diff --git a/swh/provenance/postgresql/provenance.py b/swh/provenance/postgresql/provenance.py index f559d83..623e7b3 100644 --- a/swh/provenance/postgresql/provenance.py +++ b/swh/provenance/postgresql/provenance.py @@ -1,297 +1,292 @@ # Copyright (C) 2021 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information from datetime import datetime import itertools import logging from typing import Dict, Generator, Iterable, List, Optional, Set import psycopg2.extensions import psycopg2.extras from typing_extensions import Literal from swh.core.db import BaseDb from swh.model.model import Sha1Git from ..interface import ( EntityType, ProvenanceResult, RelationData, RelationType, RevisionData, ) LOGGER = logging.getLogger(__name__) class ProvenanceStoragePostgreSql: def __init__( self, conn: psycopg2.extensions.connection, raise_on_commit: bool = False ) -> None: BaseDb.adapt_conn(conn) conn.set_isolation_level(psycopg2.extensions.ISOLATION_LEVEL_AUTOCOMMIT) conn.set_session(autocommit=True) self.conn = conn self.cursor = self.conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) # XXX: not sure this is the best place to do it! sql = "SET timezone TO 'UTC'" self.cursor.execute(sql) self._flavor: Optional[str] = None self.raise_on_commit = raise_on_commit @property def flavor(self) -> str: if self._flavor is None: sql = "SELECT swh_get_dbflavor() AS flavor" self.cursor.execute(sql) self._flavor = self.cursor.fetchone()["flavor"] assert self._flavor is not None return self._flavor @property def denormalized(self) -> bool: return "denormalized" in self.flavor def content_find_first(self, id: Sha1Git) -> Optional[ProvenanceResult]: sql = "SELECT * FROM swh_provenance_content_find_first(%s)" self.cursor.execute(sql, (id,)) row = self.cursor.fetchone() return ProvenanceResult(**row) if row is not None else None def content_find_all( self, id: Sha1Git, limit: Optional[int] = None ) -> Generator[ProvenanceResult, None, None]: sql = "SELECT * FROM swh_provenance_content_find_all(%s, %s)" self.cursor.execute(sql, (id, limit)) yield from (ProvenanceResult(**row) for row in self.cursor.fetchall()) def content_set_date(self, dates: Dict[Sha1Git, datetime]) -> bool: return self._entity_set_date("content", dates) def content_get(self, ids: Iterable[Sha1Git]) -> Dict[Sha1Git, datetime]: return self._entity_get_date("content", ids) def directory_set_date(self, dates: Dict[Sha1Git, datetime]) -> bool: return self._entity_set_date("directory", dates) def directory_get(self, ids: Iterable[Sha1Git]) -> Dict[Sha1Git, datetime]: return self._entity_get_date("directory", ids) def entity_get_all(self, entity: EntityType) -> Set[Sha1Git]: sql = f"SELECT sha1 FROM {entity.value}" self.cursor.execute(sql) return {row["sha1"] for row in self.cursor.fetchall()} def location_get(self) -> Set[bytes]: sql = "SELECT location.path AS path FROM location" self.cursor.execute(sql) return {row["path"] for row in self.cursor.fetchall()} def origin_set_url(self, urls: Dict[Sha1Git, str]) -> bool: try: if urls: sql = """ - LOCK TABLE ONLY origin; INSERT INTO origin(sha1, url) VALUES %s ON CONFLICT DO NOTHING """ psycopg2.extras.execute_values(self.cursor, sql, urls.items()) return True except: # noqa: E722 # Unexpected error occurred, rollback all changes and log message LOGGER.exception("Unexpected error") if self.raise_on_commit: raise return False def origin_get(self, ids: Iterable[Sha1Git]) -> Dict[Sha1Git, str]: urls: Dict[Sha1Git, str] = {} sha1s = tuple(ids) if sha1s: # TODO: consider splitting this query in several ones if sha1s is too big! values = ", ".join(itertools.repeat("%s", len(sha1s))) sql = f""" SELECT sha1, url FROM origin WHERE sha1 IN ({values}) """ self.cursor.execute(sql, sha1s) urls.update((row["sha1"], row["url"]) for row in self.cursor.fetchall()) return urls def revision_set_date(self, dates: Dict[Sha1Git, datetime]) -> bool: return self._entity_set_date("revision", dates) def revision_set_origin(self, origins: Dict[Sha1Git, Sha1Git]) -> bool: try: if origins: sql = """ - LOCK TABLE ONLY revision; INSERT INTO revision(sha1, origin) (SELECT V.rev AS sha1, O.id AS origin FROM (VALUES %s) AS V(rev, org) JOIN origin AS O ON (O.sha1=V.org)) ON CONFLICT (sha1) DO UPDATE