No OneTemporary
Actions

Size

104 KB

Subscribers

None

View Options

	diff --git a/sql/swh-func.sql b/sql/swh-func.sql
	index bba8d4d2f..c18389b0b 100644
	--- a/sql/swh-func.sql
	+++ b/sql/swh-func.sql
	@@ -1,1279 +1,1279 @@
	-- create a temporary table called tmp_TBLNAME, mimicking existing table
	-- TBLNAME
	--
	-- Args:
	-- tblname: name of the table to mimick
	create or replace function swh_mktemp(tblname regclass)
	returns void
	language plpgsql
	as $$
	begin
	execute format('
	create temporary table tmp_%1$I
	(like %1$I including defaults)
	on commit drop;
	alter table tmp_%1$I drop column if exists object_id;
	', tblname);
	return;
	end
	$$;

	-- create a temporary table for directory entries called tmp_TBLNAME,
	-- mimicking existing table TBLNAME with an extra dir_id (sha1_git)
	-- column, and dropping the id column.
	--
	-- This is used to create the tmp_directory_entry_<foo> tables.
	--
	-- Args:
	-- tblname: name of the table to mimick
	create or replace function swh_mktemp_dir_entry(tblname regclass)
	returns void
	language plpgsql
	as $$
	begin
	execute format('
	create temporary table tmp_%1$I
	(like %1$I including defaults, dir_id sha1_git)
	on commit drop;
	alter table tmp_%1$I drop column id;
	', tblname);
	return;
	end
	$$;


	-- create a temporary table for revisions called tmp_revisions,
	-- mimicking existing table revision, replacing the foreign keys to
	-- people with an email and name field
	--
	create or replace function swh_mktemp_revision()
	returns void
	language sql
	as $$
	create temporary table tmp_revision (
	like revision including defaults,
	author_name bytea not null default '',
	author_email bytea not null default '',
	committer_name bytea not null default '',
	committer_email bytea not null default ''
	) on commit drop;
	alter table tmp_revision drop column author;
	alter table tmp_revision drop column committer;
	alter table tmp_revision drop column object_id;
	$$;


	-- create a temporary table for releases called tmp_release,
	-- mimicking existing table release, replacing the foreign keys to
	-- people with an email and name field
	--
	create or replace function swh_mktemp_release()
	returns void
	language sql
	as $$
	create temporary table tmp_release (
	like release including defaults,
	author_name bytea not null default '',
	author_email bytea not null default ''
	) on commit drop;
	alter table tmp_release drop column author;
	alter table tmp_release drop column object_id;
	$$;

	-- create a temporary table with a single "bytea" column for fast object lookup.
	create or replace function swh_mktemp_bytea()
	returns void
	language sql
	as $$
	create temporary table tmp_bytea (
	id bytea
	) on commit drop;
	$$;

	-- create a temporary table for occurrence_history
	create or replace function swh_mktemp_occurrence_history()
	returns void
	language sql
	as $$
	create temporary table tmp_occurrence_history(
	like occurrence_history including defaults,
	date timestamptz not null
	) on commit drop;
	alter table tmp_occurrence_history
	drop column visits,
	drop column object_id;
	$$;

	-- create a temporary table for entity_history, sans id
	create or replace function swh_mktemp_entity_history()
	returns void
	language sql
	as $$
	create temporary table tmp_entity_history (
	like entity_history including defaults) on commit drop;
	alter table tmp_entity_history drop column id;
	$$;

	-- create a temporary table for entities called tmp_entity_lister,
	-- with only the columns necessary for retrieving the uuid of a listed
	-- entity.
	create or replace function swh_mktemp_entity_lister()
	returns void
	language sql
	as $$
	create temporary table tmp_entity_lister (
	id bigint,
	lister_metadata jsonb
	) on commit drop;
	$$;

	-- a content signature is a set of cryptographic checksums that we use to
	-- uniquely identify content, for the purpose of verifying if we already have
	-- some content or not during content injection
	create type content_signature as (
	sha1 sha1,
	sha1_git sha1_git,
	sha256 sha256
	);


	-- check which entries of tmp_content are missing from content
	--
	-- operates in bulk: 0. swh_mktemp(content), 1. COPY to tmp_content,
	-- 2. call this function
	create or replace function swh_content_missing()
	returns setof content_signature
	language plpgsql
	as $$
	begin
	-- This query is critical for (single-algorithm) hash collision detection,
	-- so we cannot rely only on the fact that a single hash (e.g., sha1) is
	-- missing from the table content to conclude that a given content is
	-- missing. Ideally, we would want to (try to) add to content all entries
	-- in tmp_content that, when considering all columns together, are missing
	-- from content.
	--
	-- But doing that naively would require a compound index on all checksum
	-- columns; that index would not be significantly smaller than the content
	-- table itself, and therefore won't be used. Therefore we union together
	-- all contents that differ on at least one column from what is already
	-- available. If there is a collision on some (but not all) columns, the
	-- relevant tmp_content entry will be included in the set of content to be
	-- added, causing a downstream violation of unicity constraint.
	return query
	(select sha1, sha1_git, sha256 from tmp_content as tmp
	where not exists
	(select 1 from content as c where c.sha1 = tmp.sha1))
	union
	(select sha1, sha1_git, sha256 from tmp_content as tmp
	where not exists
	(select 1 from content as c where c.sha1_git = tmp.sha1_git))
	union
	(select sha1, sha1_git, sha256 from tmp_content as tmp
	where not exists
	(select 1 from content as c where c.sha256 = tmp.sha256));
	return;
	end
	$$;

	-- check which entries of tmp_content_sha1 are missing from content
	--
	-- operates in bulk: 0. swh_mktemp_content_sha1(), 1. COPY to tmp_content_sha1,
	-- 2. call this function
	create or replace function swh_content_missing_per_sha1()
	returns setof sha1
	language plpgsql
	as $$
	begin
	return query
	(select id::sha1
	from tmp_bytea as tmp
	where not exists
	(select 1 from content as c where c.sha1=tmp.id));
	end
	$$;


	-- check which entries of tmp_skipped_content are missing from skipped_content
	--
	-- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content,
	-- 2. call this function
	create or replace function swh_skipped_content_missing()
	returns setof content_signature
	language plpgsql
	as $$
	begin
	return query
	select sha1, sha1_git, sha256 from tmp_skipped_content t
	where not exists
	(select 1 from skipped_content s where
	s.sha1 is not distinct from t.sha1 and
	s.sha1_git is not distinct from t.sha1_git and
	s.sha256 is not distinct from t.sha256);
	return;
	end
	$$;


	-- Look up content based on one or several different checksums. Return all
	-- content information if the content is found; a NULL row otherwise.
	--
	-- At least one checksum should be not NULL. If several are not NULL, they will
	-- be AND-ed together in the lookup query.
	--
	-- Note: this function is meant to be used to look up individual contents
	-- (e.g., for the web app), for batch lookup of missing content (e.g., to be
	-- added) see swh_content_missing
	create or replace function swh_content_find(
	sha1 sha1 default NULL,
	sha1_git sha1_git default NULL,
	sha256 sha256 default NULL
	)
	returns content
	language plpgsql
	as $$
	declare
	con content;
	filters text[] := array[] :: text[]; -- AND-clauses used to filter content
	q text;
	begin
	if sha1 is not null then
	filters := filters \|\| format('sha1 = %L', sha1);
	end if;
	if sha1_git is not null then
	filters := filters \|\| format('sha1_git = %L', sha1_git);
	end if;
	if sha256 is not null then
	filters := filters \|\| format('sha256 = %L', sha256);
	end if;

	if cardinality(filters) = 0 then
	return null;
	else
	q = format('select * from content where %s',
	array_to_string(filters, ' and '));
	execute q into con;
	return con;
	end if;
	end
	$$;


	-- add tmp_content entries to content, skipping duplicates
	--
	-- operates in bulk: 0. swh_mktemp(content), 1. COPY to tmp_content,
	-- 2. call this function
	create or replace function swh_content_add()
	returns void
	language plpgsql
	as $$
	begin
	insert into content (sha1, sha1_git, sha256, length, status)
	select distinct sha1, sha1_git, sha256, length, status
	from tmp_content
	where (sha1, sha1_git, sha256) in
	(select * from swh_content_missing());
	-- TODO XXX use postgres 9.5 "UPSERT" support here, when available.
	-- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid
	-- the extra swh_content_missing() query here.
	return;
	end
	$$;


	-- add tmp_skipped_content entries to skipped_content, skipping duplicates
	--
	-- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content,
	-- 2. call this function
	create or replace function swh_skipped_content_add()
	returns void
	language plpgsql
	as $$
	begin
	insert into skipped_content (sha1, sha1_git, sha256, length, status, reason, origin)
	select distinct sha1, sha1_git, sha256, length, status, reason, origin
	from tmp_skipped_content
	where (coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '')) in
	(select coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '') from swh_skipped_content_missing());
	-- TODO XXX use postgres 9.5 "UPSERT" support here, when available.
	-- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid
	-- the extra swh_content_missing() query here.
	return;
	end
	$$;


