc.last_name) from public.contacts nc where nc.id = cn.contact_id),
+ cn.contact_id, null::bigint, cn.date
+ from public.contact_notes cn
+ where cn.fts @@ %1$L::tsquery
+ order by cn.date desc nulls last
+ limit %2$s $q$,
+ 'deal_notes', $q$
+ select 'dealNote.' || dn.id, 'deal_notes', dn.id, left(dn.text, 200), nd.name, null::bigint, dn.deal_id, dn.date
+ from public.deal_notes dn
+ join public.deals nd on nd.id = dn.deal_id
+ where nd.archived_at is null and dn.fts @@ %1$L::tsquery
+ order by dn.date desc nulls last
+ limit %2$s $q$
+ );
+ branches text[];
+begin
+ if tsq is null or max_results < 1 then
+ return;
+ end if;
+
+ select array_agg(format('(%s)', format(template, tsq::text, max_results)))
+ into branches
+ from jsonb_each_text(branch_templates) as b(name, template)
+ where resources is null or b.name = any(resources);
+
+ if branches is null then
+ return;
+ end if;
+
+ return query execute format(
+ 'select * from (%s) results order by 8 desc nulls last limit %s',
+ array_to_string(branches, ' union all '),
+ max_results
+ );
+end
+$$;
+
+revoke execute on function public.global_search(text, text[], integer) from public, anon;
+grant execute on function public.global_search(text, text[], integer) to authenticated, service_role;
diff --git a/supabase/schemas/03_views.sql b/supabase/schemas/03_views.sql
index a43b80e9..1b2ec74f 100644
--- a/supabase/schemas/03_views.sql
+++ b/supabase/schemas/03_views.sql
@@ -132,114 +132,3 @@ select count(sub.id) as is_initialized
from (
select sales.id from public.sales limit 1
) sub;
-
-create or replace view public.search_index with (security_invoker = on) as
-select
- ('company.' || c.id) as id,
- 'companies' as resource,
- c.id as record_id,
- c.name as title,
- c.sector as subtitle,
- concat_ws(
- ' ',
- c.name,
- c.sector,
- c.description,
- c.website,
- c.phone_number,
- c.zipcode,
- c.city,
- c.state_abbr
- ) as content,
- null::bigint as contact_id,
- null::bigint as deal_id,
- c.created_at as date
-from public.companies c
-union all
-select
- ('contact.' || co.id) as id,
- 'contacts' as resource,
- co.id as record_id,
- concat_ws(' ', co.first_name, co.last_name) as title,
- cc.name as subtitle,
- concat_ws(
- ' ',
- co.first_name,
- co.last_name,
- co.title,
- co.background,
- cc.name,
- (jsonb_path_query_array(co.email_jsonb, '$[*]."email"'))::text,
- (jsonb_path_query_array(co.phone_jsonb, '$[*]."number"'))::text
- ) as content,
- co.id as contact_id,
- null::bigint as deal_id,
- co.first_seen as date
-from public.contacts co
- left join public.companies cc on cc.id = co.company_id
-union all
-select
- ('deal.' || d.id) as id,
- 'deals' as resource,
- d.id as record_id,
- d.name as title,
- dc.name as subtitle,
- concat_ws(' ', d.name, d.category, d.description, dc.name) as content,
- null::bigint as contact_id,
- d.id as deal_id,
- d.created_at as date
-from public.deals d
- left join public.companies dc on dc.id = d.company_id
-where d.archived_at is null
-union all
-select
- ('task.' || t.id) as id,
- 'tasks' as resource,
- t.id as record_id,
- left(t.text, 200) as title,
- concat_ws(' ', tco.first_name, tco.last_name) as subtitle,
- concat_ws(' ', t.text, t.type, tco.first_name, tco.last_name) as content,
- t.contact_id,
- null::bigint as deal_id,
- t.created_at as date
-from public.tasks t
- left join public.contacts tco on tco.id = t.contact_id
-where t.done_date is null
-union all
-select
- ('contactNote.' || cn.id) as id,
- 'contact_notes' as resource,
- cn.id as record_id,
- left(cn.text, 200) as title,
- concat_ws(' ', nco.first_name, nco.last_name) as subtitle,
- concat_ws(
- ' ',
- cn.text,
- nco.first_name,
- nco.last_name,
- (select string_agg(a ->> 'title', ' ') from unnest(cn.attachments) as a)
- ) as content,
- cn.contact_id,
- null::bigint as deal_id,
- cn.date
-from public.contact_notes cn
- left join public.contacts nco on nco.id = cn.contact_id
-union all
-select
- ('dealNote.' || dn.id) as id,
- 'deal_notes' as resource,
- dn.id as record_id,
- left(dn.text, 200) as title,
- nd.name as subtitle,
- concat_ws(
- ' ',
- dn.text,
- nd.name,
- (select string_agg(a ->> 'title', ' ') from unnest(dn.attachments) as a)
- ) as content,
- null::bigint as contact_id,
- dn.deal_id,
- dn.date
-from public.deal_notes dn
- left join public.deals nd on nd.id = dn.deal_id
-where nd.archived_at is null;
diff --git a/supabase/schemas/06_grants.sql b/supabase/schemas/06_grants.sql
index 7d2f7aac..865c9dbc 100644
--- a/supabase/schemas/06_grants.sql
+++ b/supabase/schemas/06_grants.sql
@@ -123,10 +123,6 @@ grant all on table public.init_state to anon;
grant all on table public.init_state to authenticated;
grant all on table public.init_state to service_role;
-grant all on table public.search_index to anon;
-grant all on table public.search_index to authenticated;
-grant all on table public.search_index to service_role;
-
-- Sequence grants
grant all on sequence public.companies_id_seq to anon;
grant all on sequence public.companies_id_seq to authenticated;
diff --git a/supabase/schemas/08_search.sql b/supabase/schemas/08_search.sql
new file mode 100644
index 00000000..56228c64
--- /dev/null
+++ b/supabase/schemas/08_search.sql
@@ -0,0 +1,192 @@
+--
+-- Global search
+-- Each searchable table carries a generated tsvector column with a GIN index,
+-- and global_search() queries them through RLS (security invoker), so search
+-- visibility always follows the table policies.