SET origin=EXCLUDED.origin """ psycopg2.extras.execute_values(self.cursor, sql, origins.items()) return True except: # noqa: E722 # Unexpected error occurred, rollback all changes and log message LOGGER.exception("Unexpected error") if self.raise_on_commit: raise return False def revision_get(self, ids: Iterable[Sha1Git]) -> Dict[Sha1Git, RevisionData]: result: Dict[Sha1Git, RevisionData] = {} sha1s = tuple(ids) if sha1s: # TODO: consider splitting this query in several ones if sha1s is too big! values = ", ".join(itertools.repeat("%s", len(sha1s))) sql = f""" SELECT R.sha1, R.date, O.sha1 AS origin FROM revision AS R LEFT JOIN origin AS O ON (O.id=R.origin) WHERE R.sha1 IN ({values}) """ self.cursor.execute(sql, sha1s) result.update( (row["sha1"], RevisionData(date=row["date"], origin=row["origin"])) for row in self.cursor.fetchall() ) return result def relation_add( self, relation: RelationType, data: Iterable[RelationData] ) -> bool: try: rows = [(rel.src, rel.dst, rel.path) for rel in data] if rows: rel_table = relation.value src_table, *_, dst_table = rel_table.split("_") if src_table != "origin": # Origin entries should be inserted previously as they require extra # non-null information srcs = tuple(set((sha1,) for (sha1, _, _) in rows)) sql = f""" - LOCK TABLE ONLY {src_table}; INSERT INTO {src_table}(sha1) VALUES %s ON CONFLICT DO NOTHING """ psycopg2.extras.execute_values(self.cursor, sql, srcs) if dst_table != "origin": # Origin entries should be inserted previously as they require extra # non-null information dsts = tuple(set((sha1,) for (_, sha1, _) in rows)) sql = f""" - LOCK TABLE ONLY {dst_table}; INSERT INTO {dst_table}(sha1) VALUES %s ON CONFLICT DO NOTHING """ psycopg2.extras.execute_values(self.cursor, sql, dsts) # Put the next three queries in a manual single transaction: # they use the same temp table with self.conn: with self.conn.cursor() as cur: cur.execute("SELECT swh_mktemp_relation_add()") psycopg2.extras.execute_values( cur, sql=( "INSERT INTO tmp_relation_add (src, dst, path) " "VALUES %s" ), argslist=rows, ) sql = "SELECT swh_provenance_relation_add_from_temp(%s, %s, %s)" cur.execute(sql, (rel_table, src_table, dst_table)) return True except: # noqa: E722 # Unexpected error occurred, rollback all changes and log message LOGGER.exception("Unexpected error") if self.raise_on_commit: raise return False def relation_get( self, relation: RelationType, ids: Iterable[Sha1Git], reverse: bool = False ) -> Set[RelationData]: return self._relation_get(relation, ids, reverse) def relation_get_all(self, relation: RelationType) -> Set[RelationData]: return self._relation_get(relation, None) def _entity_get_date( self, entity: Literal["content", "directory", "revision"], ids: Iterable[Sha1Git], ) -> Dict[Sha1Git, datetime]: dates: Dict[Sha1Git, datetime] = {} sha1s = tuple(ids) if sha1s: # TODO: consider splitting this query in several ones if sha1s is too big! values = ", ".join(itertools.repeat("%s", len(sha1s))) sql = f""" SELECT sha1, date FROM {entity} WHERE sha1 IN ({values}) AND date IS NOT NULL """ self.cursor.execute(sql, sha1s) dates.update((row["sha1"], row["date"]) for row in self.cursor.fetchall()) return dates def _entity_set_date( self, entity: Literal["content", "directory", "revision"], data: Dict[Sha1Git, datetime], ) -> bool: try: if data: sql = f""" - LOCK TABLE ONLY {entity}; INSERT INTO {entity}(sha1, date) VALUES %s ON CONFLICT (sha1) DO UPDATE SET date=LEAST(EXCLUDED.date,{entity}.date) """ psycopg2.extras.execute_values(self.cursor, sql, data.items()) return True except: # noqa: E722 # Unexpected error occurred, rollback all changes and log message LOGGER.exception("Unexpected error") if self.raise_on_commit: raise return False def _relation_get( self, relation: RelationType, ids: Optional[Iterable[Sha1Git]], reverse: bool = False, ) -> Set[RelationData]: result: Set[RelationData] = set() sha1s: List[Sha1Git] if ids is not None: sha1s = list(ids) filter = "filter-src" if not reverse else "filter-dst" else: sha1s = [] filter = "no-filter" if filter == "no-filter" or sha1s: rel_table = relation.value src_table, *_, dst_table = rel_table.split("_") sql = "SELECT * FROM