	-- check which entries of tmp_directory are missing from directory
	--
	-- operates in bulk: 0. swh_mktemp(directory), 1. COPY to tmp_directory,
	-- 2. call this function
	create or replace function swh_directory_missing()
	returns setof sha1_git
	language plpgsql
	as $$
	begin
	return query
	select id from tmp_directory t
	where not exists (
	select 1 from directory d
	where d.id = t.id);
	return;
	end
	$$;


	-- Retrieve information on directory from temporary table
	create or replace function swh_directory_get()
	returns setof directory
	language plpgsql
	as $$
	begin
	return query
	select d.*
	from tmp_directory t
	inner join directory d on t.id = d.id;
	return;
	end
	$$;


	create type directory_entry_type as enum('file', 'dir', 'rev');


	-- Add tmp_directory_entry_* entries to directory_entry_* and directory,
	-- skipping duplicates in directory_entry_*. This is a generic function that
	-- works on all kind of directory entries.
	--
	-- operates in bulk: 0. swh_mktemp_dir_entry('directory_entry_*'), 1 COPY to
	-- tmp_directory_entry_*, 2. call this function
	--
	-- Assumption: this function is used in the same transaction that inserts the
	-- context directory in table "directory".
	create or replace function swh_directory_entry_add(typ directory_entry_type)
	returns void
	language plpgsql
	as $$
	begin
	execute format('
	insert into directory_entry_%1$s (target, name, perms)
	select distinct t.target, t.name, t.perms
	from tmp_directory_entry_%1$s t
	where not exists (
	select 1
	from directory_entry_%1$s i
	where t.target = i.target and t.name = i.name and t.perms = i.perms)
	', typ);

	execute format('
	with new_entries as (
	select t.dir_id, array_agg(i.id) as entries
	from tmp_directory_entry_%1$s t
	inner join directory_entry_%1$s i
	using (target, name, perms)
	group by t.dir_id
	)
	update tmp_directory as d
	set %1$s_entries = new_entries.entries
	from new_entries
	where d.id = new_entries.dir_id
	', typ);

	return;
	end
	$$;

	-- Insert the data from tmp_directory, tmp_directory_entry_file,
	-- tmp_directory_entry_dir, tmp_directory_entry_rev into their final
	-- tables.
	--
	-- Prerequisites:
	-- directory ids in tmp_directory
	-- entries in tmp_directory_entry_{file,dir,rev}
	--
	create or replace function swh_directory_add()
	returns void
	language plpgsql
	as $$
	begin
	perform swh_directory_entry_add('file');
	perform swh_directory_entry_add('dir');
	perform swh_directory_entry_add('rev');

	insert into directory
	select * from tmp_directory t
	where not exists (
	select 1 from directory d
	where d.id = t.id);

	return;
	end
	$$;

	-- a directory listing entry with all the metadata
	--
	-- can be used to list a directory, and retrieve all the data in one go.
	create type directory_entry as
	(
	dir_id sha1_git, -- id of the parent directory
	type directory_entry_type, -- type of entry
	target sha1_git, -- id of target
	name unix_path, -- path name, relative to containing dir
	perms file_perms, -- unix-like permissions
	status content_status, -- visible or absent
	sha1 sha1, -- content if sha1 if type is not dir
	sha1_git sha1_git, -- content's sha1 git if type is not dir
	sha256 sha256 -- content's sha256 if type is not dir
	);


	-- List a single level of directory walked_dir_id
	-- FIXME: order by name is not correct. For git, we need to order by
	-- lexicographic order but as if a trailing / is present in directory
	-- name
	create or replace function swh_directory_walk_one(walked_dir_id sha1_git)
	returns setof directory_entry
	language sql
	stable
	as $$
	with dir as (
	select id as dir_id, dir_entries, file_entries, rev_entries
	from directory
	where id = walked_dir_id),
	ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir),
	ls_f as (select dir_id, unnest(file_entries) as entry_id from dir),
	ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir)
	(select dir_id, 'dir'::directory_entry_type as type,
	e.target, e.name, e.perms, NULL::content_status,
	NULL::sha1, NULL::sha1_git, NULL::sha256
	from ls_d
	left join directory_entry_dir e on ls_d.entry_id = e.id)
	union
	(select dir_id, 'file'::directory_entry_type as type,
	e.target, e.name, e.perms, c.status,
	c.sha1, c.sha1_git, c.sha256
	from ls_f
	left join directory_entry_file e on ls_f.entry_id = e.id
	left join content c on e.target = c.sha1_git)
	union
	(select dir_id, 'rev'::directory_entry_type as type,
	e.target, e.name, e.perms, NULL::content_status,
	NULL::sha1, NULL::sha1_git, NULL::sha256
	from ls_r
	left join directory_entry_rev e on ls_r.entry_id = e.id)
	order by name;
	$$;

	-- List recursively the content of a directory
	create or replace function swh_directory_walk(walked_dir_id sha1_git)
	returns setof directory_entry
	language sql
	stable
	as $$
	with recursive entries as (
	select dir_id, type, target, name, perms, status, sha1, sha1_git,
	sha256
	from swh_directory_walk_one(walked_dir_id)
	union all
	select dir_id, type, target, (dirname \|\| '/' \|\| name)::unix_path as name,
	perms, status, sha1, sha1_git, sha256
	from (select (swh_directory_walk_one(dirs.target)).*, dirs.name as dirname
	from (select target, name from entries where type = 'dir') as dirs) as with_parent
	)
	select dir_id, type, target, name, perms, status, sha1, sha1_git, sha256
	from entries
	$$;

	-- Find a directory entry by its path
	create or replace function swh_find_directory_entry_by_path(
	walked_dir_id sha1_git,
	dir_or_content_path bytea[])
	returns directory_entry
	language plpgsql
	as $$
	declare
	end_index integer;
	paths bytea default '';
	path bytea;
	res bytea[];
	r record;
	begin
	end_index := array_upper(dir_or_content_path, 1);
	res[1] := walked_dir_id;

	for i in 1..end_index
	loop
	path := dir_or_content_path[i];
	-- concatenate path for patching the name in the result record (if we found it)
	if i = 1 then
	paths = path;
	else
	paths := paths \|\| '/' \|\| path; -- concatenate paths
	end if;

	if i <> end_index then
	select *
	from swh_directory_walk_one(res[i] :: sha1_git)
	where name=path
	and type = 'dir'
	limit 1 into r;
	else
	select *
	from swh_directory_walk_one(res[i] :: sha1_git)
	where name=path
	limit 1 into r;
	end if;

	-- find the path
	if r is null then
	return null;
	else
	-- store the next dir to lookup the next local path from
	res[i+1] := r.target;
	end if;
	end loop;

	-- at this moment, r is the result. Patch its 'name' with the full path before returning it.
	r.name := paths;
	return r;
	end
	$$;

	-- List all revision IDs starting from a given revision, going back in time
	--
	-- TODO ordering: should be breadth-first right now (what do we want?)
	-- TODO ordering: ORDER BY parent_rank somewhere?
	create or replace function swh_revision_list(root_revisions bytea[], num_revs bigint default NULL)
	returns table (id sha1_git, parents bytea[])
	language sql
	stable
	as $$
	with recursive full_rev_list(id) as (
	(select id from revision where id = ANY(root_revisions))
	union
	(select h.parent_id
	from revision_history as h
	join full_rev_list on h.id = full_rev_list.id)
	),
	rev_list as (select id from full_rev_list limit num_revs)
	select rev_list.id as id,
	array(select rh.parent_id::bytea
	from revision_history rh
	where rh.id = rev_list.id
	order by rh.parent_rank
	) as parent
	from rev_list;
	$$;

	-- List all the children of a given revision
	create or replace function swh_revision_list_children(root_revisions bytea[], num_revs bigint default NULL)
	returns table (id sha1_git, parents bytea[])
	language sql
	stable
	as $$
	with recursive full_rev_list(id) as (
	(select id from revision where id = ANY(root_revisions))
	union
	(select h.id
	from revision_history as h
	join full_rev_list on h.parent_id = full_rev_list.id)
	),
	rev_list as (select id from full_rev_list limit num_revs)
	select rev_list.id as id,
	array(select rh.parent_id::bytea
	from revision_history rh
	where rh.id = rev_list.id
	order by rh.parent_rank
	) as parent
	from rev_list;
	$$;


	-- Detailed entry for a revision
	create type revision_entry as
	(
	id sha1_git,
	date timestamptz,
	date_offset smallint,
	date_neg_utc_offset boolean,
	committer_date timestamptz,
	committer_date_offset smallint,
	committer_date_neg_utc_offset boolean,
	type revision_type,
	directory sha1_git,
	message bytea,
	author_id bigint,
	author_name bytea,
	author_email bytea,
	committer_id bigint,
	committer_name bytea,
	committer_email bytea,
	metadata jsonb,
	synthetic boolean,
	parents bytea[]
	);