+--
+
+create extension if not exists "unaccent" with schema "extensions";
+
+-- unaccent() is only STABLE (its dictionary could change), which generated
+-- columns reject. The dictionary is fixed here, so the wrapper is immutable.
+create or replace function public.search_unaccent(value text) returns text
+ language sql immutable strict parallel safe
+ set search_path to ''
+ as $$ select extensions.unaccent('extensions.unaccent'::regdictionary, value) $$;
+
+-- Documents and queries are normalized the same way: unaccented, lowercased,
+-- split on anything that is not a letter or a digit. So "hélè[email protected]"
+-- indexes as helene, dupont, acme, fr and any of those prefixes finds it.
+create or replace function public.search_document(variadic parts text[]) returns tsvector
+ language sql immutable parallel safe
+ set search_path to ''
+ as $$
+ select to_tsvector(
+ 'simple'::regconfig,
+ regexp_replace(public.search_unaccent(array_to_string(parts, ' ')), '[^[:alnum:]]+', ' ', 'g')
+ )
+ $$;
+
+-- Every word of the query must match, each one as a prefix: "tho dup" finds
+-- "Thomas Dupont". Returns null (matching nothing) for a blank query.
+create or replace function public.search_query(query text) returns tsquery
+ language sql immutable strict parallel safe
+ set search_path to ''
+ as $$
+ select to_tsquery('simple'::regconfig, string_agg(word || ':*', ' & '))
+ from regexp_split_to_table(lower(public.search_unaccent(query)), '[^[:alnum:]]+') as word
+ where word <> ''
+ $$;
+
+create or replace function public.search_attachment_titles(attachments jsonb[]) returns text
+ language sql immutable parallel safe
+ set search_path to ''
+ as $$ select string_agg(attachment ->> 'title', ' ') from unnest(attachments) as attachment $$;
+
+alter table public.companies add column fts tsvector generated always as (
+ public.search_document(name, sector, description, website::text, phone_number, zipcode, city, state_abbr)
+) stored;
+
+alter table public.contacts add column fts tsvector generated always as (
+ public.search_document(
+ first_name,
+ last_name,
+ title,
+ background,
+ jsonb_path_query_array(email_jsonb, '$[*]."email"')::text,
+ jsonb_path_query_array(phone_jsonb, '$[*]."number"')::text
+ )
+) stored;
+
+alter table public.deals add column fts tsvector generated always as (
+ public.search_document(name, category, description)
+) stored;
+
+alter table public.tasks add column fts tsvector generated always as (
+ public.search_document(text, type)
+) stored;
+
+alter table public.contact_notes add column fts tsvector generated always as (
+ public.search_document(text, public.search_attachment_titles(attachments))
+) stored;
+
+alter table public.deal_notes add column fts tsvector generated always as (
+ public.search_document(text, public.search_attachment_titles(attachments))
+) stored;
+
+-- Partial indexes mirror what global_search() returns: open tasks, live deals.
+create index companies_fts_idx on public.companies using gin (fts);
+create index contacts_fts_idx on public.contacts using gin (fts);
+create index deals_fts_idx on public.deals using gin (fts) where archived_at is null;
+create index tasks_fts_idx on public.tasks using gin (fts) where done_date is null;
+create index contact_notes_fts_idx on public.contact_notes using gin (fts);
+create index deal_notes_fts_idx on public.deal_notes using gin (fts);
+
+-- A word found in most rows (a common word, a 2-letter prefix) makes the GIN
+-- path fetch and sort every match. Walking a date index backwards instead stops
+-- at the first max_results matches. The planner picks between the two per
+-- query, which needs fine-grained lexeme statistics on the large tables.