swh_provenance_relation_get(%s, %s, %s, %s, %s)" self.cursor.execute(sql, (rel_table, src_table, dst_table, filter, sha1s)) result.update(RelationData(**row) for row in self.cursor.fetchall()) return result def with_path(self) -> bool: return "with-path" in self.flavor diff --git a/swh/provenance/sql/40-funcs.sql b/swh/provenance/sql/40-funcs.sql index 435b074..d6dffa4 100644 --- a/swh/provenance/sql/40-funcs.sql +++ b/swh/provenance/sql/40-funcs.sql @@ -1,717 +1,711 @@ -- psql variables to get the current database flavor select position('denormalized' in swh_get_dbflavor()::text) = 0 as dbflavor_norm \gset select position('with-path' in swh_get_dbflavor()::text) != 0 as dbflavor_with_path \gset create or replace function swh_mktemp_relation_add() returns void language sql as $$ create temp table tmp_relation_add ( src sha1_git not null, dst sha1_git not null, path unix_path ) on commit drop $$; \if :dbflavor_norm \if :dbflavor_with_path -- -- with path and normalized -- create or replace function swh_provenance_content_find_first(content_id sha1_git) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ select C.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, L.path as path from content as C inner join content_in_revision as CR on (CR.content = C.id) inner join location as L on (L.id = CR.location) inner join revision as R on (R.id = CR.revision) left join origin as O on (O.id = R.origin) where C.sha1 = content_id order by date, revision, origin, path asc limit 1 $$; create or replace function swh_provenance_content_find_all(content_id sha1_git, early_cut int) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ (select C.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, L.path as path from content as C inner join content_in_revision as CR on (CR.content = C.id) inner join location as L on (L.id = CR.location) inner join revision as R on (R.id = CR.revision) left join origin as O on (O.id = R.origin) where C.sha1 = content_id) union (select C.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, case DL.path when '' then CL.path when '.' then CL.path else (DL.path || '/' || CL.path)::unix_path end as path from content as C inner join content_in_directory as CD on (CD.content = C.id) inner join directory_in_revision as DR on (DR.directory = CD.directory) inner join revision as R on (R.id = DR.revision) inner join location as CL on (CL.id = CD.location) inner join location as DL on (DL.id = DR.location) left join origin as O on (O.id = R.origin) where C.sha1 = content_id) order by date, revision, origin, path limit early_cut $$; create or replace function swh_provenance_relation_add_from_temp( rel_table regclass, src_table regclass, dst_table regclass ) returns void language plpgsql volatile as $$ declare select_fields text; join_location text; begin if src_table in ('content'::regclass, 'directory'::regclass) then - lock table only location; insert into location(path) select V.path from tmp_relation_add as V on conflict (path) do nothing; select_fields := 'D.id, L.id'; join_location := 'inner join location as L on (L.path = V.path)'; else select_fields := 'D.id'; join_location := ''; end if; execute format( - 'lock table only %s; - insert into %s + 'insert into %s select S.id, ' || select_fields || ' from tmp_relation_add as V inner join %s as S on (S.sha1 = V.src) inner join %s as D on (D.sha1 = V.dst) ' || join_location || ' on conflict do nothing', - rel_table, rel_table, src_table, dst_table + rel_table, src_table, dst_table ); end; $$; create or replace function swh_provenance_relation_get( rel_table regclass, src_table regclass, dst_table regclass, filter rel_flt, sha1s sha1_git[] ) returns table ( src sha1_git, dst sha1_git, path unix_path ) language plpgsql stable as $$ declare src_field text; dst_field text; join_location text; proj_location text; filter_result text; begin if rel_table = 'revision_before_revision'::regclass then src_field := 'prev'; dst_field := 'next'; else src_field := src_table::text; dst_field := dst_table::text; end if; if src_table in ('content'::regclass, 'directory'::regclass) then join_location := 'inner join location as L on (L.id = R.location)'; proj_location := 'L.path'; else join_location := ''; proj_location := 'NULL::unix_path'; end if; case filter