	-- "git style" revision log. Similar to swh_revision_list(), but returning all
	-- information associated to each revision, and expanding authors/committers
	create or replace function swh_revision_log(root_revisions bytea[], num_revs bigint default NULL)
	returns setof revision_entry
	language sql
	stable
	as $$
	select t.id, r.date, r.date_offset, r.date_neg_utc_offset,
	r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset,
	r.type, r.directory, r.message,
	a.id, a.name, a.email,
	c.id, c.name, c.email,
	r.metadata, r.synthetic, t.parents
	from swh_revision_list(root_revisions, num_revs) as t
	left join revision r on t.id = r.id
	left join person a on a.id = r.author
	left join person c on c.id = r.committer;
	$$;


	--- Retrieve revisions from tmp_revision in bulk
	+-- Retrieve revisions from tmp_bytea in bulk
	create or replace function swh_revision_get()
	returns setof revision_entry
	language plpgsql
	as $$
	begin
	return query
	- select t.id, r.date, r.date_offset, r.date_neg_utc_offset,
	+ select r.id, r.date, r.date_offset, r.date_neg_utc_offset,
	r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset,
	r.type, r.directory, r.message,
	a.id, a.name, a.email, c.id, c.name, c.email, r.metadata, r.synthetic,
	array(select rh.parent_id::bytea from revision_history rh where rh.id = t.id order by rh.parent_rank)
	as parents
	- from tmp_revision t
	+ from tmp_bytea t
	left join revision r on t.id = r.id
	left join person a on a.id = r.author
	left join person c on c.id = r.committer;
	return;
	end
	$$;

	--- List missing revisions from tmp_revision
	+-- List missing revisions from tmp_bytea
	create or replace function swh_revision_missing()
	returns setof sha1_git
	language plpgsql
	as $$
	begin
	return query
	- select id from tmp_revision t
	+ select id::sha1_git from tmp_bytea t
	where not exists (
	select 1 from revision r
	where r.id = t.id);
	return;
	end
	$$;

	-- Detailed entry for a release
	create type release_entry as
	(
	id sha1_git,
	target sha1_git,
	target_type object_type,
	date timestamptz,
	date_offset smallint,
	date_neg_utc_offset boolean,
	name bytea,
	comment bytea,
	synthetic boolean,
	author_id bigint,
	author_name bytea,
	author_email bytea
	);

	-- Detailed entry for release
	create or replace function swh_release_get()
	returns setof release_entry
	language plpgsql
	as $$
	begin
	return query
	select r.id, r.target, r.target_type, r.date, r.date_offset, r.date_neg_utc_offset, r.name, r.comment,
	r.synthetic, p.id as author_id, p.name as author_name, p.email as author_email
	from tmp_bytea t
	inner join release r on t.id = r.id
	inner join person p on p.id = r.author;
	return;
	end
	$$;

	-- Create entries in person from tmp_revision
	create or replace function swh_person_add_from_revision()
	returns void
	language plpgsql
	as $$
	begin
	with t as (
	select author_name as name, author_email as email from tmp_revision
	union
	select committer_name as name, committer_email as email from tmp_revision
	) insert into person (name, email)
	select distinct name, email from t
	where not exists (
	select 1
	from person p
	where t.name = p.name and t.email = p.email
	);
	return;
	end
	$$;


	-- Create entries in revision from tmp_revision
	create or replace function swh_revision_add()
	returns void
	language plpgsql
	as $$
	begin
	perform swh_person_add_from_revision();

	insert into revision (id, date, date_offset, date_neg_utc_offset, committer_date, committer_date_offset, committer_date_neg_utc_offset, type, directory, message, author, committer, metadata, synthetic)
	select t.id, t.date, t.date_offset, t.date_neg_utc_offset, t.committer_date, t.committer_date_offset, t.committer_date_neg_utc_offset, t.type, t.directory, t.message, a.id, c.id, t.metadata, t.synthetic
	from tmp_revision t
	left join person a on a.name = t.author_name and a.email = t.author_email
	left join person c on c.name = t.committer_name and c.email = t.committer_email;
	return;
	end
	$$;


	-- List missing releases from tmp_bytea
	create or replace function swh_release_missing()
	returns setof sha1_git
	language plpgsql
	as $$
	begin
	return query
	select id::sha1_git from tmp_bytea t
	where not exists (
	select 1 from release r
	where r.id = t.id);
	end
	$$;


	-- Create entries in person from tmp_release
	create or replace function swh_person_add_from_release()
	returns void
	language plpgsql
	as $$
	begin
	with t as (
	select distinct author_name as name, author_email as email from tmp_release
	) insert into person (name, email)
	select name, email from t
	where not exists (
	select 1
	from person p
	where t.name = p.name and t.email = p.email
	);
	return;
	end
	$$;


	-- Create entries in release from tmp_release
	create or replace function swh_release_add()
	returns void
	language plpgsql
	as $$
	begin
	perform swh_person_add_from_release();

	insert into release (id, target, target_type, date, date_offset, date_neg_utc_offset, name, comment, author, synthetic)
	select t.id, t.target, t.target_type, t.date, t.date_offset, t.date_neg_utc_offset, t.name, t.comment, a.id, t.synthetic
	from tmp_release t
	left join person a on a.name = t.author_name and a.email = t.author_email;
	return;
	end
	$$;

	create or replace function swh_occurrence_update_for_origin(origin_id bigint)
	returns void
	language sql
	as $$
	delete from occurrence where origin = origin_id;
	insert into occurrence (origin, branch, target, target_type)
	select origin, branch, target, target_type
	from occurrence_history
	where origin = origin_id and
	(select visit from origin_visit
	where origin = origin_id
	order by date desc
	limit 1) = any(visits);
	$$;

	create or replace function swh_occurrence_update_all()
	returns void
	language plpgsql
	as $$
	declare
	origin_id origin.id%type;
	begin
	for origin_id in
	select distinct id from origin
	loop
	perform swh_occurrence_update_for_origin(origin_id);
	end loop;
	return;
	end;
	$$;



	-- add tmp_occurrence_history entries to occurrence_history
	--
	-- operates in bulk: 0. swh_mktemp(occurrence_history), 1. COPY to tmp_occurrence_history,
	-- 2. call this function
	create or replace function swh_occurrence_history_add()
	returns void
	language plpgsql
	as $$
	declare
	origin_id origin.id%type;
	begin
	-- Create new visits
	with current_visits as (
	select distinct origin, date from tmp_occurrence_history
	),
	new_visits as (
	select origin, date, (select coalesce(max(visit), 0)
	from origin_visit ov
	where ov.origin = cv.origin) as max_visit
	from current_visits cv
	where not exists (select 1 from origin_visit ov
	where ov.origin = cv.origin and
	ov.date = cv.date)
	)
	insert into origin_visit (origin, date, visit)
	select origin, date, max_visit + row_number() over
	(partition by origin
	order by origin, date)
	from new_visits;

	-- Create or update occurrence_history
	with occurrence_history_id_visit as (
	select tmp_occurrence_history.*, object_id, visits, visit from tmp_occurrence_history
	left join occurrence_history using(origin, branch, target, target_type)
	left join origin_visit using(origin, date)
	),
	occurrences_to_update as (
	select object_id, visit from occurrence_history_id_visit where object_id is not null
	),
	update_occurrences as (
	update occurrence_history
	set visits = array(select unnest(occurrence_history.visits) as e
	union
	select occurrences_to_update.visit as e
	order by e)
	from occurrences_to_update
	where occurrence_history.object_id = occurrences_to_update.object_id
	)
	insert into occurrence_history (origin, branch, target, target_type, visits)
	select origin, branch, target, target_type, ARRAY[visit]
	from occurrence_history_id_visit
	where object_id is null;

	-- update occurrence
	for origin_id in
	select distinct origin from tmp_occurrence_history
	loop
	perform swh_occurrence_update_for_origin(origin_id);
	end loop;
	return;
	end
	$$;


	-- Absolute path: directory reference + complete path relative to it
	create type content_dir as (
	directory sha1_git,
	path unix_path
	);


	-- Find the containing directory of a given content, specified by sha1
	-- (note: not sha1_git).
	--
	-- Return a pair (dir_it, path) where path is a UNIX path that, from the
	-- directory root, reach down to a file with the desired content. Return NULL
	-- if no match is found.
	--
	-- In case of multiple paths (i.e., pretty much always), an arbitrary one is
	-- chosen.
	create or replace function swh_content_find_directory(content_id sha1)
	returns content_dir
	language sql
	stable
	as $$
	with recursive path as (
	-- Recursively build a path from the requested content to a root
	-- directory. Each iteration returns a pair (dir_id, filename) where
	-- filename is relative to dir_id. Stops when no parent directory can
	-- be found.
	(select dir.id as dir_id, dir_entry_f.name as name, 0 as depth
	from directory_entry_file as dir_entry_f
	join content on content.sha1_git = dir_entry_f.target
	join directory as dir on dir.file_entries @> array[dir_entry_f.id]
	where content.sha1 = content_id
	limit 1)
	union all
	(select dir.id as dir_id,
	(dir_entry_d.name \|\| '/' \|\| path.name)::unix_path as name,
	path.depth + 1
	from path
	join directory_entry_dir as dir_entry_d on dir_entry_d.target = path.dir_id
	join directory as dir on dir.dir_entries @> array[dir_entry_d.id]
	limit 1)
	)
	select dir_id, name from path order by depth desc limit 1;
	$$;