+create index contacts_first_seen_idx on public.contacts using btree (first_seen desc nulls last);
+create index tasks_open_created_at_idx on public.tasks using btree (created_at desc) where done_date is null;
+create index contact_notes_date_idx on public.contact_notes using btree (date desc nulls last);
+create index deal_notes_date_idx on public.deal_notes using btree (date desc nulls last);
+
+alter table public.contacts alter column fts set statistics 10000;
+alter table public.tasks alter column fts set statistics 10000;
+alter table public.contact_notes alter column fts set statistics 10000;
+alter table public.deal_notes alter column fts set statistics 10000;
+
+-- Each branch takes its own newest max_results matches, so the union never
+-- holds more than 6 * max_results rows, and the lookups that build display
+-- labels only run on those. The SQL is built per call with the tsquery inlined
+-- as a literal: with a parameter the planner could not tell a rare word (use
+-- the GIN index) from a common one (walk the date index).
+create or replace function public.global_search(
+ query text,
+ resources text[] default null,
+ max_results integer default 50
+) returns table (
+ id text,
+ resource text,
+ record_id bigint,
+ title text,
+ subtitle text,
+ contact_id bigint,
+ deal_id bigint,
+ date timestamp with time zone
+)
+ language plpgsql stable security invoker
+ set search_path to ''
+ as $$
+declare
+ tsq tsquery := public.search_query(query);
+ branch_templates constant jsonb := jsonb_build_object(
+ 'companies', $q$
+ select 'company.' || c.id, 'companies', c.id, c.name, c.sector, null::bigint, null::bigint, c.created_at
+ from public.companies c
+ where c.fts @@ %1$L::tsquery
+ order by c.created_at desc
+ limit %2$s $q$,
+ 'contacts', $q$
+ select 'contact.' || co.id, 'contacts', co.id, concat_ws(' ', co.first_name, co.last_name),
+ (select cc.name from public.companies cc where cc.id = co.company_id), co.id, null::bigint, co.first_seen
+ from public.contacts co
+ where co.fts @@ %1$L::tsquery
+ order by co.first_seen desc nulls last
+ limit %2$s $q$,
+ 'deals', $q$
+ select 'deal.' || d.id, 'deals', d.id, d.name,
+ (select dc.name from public.companies dc where dc.id = d.company_id), null::bigint, d.id, d.created_at
+ from public.deals d
+ where d.archived_at is null and d.fts @@ %1$L::tsquery
+ order by d.created_at desc
+ limit %2$s $q$,
+ 'tasks', $q$
+ select 'task.' || t.id, 'tasks', t.id, left(t.text, 200),
+ (select concat_ws(' ', tc.first_name, tc.last_name) from public.contacts tc where tc.id = t.contact_id),
+ t.contact_id, null::bigint, t.created_at
+ from public.tasks t
+ where t.done_date is null and t.fts @@ %1$L::tsquery
+ order by t.created_at desc
+ limit %2$s $q$,
+ 'contact_notes', $q$
+ select 'contactNote.' || cn.id, 'contact_notes', cn.id, left(cn.text, 200),
+ (select concat_ws(' ', nc.first_name, nc.last_name) from public.contacts nc where nc.id = cn.contact_id),
+ cn.contact_id, null::bigint, cn.date
+ from public.contact_notes cn
+ where cn.fts @@ %1$L::tsquery
+ order by cn.date desc nulls last
+ limit %2$s $q$,
+ 'deal_notes', $q$
+ select 'dealNote.' || dn.id, 'deal_notes', dn.id, left(dn.text, 200), nd.name, null::bigint, dn.deal_id, dn.date
+ from public.deal_notes dn
+ join public.deals nd on nd.id = dn.deal_id
+ where nd.archived_at is null and dn.fts @@ %1$L::tsquery
+ order by dn.date desc nulls last
+ limit %2$s $q$
+ );
+ branches text[];
+begin
+ if tsq is null or max_results < 1 then
+ return;
+ end if;
+
+ select array_agg(format('(%s)', format(template, tsq::text, max_results)))
+ into branches
+ from jsonb_each_text(branch_templates) as b(name, template)
+ where resources is null or b.name = any(resources);
+
+ if branches is null then
+ return;
+ end if;
+
+ return query execute format(
+ 'select * from (%s) results order by 8 desc nulls last limit %s',
+ array_to_string(branches, ' union all '),
+ max_results
+ );
+end
+$$;
+
+revoke execute on function public.global_search(text, text[], integer) from public, anon;
+grant execute on function public.global_search(text, text[], integer) to authenticated, service_role;