when 'filter-src'::rel_flt then filter_result := 'where S.sha1 = any($1)'; when 'filter-dst'::rel_flt then filter_result := 'where D.sha1 = any($1)'; else filter_result := ''; end case; return query execute format( 'select S.sha1 as src, D.sha1 as dst, ' || proj_location || ' as path from %s as R inner join %s as S on (S.id = R.' || src_field || ') inner join %s as D on (D.id = R.' || dst_field || ') ' || join_location || ' ' || filter_result, rel_table, src_table, dst_table ) using sha1s; end; $$; \else -- -- without path and normalized -- create or replace function swh_provenance_content_find_first(content_id sha1_git) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ select C.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, '\x'::unix_path as path from content as C inner join content_in_revision as CR on (CR.content = C.id) inner join revision as R on (R.id = CR.revision) left join origin as O on (O.id = R.origin) where C.sha1 = content_id order by date, revision, origin asc limit 1 $$; create or replace function swh_provenance_content_find_all(content_id sha1_git, early_cut int) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ (select C.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, '\x'::unix_path as path from content as C inner join content_in_revision as CR on (CR.content = C.id) inner join revision as R on (R.id = CR.revision) left join origin as O on (O.id = R.origin) where C.sha1 = content_id) union (select C.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, '\x'::unix_path as path from content as C inner join content_in_directory as CD on (CD.content = C.id) inner join directory_in_revision as DR on (DR.directory = CD.directory) inner join revision as R on (R.id = DR.revision) left join origin as O on (O.id = R.origin) where C.sha1 = content_id) order by date, revision, origin, path limit early_cut $$; create or replace function swh_provenance_relation_add_from_temp( rel_table regclass, src_table regclass, dst_table regclass ) returns void language plpgsql volatile as $$ begin execute format( - 'lock table only %s; - insert into %s + 'insert into %s select S.id, D.id from tmp_relation_add as V inner join %s as S on (S.sha1 = V.src) inner join %s as D on (D.sha1 = V.dst) on conflict do nothing', - rel_table, rel_table, src_table, dst_table + rel_table, src_table, dst_table ); end; $$; create or replace function swh_provenance_relation_get( rel_table regclass, src_table regclass, dst_table regclass, filter rel_flt, sha1s sha1_git[] ) returns table ( src sha1_git, dst sha1_git, path unix_path ) language plpgsql stable as $$ declare src_field text; dst_field text; filter_result text; begin if rel_table = 'revision_before_revision'::regclass then src_field := 'prev'; dst_field := 'next'; else src_field := src_table::text; dst_field := dst_table::text; end if; case filter when 'filter-src'::rel_flt then filter_result := 'where S.sha1 = any($1)'; when 'filter-dst'::rel_flt then filter_result := 'where D.sha1 = any($1)'; else filter_result := ''; end case; return query execute format( 'select S.sha1 as src, D.sha1 as dst, NULL::unix_path as path from %s as R inner join %s as S on (S.id = R.' || src_field || ') inner join %s as D on (D.id = R.' || dst_field || ') ' || filter_result, rel_table, src_table, dst_table ) using sha1s; end; $$; -- :dbflavor_with_path \endif -- :dbflavor_norm \else \if :dbflavor_with_path -- -- with path and denormalized -- create or replace function swh_provenance_content_find_first(content_id sha1_git) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ select CL.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, L.path as path from ( select C.sha1 as sha1, unnest(CR.revision) as revision from content_in_revision as CR inner join content as C on (C.id = CR.content) where C.sha1 = content_id ) as CL inner join revision as R on (R.id = (CL.revision).id) inner join location as L on (L.id = (CL.revision).loc) left join origin as O on (O.id = R.origin) order by date, revision, origin, path asc limit 1 $$; create or replace function swh_provenance_content_find_all(content_id sha1_git, early_cut int) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ (with cntrev as ( select C.sha1 as sha1, unnest(CR.revision) as revision from content_in_revision as CR inner join content as C on (C.id = CR.content) where C.sha1 = content_id) select CR.