	-- Walk the revision history starting from a given revision, until a matching
	-- occurrence is found. Return all occurrence information if one is found, NULL
	-- otherwise.
	create or replace function swh_revision_find_occurrence(revision_id sha1_git)
	returns occurrence
	language sql
	stable
	as $$
	select origin, branch, target, target_type
	from swh_revision_list_children(ARRAY[revision_id] :: bytea[]) as rev_list
	left join occurrence_history occ_hist
	on rev_list.id = occ_hist.target
	where occ_hist.origin is not null and
	occ_hist.target_type = 'revision'
	limit 1;
	$$;

	-- Find the visit of origin id closest to date visit_date
	create or replace function swh_visit_find_by_date(origin bigint, visit_date timestamptz default NOW())
	returns origin_visit
	language sql
	stable
	as $$
	with closest_two_visits as ((
	select origin_visit, (date - visit_date) as interval
	from origin_visit
	where date >= visit_date
	order by date asc
	limit 1
	) union (
	select origin_visit, (visit_date - date) as interval
	from origin_visit
	where date < visit_date
	order by date desc
	limit 1
	)) select (origin_visit).* from closest_two_visits order by interval limit 1
	$$;


	-- Retrieve occurrence by filtering on origin_id and optionally on
	-- branch_name and/or validity range
	create or replace function swh_occurrence_get_by(
	origin_id bigint,
	branch_name bytea default NULL,
	date timestamptz default NULL)
	returns setof occurrence_history
	language plpgsql
	as $$
	declare
	filters text[] := array[] :: text[]; -- AND-clauses used to filter content
	visit_id bigint;
	q text;
	begin
	if origin_id is not null then
	filters := filters \|\| format('origin = %L', origin_id);
	end if;
	if branch_name is not null then
	filters := filters \|\| format('branch = %L', branch_name);
	end if;
	if date is not null then
	if origin_id is null then
	raise exception 'Needs an origin_id to filter by date.';
	end if;
	select visit from swh_visit_find_by_date(origin_id, date) into visit_id;
	if visit_id is null then
	return;
	end if;
	filters := filters \|\| format('%L = any(visits)', visit_id);
	end if;

	if cardinality(filters) = 0 then
	raise exception 'At least one filter amongst (origin_id, branch_name, date) is needed';
	else
	q = format('select * ' \|\|
	'from occurrence_history ' \|\|
	'where %s',
	array_to_string(filters, ' and '));
	return query execute q;
	end if;
	end
	$$;


	-- Retrieve revisions by occurrence criterion filtering
	create or replace function swh_revision_get_by(
	origin_id bigint,
	branch_name bytea default NULL,
	date timestamptz default NULL)
	returns setof revision_entry
	language sql
	stable
	as $$
	select r.id, r.date, r.date_offset, r.date_neg_utc_offset,
	r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset,
	r.type, r.directory, r.message,
	a.id, a.name, a.email, c.id, c.name, c.email, r.metadata, r.synthetic,
	array(select rh.parent_id::bytea
	from revision_history rh
	where rh.id = r.id
	order by rh.parent_rank
	) as parents
	from swh_occurrence_get_by(origin_id, branch_name, date) as occ
	inner join revision r on occ.target = r.id
	left join person a on a.id = r.author
	left join person c on c.id = r.committer;
	$$;

	-- Retrieve a release by occurrence criterion
	create or replace function swh_release_get_by(
	origin_id bigint)
	returns setof release_entry
	language sql
	stable
	as $$
	select r.id, r.target, r.target_type, r.date, r.date_offset, r.date_neg_utc_offset,
	r.name, r.comment, r.synthetic, a.id as author_id,
	a.name as author_name, a.email as author_email
	from release r
	inner join occurrence_history occ on occ.target = r.target
	left join person a on a.id = r.author
	where occ.origin = origin_id and occ.target_type = 'revision' and r.target_type = 'revision';
	$$;


	-- Occurrence of some content in a given context
	create type content_occurrence as (
	origin_type text,
	origin_url text,
	branch bytea,
	target sha1_git,
	target_type object_type,
	path unix_path
	);


	-- Given the sha1 of some content, look up an occurrence that points to a
	-- revision, which in turns reference (transitively) a tree containing the
	-- content. Answer the question: "where/when did SWH see a given content"?
	-- Return information about an arbitrary occurrence/revision/tree if one is
	-- found, NULL otherwise.
	create or replace function swh_content_find_occurrence(content_id sha1)
	returns content_occurrence
	language plpgsql
	as $$
	declare
	dir content_dir;
	rev sha1_git;
	occ occurrence%ROWTYPE;
	coc content_occurrence;
	begin
	-- each step could fail if no results are found, and that's OK
	select * from swh_content_find_directory(content_id) -- look up directory
	into dir;
	if not found then return null; end if;

	select id from revision where directory = dir.directory -- look up revision
	limit 1
	into rev;
	if not found then return null; end if;

	select * from swh_revision_find_occurrence(rev) -- look up occurrence
	into occ;
	if not found then return null; end if;

	select origin.type, origin.url, occ.branch, occ.target, occ.target_type, dir.path
	from origin
	where origin.id = occ.origin
	into coc;

	return coc; -- might be NULL
	end
	$$;

	create type object_found as (
	sha1_git sha1_git,
	type object_type,
	id bytea, -- sha1 or sha1_git depending on object_type
	object_id bigint
	);

	-- Find objects by sha1_git, return their type and their main identifier
	create or replace function swh_object_find_by_sha1_git()
	returns setof object_found
	language plpgsql
	as $$
	begin
	return query
	with known_objects as ((
	select id as sha1_git, 'release'::object_type as type, id, object_id from release r
	where exists (select 1 from tmp_bytea t where t.id = r.id)
	) union all (
	select id as sha1_git, 'revision'::object_type as type, id, object_id from revision r
	where exists (select 1 from tmp_bytea t where t.id = r.id)
	) union all (
	select id as sha1_git, 'directory'::object_type as type, id, object_id from directory d
	where exists (select 1 from tmp_bytea t where t.id = d.id)
	) union all (
	select sha1_git as sha1_git, 'content'::object_type as type, sha1 as id, object_id from content c
	where exists (select 1 from tmp_bytea t where t.id = c.sha1_git)
	))
	select t.id::sha1_git as sha1_git, k.type, k.id, k.object_id from tmp_bytea t
	left join known_objects k on t.id = k.sha1_git;
	end
	$$;

	-- Create entries in entity_history from tmp_entity_history
	--
	-- TODO: do something smarter to compress the entries if the data
	-- didn't change.
	create or replace function swh_entity_history_add()
	returns void
	language plpgsql
	as $$
	begin
	insert into entity_history (
	uuid, parent, name, type, description, homepage, active, generated, lister_metadata, metadata, validity
	) select * from tmp_entity_history;
	return;
	end
	$$;


	create or replace function swh_update_entity_from_entity_history()
	returns trigger
	language plpgsql
	as $$
	begin
	insert into entity (uuid, parent, name, type, description, homepage, active, generated,
	lister_metadata, metadata, last_seen, last_id)
	select uuid, parent, name, type, description, homepage, active, generated,
	lister_metadata, metadata, unnest(validity), id
	from entity_history
	where uuid = NEW.uuid
	order by unnest(validity) desc limit 1
	on conflict (uuid) do update set
	parent = EXCLUDED.parent,
	name = EXCLUDED.name,
	type = EXCLUDED.type,
	description = EXCLUDED.description,
	homepage = EXCLUDED.homepage,
	active = EXCLUDED.active,
	generated = EXCLUDED.generated,
	lister_metadata = EXCLUDED.lister_metadata,
	metadata = EXCLUDED.metadata,
	last_seen = EXCLUDED.last_seen,
	last_id = EXCLUDED.last_id;

	return null;
	end
	$$;

	create trigger update_entity
	after insert or update
	on entity_history
	for each row
	execute procedure swh_update_entity_from_entity_history();

	-- map an id of tmp_entity_lister to a full entity
	create type entity_id as (
	id bigint,
	uuid uuid,
	parent uuid,
	name text,
	type entity_type,
	description text,
	homepage text,
	active boolean,
	generated boolean,
	lister_metadata jsonb,
	metadata jsonb,
	last_seen timestamptz,
	last_id bigint
	);

	-- find out the uuid of the entries of entity with the metadata
	-- contained in tmp_entity_lister
	create or replace function swh_entity_from_tmp_entity_lister()
	returns setof entity_id
	language plpgsql
	as $$
	begin
	return query
	select t.id, e.*
	from tmp_entity_lister t
	left join entity e
	on e.lister_metadata @> t.lister_metadata;
	return;
	end
	$$;

	create or replace function swh_entity_get(entity_uuid uuid)
	returns setof entity
	language sql
	stable
	as $$
	with recursive entity_hierarchy as (
	select e.*
	from entity e where uuid = entity_uuid
	union
	select p.*
	from entity_hierarchy e
	join entity p on e.parent = p.uuid
	)
	select *
	from entity_hierarchy;
	$$;

	-- simple counter mapping a textual label to an integer value
	create type counter as (
	label text,
	value bigint
	);

	-- return statistics about the number of tuples in various SWH tables
	--
	-- Note: the returned values are based on postgres internal statistics
	-- (pg_class table), which are only updated daily (by autovacuum) or so
	create or replace function swh_stat_counters()
	returns setof counter
	language sql
	stable
	as $$
	select relname::text as label, reltuples::bigint as value
	from pg_class
	where oid in (
	'public.content'::regclass,
	'public.directory'::regclass,
	'public.directory_entry_dir'::regclass,
	'public.directory_entry_file'::regclass,
	'public.directory_entry_rev'::regclass,
	'public.occurrence'::regclass,
	'public.occurrence_history'::regclass,
	'public.origin'::regclass,
	'public.person'::regclass,
	'public.entity'::regclass,
	'public.entity_history'::regclass,
	'public.release'::regclass,
	'public.revision'::regclass,
	'public.revision_history'::regclass,
	'public.skipped_content'::regclass
	);
	$$;
	diff --git a/sql/swh-schema.sql b/sql/swh-schema.sql
	index 8749482da..26abf71a2 100644
	--- a/sql/swh-schema.sql
	+++ b/sql/swh-schema.sql
	@@ -1,424 +1,424 @@
	---
	--- Software Heritage Data Model
	---

	-- drop schema if exists swh cascade;
	-- create schema swh;
	-- set search_path to swh;

	create table dbversion
	(
	version int primary key,
	release timestamptz,
	description text
	);

	insert into dbversion(version, release, description)
	- values(66, now(), 'Work In Progress');
	+ values(67, now(), 'Work In Progress');

	-- a SHA1 checksum (not necessarily originating from Git)
	create domain sha1 as bytea check (length(value) = 20);

	-- a Git object ID, i.e., a SHA1 checksum
	create domain sha1_git as bytea check (length(value) = 20);

	-- a SHA256 checksum
	create domain sha256 as bytea check (length(value) = 32);

	-- UNIX path (absolute, relative, individual path component, etc.)
	create domain unix_path as bytea;

	-- a set of UNIX-like access permissions, as manipulated by, e.g., chmod
	create domain file_perms as int;

	create type content_status as enum ('absent', 'visible', 'hidden');

	-- Checksums about actual file content. Note that the content itself is not
	-- stored in the DB, but on external (key-value) storage. A single checksum is
	-- used as key there, but the other can be used to verify that we do not inject
	-- content collisions not knowingly.
	create table content
	(
	sha1 sha1 primary key,
	sha1_git sha1_git not null,
	sha256 sha256 not null,
	length bigint not null,
	ctime timestamptz not null default now(),
	-- creation time, i.e. time of (first) injection into the storage
	status content_status not null default 'visible',
	object_id bigserial
	);

	create unique index on content(sha1_git);
	create unique index on content(sha256);
	create index on content(ctime); -- TODO use a BRIN index here (postgres >= 9.5)

	-- Entities constitute a typed hierarchy of organization, hosting
	-- facilities, groups, people and software projects.
	--
	-- Examples of entities: Software Heritage, Debian, GNU, GitHub,
	-- Apache, The Linux Foundation, the Debian Python Modules Team, the
	-- torvalds GitHub user, the torvalds/linux GitHub project.
	--
	-- The data model is hierarchical (via the parent attribute) and might
	-- store sub-branches of existing entities. The key feature of an
	-- entity is might be listed (if it is available in listable_entity)
	-- to retrieve information about its content, i.e: sub-entities,
	-- projects, origins.

	-- Types of entities.
	--
	-- - organization: a root entity, usually backed by a non-profit, a
	-- company, or another kind of "association". (examples: Software
	-- Heritage, Debian, GNU, GitHub)
	--
	-- - group_of_entities: used for hierarchies, doesn't need to have a
	-- concrete existence. (examples: GNU hosting facilities, Debian
	-- hosting facilities, GitHub users, ...)
	--
	-- - hosting: a hosting facility, can usually be listed to generate
	-- other data. (examples: GitHub git hosting, alioth.debian.org,
	-- snapshot.debian.org)
	--
	-- - group_of_persons: an entity representing a group of
	-- persons. (examples: a GitHub organization, a Debian team)
	--
	-- - person: an entity representing a person. (examples:
	-- a GitHub user, a Debian developer)
	--
	-- - project: an entity representing a software project. (examples: a
	-- GitHub project, Apache httpd, a Debian source package, ...)
	create type entity_type as enum (
	'organization',
	'group_of_entities',
	'hosting',
	'group_of_persons',
	'person',
	'project'
	);

	-- The history of entities. Allows us to keep historical metadata
	-- about entities. The temporal invariant is the uuid. Root
	-- organization uuids are manually generated (and available in
	-- swh-data.sql).
	--
	-- For generated entities (generated = true), we can provide
	-- generation_metadata to allow listers to retrieve the uuids of previous
	-- iterations of the entity.
	--
	-- Inactive entities that have been active in the past (active =
	-- false) should register the timestamp at which we saw them
	-- deactivate, in a new entry of entity_history.
	create table entity_history
	(
	id bigserial primary key,
	uuid uuid,
	parent uuid, -- should reference entity_history(uuid)
	name text not null,
	type entity_type not null,
	description text,
	homepage text,
	active boolean not null, -- whether the entity was seen on the last listing
	generated boolean not null, -- whether this entity has been generated by a lister
	lister_metadata jsonb, -- lister-specific metadata, used for queries
	metadata jsonb,
	validity timestamptz[] -- timestamps at which we have seen this entity
	);

	create index on entity_history(uuid);
	create index on entity_history(name);

	-- The entity table provides a view of the latest information on a
	-- given entity. It is updated via a trigger on entity_history.
	create table entity
	(
	uuid uuid primary key,
	parent uuid references entity(uuid) deferrable initially deferred,
	name text not null,
	type entity_type not null,
	description text,
	homepage text,
	active boolean not null, -- whether the entity was seen on the last listing
	generated boolean not null, -- whether this entity has been generated by a lister
	lister_metadata jsonb, -- lister-specific metadata, used for queries
	metadata jsonb,
	last_seen timestamptz, -- last listing time or disappearance time for active=false
	last_id bigint references entity_history(id) -- last listing id
	);

	create index on entity(name);
	create index on entity using gin(lister_metadata jsonb_path_ops);

	-- Register the equivalence between two entities. Allows sideways
	-- navigation in the entity table
	create table entity_equivalence
	(
	entity1 uuid references entity(uuid),
	entity2 uuid references entity(uuid),
	primary key (entity1, entity2),
	constraint order_entities check (entity1 < entity2)
	);

	-- Register a lister for a specific entity.
	create table listable_entity
	(
	uuid uuid references entity(uuid) primary key,
	enabled boolean not null default true, -- do we list this entity automatically?
	list_engine text, -- crawler to be used to list entity's content
	list_url text, -- root URL to start the listing
	list_params jsonb, -- org-specific listing parameter
	latest_list timestamptz -- last time the entity's content has been listed
	);

	-- Log of all entity listings (i.e., entity crawling) that have been
	-- done in the past, or are still ongoing.
	create table list_history
	(
	id bigserial primary key,
	entity uuid references listable_entity(uuid),
	date timestamptz not null,
	status boolean, -- true if and only if the listing has been successful
	result jsonb, -- more detailed return value, depending on status
	stdout text,
	stderr text,
	duration interval -- fetch duration of NULL if still ongoing
	);


	-- An origin is a place, identified by an URL, where software can be found. We
	-- support different kinds of origins, e.g., git and other VCS repositories,
	-- web pages that list tarballs URLs (e.g., http://www.kernel.org), indirect
	-- tarball URLs (e.g., http://www.example.org/latest.tar.gz), etc. The key
	-- feature of an origin is that it can be fetched (wget, git clone, svn
	-- checkout, etc.) to retrieve all the contained software.
	create table origin
	(
	id bigserial primary key,
	type text, -- TODO use an enum here (?)
	url text not null,
	lister uuid references listable_entity(uuid),
	project uuid references entity(uuid)
	);

	create index on origin(type, url);