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, L.path as path from cntrev as CR inner join revision as R on (R.id = (CR.revision).id) inner join location as L on (L.id = (CR.revision).loc) left join origin as O on (O.id = R.origin)) union (with cntdir as ( select C.sha1 as sha1, unnest(CD.directory) as directory from content as C inner join content_in_directory as CD on (CD.content = C.id) where C.sha1 = content_id), cntrev as ( select CD.sha1 as sha1, L.path as path, unnest(DR.revision) as revision from cntdir as CD inner join directory_in_revision as DR on (DR.directory = (CD.directory).id) inner join location as L on (L.id = (CD.directory).loc)) select CR.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, case DL.path when '' then CR.path when '.' then CR.path else (DL.path || '/' || CR.path)::unix_path end as path from cntrev as CR inner join revision as R on (R.id = (CR.revision).id) inner join location as DL on (DL.id = (CR.revision).loc) left join origin as O on (O.id = R.origin)) order by date, revision, origin, path limit early_cut $$; create or replace function swh_provenance_relation_add_from_temp( rel_table regclass, src_table regclass, dst_table regclass ) returns void language plpgsql volatile as $$ declare select_fields text; join_location text; group_entries text; on_conflict text; begin if src_table in ('content'::regclass, 'directory'::regclass) then - lock table only location; insert into location(path) select V.path from tmp_relation_add as V on conflict (path) do nothing; select_fields := 'array_agg((D.id, L.id)::rel_dst)'; join_location := 'inner join location as L on (L.path = V.path)'; group_entries := 'group by S.id'; on_conflict := format(' (%s) do update set %s=array( select distinct unnest( %s.' || dst_table::text || ' || excluded.' || dst_table::text || ' ) )', src_table, dst_table, rel_table, rel_table, rel_table ); else select_fields := 'D.id'; join_location := ''; group_entries := ''; on_conflict := 'do nothing'; end if; execute format( - 'lock table only %s; - insert into %s + 'insert into %s select S.id, ' || select_fields || ' from tmp_relation_add as V inner join %s as S on (S.sha1 = V.src) inner join %s as D on (D.sha1 = V.dst) ' || join_location || ' ' || group_entries || ' on conflict ' || on_conflict, - rel_table, rel_table, src_table, dst_table + rel_table, src_table, dst_table ); end; $$; create or replace function swh_provenance_relation_get( rel_table regclass, src_table regclass, dst_table regclass, filter rel_flt, sha1s sha1_git[] ) returns table ( src sha1_git, dst sha1_git, path unix_path ) language plpgsql stable as $$ declare src_field text; dst_field text; proj_dst_id text; proj_unnested text; proj_location text; join_location text; filter_inner_result text; filter_outer_result text; begin if rel_table = 'revision_before_revision'::regclass then src_field := 'prev'; dst_field := 'next'; else src_field := src_table::text; dst_field := dst_table::text; end if; if src_table in ('content'::regclass, 'directory'::regclass) then proj_unnested := 'unnest(R.' || dst_field || ') as dst'; proj_dst_id := '(CL.dst).id'; join_location := 'inner join location as L on (L.id = (CL.dst).loc)'; proj_location := 'L.path'; else proj_unnested := 'R.' || dst_field || ' as dst'; proj_dst_id := 'CL.dst'; join_location := ''; proj_location := 'NULL::unix_path'; end if; case filter when 'filter-src'::rel_flt then filter_inner_result := 'where S.sha1 = any($1)'; filter_outer_result := ''; when 'filter-dst'::rel_flt then filter_inner_result := ''; filter_outer_result := 'where D.sha1 = any($1)'; else filter_inner_result := ''; filter_outer_result := ''; end case; return query execute format( 'select CL.src, D.sha1 as dst, ' || proj_location || ' as path from (select S.sha1 as src, ' || proj_unnested || ' from %s as R inner join %s as S on (S.id = R.' || src_field || ') ' || filter_inner_result || ') as CL inner join %s as D on (D.id = ' || proj_dst_id || ') ' || join_location || ' ' || filter_outer_result, rel_table, src_table, dst_table ) using sha1s; end; $$; \else -- -- without path and denormalized -- create or replace function swh_provenance_content_find_first(content_id sha1_git) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ select CL.