	-- Content we have seen but skipped for some reason. This table is
	-- separate from the content table as we might not have the sha1
	-- checksum of that data (for instance when we inject git
	-- repositories, objects that are too big will be skipped here, and we
	-- will only know their sha1_git). 'reason' contains the reason the
	-- content was skipped. origin is a nullable column allowing to find
	-- out which origin contains that skipped content.
	create table skipped_content
	(
	sha1 sha1,
	sha1_git sha1_git,
	sha256 sha256,
	length bigint not null,
	ctime timestamptz not null default now(),
	status content_status not null default 'absent',
	reason text not null,
	origin bigint references origin(id),
	object_id bigserial,
	unique (sha1, sha1_git, sha256)
	);

	-- those indexes support multiple NULL values.
	create unique index on skipped_content(sha1);
	create unique index on skipped_content(sha1_git);
	create unique index on skipped_content(sha256);


	-- Log of all origin fetches (i.e., origin crawling) that have been done in the
	-- past, or are still ongoing. Similar to list_history, but for origins.
	create table fetch_history
	(
	id bigserial primary key,
	origin bigint references origin(id),
	date timestamptz not null,
	status boolean, -- true if and only if the fetch has been successful
	result jsonb, -- more detailed returned values, times, etc...
	stdout text,
	stderr text, -- null when status is true, filled otherwise
	duration interval -- fetch duration of NULL if still ongoing
	);


	-- A file-system directory. A directory is a list of directory entries (see
	-- tables: directory_entry_{dir,file}).
	--
	-- To list the contents of a directory:
	-- 1. list the contained directory_entry_dir using array dir_entries
	-- 2. list the contained directory_entry_file using array file_entries
	-- 3. list the contained directory_entry_rev using array rev_entries
	-- 4. UNION
	--
	-- Synonyms/mappings:
	-- * git: tree
	create table directory
	(
	id sha1_git primary key,
	dir_entries bigint[], -- sub-directories, reference directory_entry_dir
	file_entries bigint[], -- contained files, reference directory_entry_file
	rev_entries bigint[], -- mounted revisions, reference directory_entry_rev
	object_id bigserial -- short object identifier
	);

	create index on directory using gin (dir_entries);
	create index on directory using gin (file_entries);
	create index on directory using gin (rev_entries);

	-- A directory entry pointing to a sub-directory.
	create table directory_entry_dir
	(
	id bigserial primary key,
	target sha1_git, -- id of target directory
	name unix_path, -- path name, relative to containing dir
	perms file_perms -- unix-like permissions
	);

	create unique index on directory_entry_dir(target, name, perms);

	-- A directory entry pointing to a file.
	create table directory_entry_file
	(
	id bigserial primary key,
	target sha1_git, -- id of target file
	name unix_path, -- path name, relative to containing dir
	perms file_perms -- unix-like permissions
	);

	create unique index on directory_entry_file(target, name, perms);

	-- A directory entry pointing to a revision.
	create table directory_entry_rev
	(
	id bigserial primary key,
	target sha1_git, -- id of target revision
	name unix_path, -- path name, relative to containing dir
	perms file_perms -- unix-like permissions
	);

	create unique index on directory_entry_rev(target, name, perms);

	create table person
	(
	id bigserial primary key,
	name bytea not null default '',
	email bytea not null default ''
	);

	create unique index on person(name, email);

	create type revision_type as enum ('git', 'tar', 'dsc', 'svn');

	-- the data object types stored in our data model
	create type object_type as enum ('content', 'directory', 'revision', 'release');

	-- A snapshot of a software project at a specific point in time.
	--
	-- Synonyms/mappings:
	-- * git / subversion / etc: commit
	-- * tarball: a specific tarball
	--
	-- Revisions are organized as DAGs. Each revision points to 0, 1, or more (in
	-- case of merges) parent revisions. Each revision points to a directory, i.e.,
	-- a file-system tree containing files and directories.
	create table revision
	(
	id sha1_git primary key,
	date timestamptz,
	date_offset smallint,
	date_neg_utc_offset boolean,
	committer_date timestamptz,
	committer_date_offset smallint,
	committer_date_neg_utc_offset boolean,
	type revision_type not null,
	directory sha1_git, -- file-system tree
	message bytea,
	author bigint references person(id),
	committer bigint references person(id),
	metadata jsonb, -- extra metadata (tarball checksums, extra commit information, etc...)
	synthetic boolean not null default false, -- true if synthetic (cf. swh-loader-tar)
	object_id bigserial
	);

	create index on revision(directory);

	-- either this table or the sha1_git[] column on the revision table
	create table revision_history
	(
	id sha1_git references revision(id),
	parent_id sha1_git,
	parent_rank int not null default 0,
	-- parent position in merge commits, 0-based
	primary key (id, parent_rank)
	);

	create index on revision_history(parent_id);

	-- The timestamps at which Software Heritage has made a visit of the given origin.
	create table origin_visit
	(
	origin bigint not null references origin(id),
	visit bigint not null,
	date timestamptz not null,
	primary key (origin, visit)
	);

	create index on origin_visit(date);

	-- The content of software origins is indexed starting from top-level pointers
	-- called "branches". Every time we fetch some origin we store in this table
	-- where the branches pointed to at fetch time.
	--
	-- Synonyms/mappings:
	-- * git: ref (in the "git update-ref" sense)
	create table occurrence_history
	(
	origin bigint references origin(id) not null,
	branch bytea not null, -- e.g., b"master" (for VCS), or b"sid" (for Debian)
	target sha1_git not null, -- ref target, e.g., commit id
	target_type object_type not null, -- ref target type
	object_id bigserial not null, -- short object identifier
	visits bigint[] not null, -- the visits where that occurrence was valid. References
	-- origin_visit(visit), where o_h.origin = origin_visit.origin.
	primary key (object_id)
	);

	create index on occurrence_history(target, target_type);
	create index on occurrence_history(origin, branch);
	create unique index on occurrence_history(origin, branch, target, target_type);

	-- Materialized view of occurrence_history, storing the current value of each
	-- branch, as last seen by SWH.
	create table occurrence
	(
	origin bigint references origin(id) not null,
	branch bytea not null,
	target sha1_git not null,
	target_type object_type not null,
	primary key(origin, branch)
	);

	-- A "memorable" point in the development history of a project.
	--
	-- Synonyms/mappings:
	-- * git: tag (of the annotated kind, otherwise they are just references)
	-- * tarball: the release version number
	create table release
	(
	id sha1_git primary key,
	target sha1_git,
	target_type object_type,
	date timestamptz,
	date_offset smallint,
	date_neg_utc_offset boolean,
	name bytea,
	comment bytea,
	author bigint references person(id),
	synthetic boolean not null default false, -- true if synthetic (cf. swh-loader-tar)
	object_id bigserial
	);

	create index on release(target, target_type);
	diff --git a/sql/upgrades/067.sql b/sql/upgrades/067.sql
	new file mode 100644
	index 000000000..a4dce40d3
	--- /dev/null
	+++ b/sql/upgrades/067.sql
	@@ -0,0 +1,39 @@
	+-- SWH DB schema upgrade
	+-- from_version: 66
	+-- to_version: 67
	+-- description: use tmp_bytea instead of tmp_revision to retrieve revision info
	+
	+insert into dbversion(version, release, description)
	+ values(67, now(), 'Work In Progress');
	+
	+CREATE OR REPLACE FUNCTION swh_revision_get() RETURNS SETOF revision_entry
	+ LANGUAGE plpgsql
	+ AS $$
	+begin
	+ return query
	+ select r.id, r.date, r.date_offset, r.date_neg_utc_offset,
	+ r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset,
	+ r.type, r.directory, r.message,
	+ a.id, a.name, a.email, c.id, c.name, c.email, r.metadata, r.synthetic,
	+ array(select rh.parent_id::bytea from revision_history rh where rh.id = t.id order by rh.parent_rank)
	+ as parents
	+ from tmp_bytea t
	+ left join revision r on t.id = r.id
	+ left join person a on a.id = r.author
	+ left join person c on c.id = r.committer;
	+ return;
	+end
	+$$;
	+
	+CREATE OR REPLACE FUNCTION swh_revision_missing() RETURNS SETOF sha1_git
	+ LANGUAGE plpgsql
	+ AS $$
	+begin
	+ return query
	+ select id::sha1_git from tmp_bytea t
	+ where not exists (
	+ select 1 from revision r
	+ where r.id = t.id);
	+ return;
	+end
	+$$;
	diff --git a/swh/storage/api/client.py b/swh/storage/api/client.py
	index cdc39957f..937d7f6ae 100644
	--- a/swh/storage/api/client.py
	+++ b/swh/storage/api/client.py
	@@ -1,200 +1,199 @@
	# Copyright (C) 2015 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

	import pickle

	import requests

	from requests.exceptions import ConnectionError
	from swh.core.serializers import msgpack_dumps, msgpack_loads, SWHJSONDecoder
	from swh.storage.exc import StorageAPIError


	def encode_data(data):
	try:
	return msgpack_dumps(data)
	except OverflowError as e:
	raise ValueError('Limits were reached. Please, check your input.\n' +
	str(e))


	def decode_response(response):
	content_type = response.headers['content-type']

	if content_type.startswith('application/x-msgpack'):
	r = msgpack_loads(response.content)
	elif content_type.startswith('application/json'):
	r = response.json(cls=SWHJSONDecoder)
	else:
	raise ValueError('Wrong content type `%s` for API response'
	% content_type)

	return r


	class RemoteStorage():
	"""Proxy to a remote storage API"""
	def __init__(self, base_url):
	self.base_url = base_url
	self.session = requests.Session()

	def url(self, endpoint):
	return '%s%s' % (self.base_url, endpoint)

	def post(self, endpoint, data):
	try:
	response = self.session.post(
	self.url(endpoint),
	data=encode_data(data),
	headers={'content-type': 'application/x-msgpack'},
	)
	except ConnectionError as e:
	print(str(e))
	raise StorageAPIError(e)

	# XXX: this breaks language-independence and should be
	# replaced by proper unserialization
	if response.status_code == 400:
	raise pickle.loads(decode_response(response))

	return decode_response(response)

	def get(self, endpoint, data=None):
	try:
	response = self.session.get(
	self.url(endpoint),
	params=data,
	)
	except ConnectionError as e:
	print(str(e))
	raise StorageAPIError(e)

	if response.status_code == 404:
	return None

	# XXX: this breaks language-independence and should be
	# replaced by proper unserialization
	if response.status_code == 400:
	raise pickle.loads(decode_response(response))

	else:
	return decode_response(response)

	def content_add(self, content):
	return self.post('content/add', {'content': content})

	def content_missing(self, content, key_hash='sha1'):
	return self.post('content/missing', {'content': content,
	'key_hash': key_hash})

	def content_missing_per_sha1(self, contents):
	return self.post('content/missing/sha1', {'contents': contents})

	def content_get(self, content):
	return self.post('content/data', {'content': content})

	def content_find(self, content):
	return self.post('content/present', {'content': content})

	def content_find_occurrence(self, content):
	return self.post('content/occurrence', {'content': content})

	def directory_add(self, directories):
	return self.post('directory/add', {'directories': directories})

	def directory_missing(self, directories):
	return self.post('directory/missing', {'directories': directories})

	def directory_get(self, directories):
	return self.post('directory', dict(directories=directories))

	def directory_ls(self, directory, recursive=False):
	return self.get('directory/ls', {'directory': directory,
	'recursive': recursive})

	def revision_get(self, revisions):
	return self.post('revision', {'revisions': revisions})

	def revision_get_by(self, origin_id, branch_name, timestamp, limit=None):
	return self.post('revision/by', dict(origin_id=origin_id,
	branch_name=branch_name,
	timestamp=timestamp,
	limit=limit))

	def revision_log(self, revisions, limit=None):
	return self.post('revision/log', {'revisions': revisions,
	'limit': limit})

	def revision_shortlog(self, revisions, limit=None):
	return self.post('revision/shortlog', {'revisions': revisions,
	'limit': limit})

	def revision_add(self, revisions):
	return self.post('revision/add', {'revisions': revisions})

	- def revision_missing(self, revisions, type='git'):
	- return self.post('revision/missing', {'revisions': revisions,
	- 'type': type})
	+ def revision_missing(self, revisions):
	+ return self.post('revision/missing', {'revisions': revisions})

	def release_add(self, releases):
	return self.post('release/add', {'releases': releases})

	def release_get(self, releases):
	return self.post('release', {'releases': releases})

	def release_get_by(self, origin_id, limit=None):
	return self.post('release/by', dict(origin_id=origin_id,
	limit=limit))

	def release_missing(self, releases):
	return self.post('release/missing', {'releases': releases})

	def object_find_by_sha1_git(self, ids):
	return self.post('object/find_by_sha1_git', {'ids': ids})

	def occurrence_get(self, origin_id):
	return self.post('occurrence', {'origin_id': origin_id})

	def occurrence_add(self, occurrences):
	return self.post('occurrence/add', {'occurrences': occurrences})

	def origin_get(self, origin):
	return self.post('origin/get', {'origin': origin})

	def origin_add_one(self, origin):
	return self.post('origin/add', {'origin': origin})

	def person_get(self, person):
	return self.post('person', {'person': person})

	def fetch_history_start(self, origin_id):
	return self.post('fetch_history/start', {'origin_id': origin_id})

	def fetch_history_end(self, fetch_history_id, data):
	return self.post('fetch_history/end',
	{'fetch_history_id': fetch_history_id,
	'data': data})

	def fetch_history_get(self, fetch_history_id):
	return self.get('fetch_history', {'id': fetch_history_id})

	def entity_add(self, entities):
	return self.post('entity/add', {'entities': entities})

	def entity_get(self, uuid):
	return self.post('entity/get', {'uuid': uuid})

	def entity_get_one(self, uuid):
	return self.get('entity', {'uuid': uuid})

	def entity_get_from_lister_metadata(self, entities):
	return self.post('entity/from_lister_metadata', {'entities': entities})

	def stat_counters(self):
	return self.get('stat/counters')

	def directory_entry_get_by_path(self, directory, paths):
	return self.post('directory/path', dict(directory=directory,
	paths=paths))
	diff --git a/swh/storage/storage.py b/swh/storage/storage.py
	index f91117d4a..93dd72612 100644
	--- a/swh/storage/storage.py
	+++ b/swh/storage/storage.py
	@@ -1,1123 +1,1111 @@
	# Copyright (C) 2015 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 collections import defaultdict
	import datetime
	import functools
	import itertools
	import dateutil.parser
	import psycopg2

	from . import converters
	from .db import Db
	from .objstorage import ObjStorage
	from .exc import ObjNotFoundError, StorageDBError

	from swh.core.hashutil import ALGORITHMS

	# Max block size of contents to return
	BULK_BLOCK_CONTENT_LEN_MAX = 10000


	def db_transaction(meth):
	"""decorator to execute Storage methods within DB transactions

	The decorated method must accept a `cur` keyword argument
	"""
	@functools.wraps(meth)
	def _meth(self, args, *kwargs):
	with self.db.transaction() as cur:
	return meth(self, args, cur=cur, *kwargs)
	return _meth


	def db_transaction_generator(meth):
	"""decorator to execute Storage methods within DB transactions, while
	returning a generator

	The decorated method must accept a `cur` keyword argument

	"""
	@functools.wraps(meth)
	def _meth(self, args, *kwargs):
	with self.db.transaction() as cur:
	yield from meth(self, args, cur=cur, *kwargs)
	return _meth


	class Storage():
	"""SWH storage proxy, encompassing DB and object storage

	"""

	def __init__(self, db_conn, obj_root):
	"""
	Args:
	db_conn: either a libpq connection string, or a psycopg2 connection
	obj_root: path to the root of the object storage

	"""
	try:
	if isinstance(db_conn, psycopg2.extensions.connection):
	self.db = Db(db_conn)
	else:
	self.db = Db.connect(db_conn)
	except psycopg2.OperationalError as e:
	raise StorageDBError(e)

	self.objstorage = ObjStorage(obj_root)

	def content_add(self, content):
	"""Add content blobs to the storage

	Note: in case of DB errors, objects might have already been added to
	the object storage and will not be removed. Since addition to the
	object storage is idempotent, that should not be a problem.

	Args:
	content: iterable of dictionaries representing individual pieces of
	content to add. Each dictionary has the following keys:
	- data (bytes): the actual content
	- length (int): content length (default: -1)
	- one key for each checksum algorithm in
	swh.core.hashutil.ALGORITHMS, mapped to the corresponding
	checksum
	- status (str): one of visible, hidden, absent
	- reason (str): if status = absent, the reason why
	- origin (int): if status = absent, the origin we saw the
	content in

	"""
	db = self.db

	content_by_status = defaultdict(list)
	for d in content:
	if 'status' not in d:
	d['status'] = 'visible'
	if 'length' not in d:
	d['length'] = -1
	content_by_status[d['status']].append(d)

	content_with_data = content_by_status['visible']
	content_without_data = content_by_status['absent']

	missing_content = set(self.content_missing(content_with_data))
	missing_skipped = set(
	sha1_git for sha1, sha1_git, sha256
	in self.skipped_content_missing(content_without_data))

	with db.transaction() as cur:
	if missing_content:
	# create temporary table for metadata injection
	db.mktemp('content', cur)

	def add_to_objstorage(cont):
	self.objstorage.add_bytes(cont['data'],
	obj_id=cont['sha1'])

	content_filtered = (cont for cont in content_with_data
	if cont['sha1'] in missing_content)

	db.copy_to(content_filtered, 'tmp_content',
	['sha1', 'sha1_git', 'sha256', 'length', 'status'],
	cur, item_cb=add_to_objstorage)

	# move metadata in place
	db.content_add_from_temp(cur)

	if missing_skipped:
	missing_filtered = (cont for cont in content_without_data
	if cont['sha1_git'] in missing_skipped)
	db.mktemp('skipped_content', cur)
	db.copy_to(missing_filtered, 'tmp_skipped_content',
	['sha1', 'sha1_git', 'sha256', 'length',
	'reason', 'status', 'origin'], cur)

	# move metadata in place
	db.skipped_content_add_from_temp(cur)

	def content_get(self, content):
	"""Retrieve in bulk contents and their data.