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, '\x'::unix_path as path from ( select C.sha1, unnest(revision) as revision from content_in_revision as CR inner join content as C on (C.id = CR.content) where C.sha1=content_id ) as CL inner join revision as R on (R.id = CL.revision) left join origin as O on (O.id = R.origin) order by date, revision, origin, path asc limit 1 $$; create or replace function swh_provenance_content_find_all(content_id sha1_git, early_cut int) returns table ( content sha1_git, revision sha1_git, date timestamptz, origin text, path unix_path ) language sql stable as $$ (with cntrev as ( select C.sha1 as sha1, unnest(CR.revision) as revision from content_in_revision as CR inner join content as C on (C.id = CR.content) where C.sha1 = content_id) select CR.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, '\x'::unix_path as path from cntrev as CR inner join revision as R on (R.id = CR.revision) left join origin as O on (O.id = R.origin)) union (with cntdir as ( select C.sha1 as sha1, unnest(CD.directory) as directory from content as C inner join content_in_directory as CD on (CD.content = C.id) where C.sha1 = content_id), cntrev as ( select CD.sha1 as sha1, unnest(DR.revision) as revision from cntdir as CD inner join directory_in_revision as DR on (DR.directory = CD.directory)) select CR.sha1 as content, R.sha1 as revision, R.date as date, O.url as origin, '\x'::unix_path as path from cntrev as CR inner join revision as R on (R.id = CR.revision) left join origin as O on (O.id = R.origin)) order by date, revision, origin, path limit early_cut $$; create or replace function swh_provenance_relation_add_from_temp( rel_table regclass, src_table regclass, dst_table regclass ) returns void language plpgsql volatile as $$ declare select_fields text; group_entries text; on_conflict text; begin if src_table in ('content'::regclass, 'directory'::regclass) then select_fields := 'array_agg(D.id)'; group_entries := 'group by S.id'; on_conflict := format(' (%s) do update set %s=array( select distinct unnest( %s.' || dst_table::text || ' || excluded.' || dst_table::text || ' ) )', src_table, dst_table, rel_table, rel_table ); else select_fields := 'D.id'; group_entries := ''; on_conflict := 'do nothing'; end if; execute format( - 'lock table only %s; - insert into %s + 'insert into %s select S.id, ' || select_fields || ' from tmp_relation_add as V inner join %s as S on (S.sha1 = V.src) inner join %s as D on (D.sha1 = V.dst) ' || group_entries || ' on conflict ' || on_conflict, - rel_table, rel_table, src_table, dst_table + rel_table, src_table, dst_table ); end; $$; create or replace function swh_provenance_relation_get( rel_table regclass, src_table regclass, dst_table regclass, filter rel_flt, sha1s sha1_git[] ) returns table ( src sha1_git, dst sha1_git, path unix_path ) language plpgsql stable as $$ declare src_field text; dst_field text; proj_unnested text; filter_inner_result text; filter_outer_result text; begin if rel_table = 'revision_before_revision'::regclass then src_field := 'prev'; dst_field := 'next'; else src_field := src_table::text; dst_field := dst_table::text; end if; if src_table in ('content'::regclass, 'directory'::regclass) then proj_unnested := 'unnest(R.' || dst_field || ') as dst'; else proj_unnested := 'R.' || dst_field || ' as dst'; end if; case filter when 'filter-src'::rel_flt then filter_inner_result := 'where S.sha1 = any($1)'; filter_outer_result := ''; when 'filter-dst'::rel_flt then filter_inner_result := ''; filter_outer_result := 'where D.sha1 = any($1)'; else filter_inner_result := ''; filter_outer_result := ''; end case; return query execute format( 'select CL.src, D.sha1 as dst, NULL::unix_path as path from (select S.sha1 as src, ' || proj_unnested || ' from %s as R inner join %s as S on (S.id = R.' || src_field || ') ' || filter_inner_result || ') as CL inner join %s as D on (D.id = CL.dst) ' || filter_outer_result, rel_table, src_table, dst_table ) using sha1s; end; $$; \endif -- :dbflavor_with_path \endif -- :dbflavor_norm