	Args:
	content: iterables of sha1

	Returns:
	Generates streams of contents as dict with their raw data:
	- sha1: sha1's content
	- data: bytes data of the content

	Raises:
	ValueError in case of too much contents are required.
	cf. BULK_BLOCK_CONTENT_LEN_MAX

	"""
	# FIXME: Improve on server module to slice the result
	if len(content) > BULK_BLOCK_CONTENT_LEN_MAX:
	raise ValueError(
	"Send at maximum %s contents." % BULK_BLOCK_CONTENT_LEN_MAX)

	for obj_id in content:
	try:
	data = self.objstorage.get_bytes(obj_id)
	except ObjNotFoundError:
	yield None
	continue

	yield {'sha1': obj_id, 'data': data}

	@db_transaction_generator
	def content_missing(self, content, key_hash='sha1', cur=None):
	"""List content missing from storage

	Args:
	content: iterable of dictionaries containing one key for each
	checksum algorithm in swh.core.hashutil.ALGORITHMS, mapped to
	the corresponding checksum, and a length key mapped to the
	content length.
	key_hash: the name of the hash used as key (default: 'sha1')

	Returns:
	an iterable of `key_hash`es missing from the storage

	Raises:
	TODO: an exception when we get a hash collision.

	"""
	db = self.db

	keys = ['sha1', 'sha1_git', 'sha256']

	if key_hash not in keys:
	raise ValueError("key_hash should be one of %s" % keys)

	key_hash_idx = keys.index(key_hash)

	# Create temporary table for metadata injection
	db.mktemp('content', cur)

	db.copy_to(content, 'tmp_content', keys + ['length'], cur)

	for obj in db.content_missing_from_temp(cur):
	yield obj[key_hash_idx]

	@db_transaction_generator
	def content_missing_per_sha1(self, contents, cur=None):
	"""List content missing from storage based only on sha1.

	Args:
	contents: Iterable of sha1 to check for absence.

	Returns:
	an iterable of `sha1`s missing from the storage.

	Raises:
	TODO: an exception when we get a hash collision.

	"""
	db = self.db

	db.store_tmp_bytea(contents, cur)
	for obj in db.content_missing_per_sha1_from_temp(cur):
	yield obj[0]

	@db_transaction_generator
	def skipped_content_missing(self, content, cur=None):
	"""List skipped_content missing from storage

	Args:
	content: iterable of dictionaries containing the data for each
	checksum algorithm.

	Returns:
	an iterable of signatures missing from the storage
	"""
	keys = ['sha1', 'sha1_git', 'sha256']

	db = self.db

	db.mktemp('skipped_content', cur)
	db.copy_to(content, 'tmp_skipped_content',
	keys + ['length', 'reason'], cur)

	yield from db.skipped_content_missing_from_temp(cur)

	@db_transaction
	def content_find(self, content, cur=None):
	"""Find a content hash in db.

	Args:
	content: a dictionary representing one content hash, mapping
	checksum algorithm names (see swh.core.hashutil.ALGORITHMS) to
	checksum values

	Returns:
	a triplet (sha1, sha1_git, sha256) if the content exist
	or None otherwise.

	Raises:
	ValueError in case the key of the dictionary is not sha1, sha1_git
	nor sha256.

	"""
	db = self.db

	if not set(content).intersection(ALGORITHMS):
	raise ValueError('content keys must contain at least one of: '
	'sha1, sha1_git, sha256')

	c = db.content_find(sha1=content.get('sha1'),
	sha1_git=content.get('sha1_git'),
	sha256=content.get('sha256'),
	cur=cur)
	if c:
	keys = ['sha1', 'sha1_git', 'sha256', 'length', 'ctime', 'status']
	return dict(zip(keys, c))
	return None

	@db_transaction
	def content_find_occurrence(self, content, cur=None):
	"""Find the content's occurrence.

	Args:
	content: a dictionary entry representing one content hash.
	The dictionary key is one of swh.core.hashutil.ALGORITHMS.
	The value mapped to the corresponding checksum.

	Returns:
	The occurrence of the content.

	Raises:
	ValueError in case the key of the dictionary is not sha1, sha1_git
	nor sha256.

	"""
	db = self.db

	c = self.content_find(content)

	if not c:
	return None

	sha1 = c['sha1']

	found_occ = db.content_find_occurrence(sha1, cur=cur)
	if found_occ:
	keys = ['origin_type', 'origin_url', 'branch', 'target',
	'target_type', 'path']
	return dict(zip(keys, found_occ))
	return None

	def directory_add(self, directories):
	"""Add directories to the storage

	Args:
	directories: iterable of dictionaries representing the individual
	directories to add. Each dict has the following keys:
	- id (sha1_git): the id of the directory to add
	- entries (list): list of dicts for each entry in the
	directory. Each dict has the following keys:
	- name (bytes)
	- type (one of 'file', 'dir', 'rev'):
	type of the directory entry (file, directory, revision)
	- target (sha1_git): id of the object pointed at by the
	directory entry
	- perms (int): entry permissions
	"""
	dirs = set()
	dir_entries = {
	'file': defaultdict(list),
	'dir': defaultdict(list),
	'rev': defaultdict(list),
	}

	for cur_dir in directories:
	dir_id = cur_dir['id']
	dirs.add(dir_id)
	for src_entry in cur_dir['entries']:
	entry = src_entry.copy()
	entry['dir_id'] = dir_id
	dir_entries[entry['type']][dir_id].append(entry)

	dirs_missing = set(self.directory_missing(dirs))
	if not dirs_missing:
	return

	db = self.db
	with db.transaction() as cur:
	# Copy directory ids
	dirs_missing_dict = ({'id': dir} for dir in dirs_missing)
	db.mktemp('directory', cur)
	db.copy_to(dirs_missing_dict, 'tmp_directory', ['id'], cur)

	# Copy entries
	for entry_type, entry_list in dir_entries.items():
	entries = itertools.chain.from_iterable(
	entries_for_dir
	for dir_id, entries_for_dir
	in entry_list.items()
	if dir_id in dirs_missing)

	db.mktemp_dir_entry(entry_type)

	db.copy_to(
	entries,
	'tmp_directory_entry_%s' % entry_type,
	['target', 'name', 'perms', 'dir_id'],
	cur,
	)

	# Do the final copy
	db.directory_add_from_temp(cur)

	@db_transaction_generator
	def directory_missing(self, directories, cur):
	"""List directories missing from storage

	Args: an iterable of directory ids
	Returns: a list of missing directory ids
	"""
	db = self.db

	# Create temporary table for metadata injection
	db.mktemp('directory', cur)

	directories_dicts = ({'id': dir} for dir in directories)

	db.copy_to(directories_dicts, 'tmp_directory', ['id'], cur)

	for obj in db.directory_missing_from_temp(cur):
	yield obj[0]

	@db_transaction_generator
	def directory_get(self,
	directories,
	cur=None):
	"""Get information on directories.

	Args:
	- directories: an iterable of directory ids

	Returns:
	List of directories as dict with keys and associated values.

	"""
	db = self.db
	keys = ('id', 'dir_entries', 'file_entries', 'rev_entries')

	db.mktemp('directory', cur)
	db.copy_to(({'id': dir_id} for dir_id in directories),
	'tmp_directory', ['id'], cur)

	dirs = db.directory_get_from_temp(cur)
	for line in dirs:
	yield dict(zip(keys, line))

	@db_transaction_generator
	def directory_ls(self, directory, recursive=False, cur=None):
	"""Get entries for one directory.

	Args:
	- directory: the directory to list entries from.
	- recursive: if flag on, this list recursively from this directory.

	Returns:
	List of entries for such directory.

	"""
	db = self.db
	keys = ['dir_id', 'type', 'target', 'name', 'perms', 'status',
	'sha1', 'sha1_git', 'sha256']

	if recursive:
	res_gen = db.directory_walk(directory)
	else:
	res_gen = db.directory_walk_one(directory)

	for line in res_gen:
	yield dict(zip(keys, line))

	@db_transaction
	def directory_entry_get_by_path(self, directory, paths, cur=None):
	"""Get the directory entry (either file or dir) from directory with
	path.

	Args:
	- directory: sha1 of the top level directory
	- paths: path to lookup from the top level directory. From left
	(top) to right (bottom).

	Returns:
	The corresponding directory entry if found, None otherwise.

	"""
	db = self.db
	keys = ('dir_id', 'type', 'target', 'name', 'perms', 'status',
	'sha1', 'sha1_git', 'sha256')

	res = db.directory_entry_get_by_path(directory, paths, cur)
	if res:
	return dict(zip(keys, res))

	def revision_add(self, revisions):
	"""Add revisions to the storage

	Args:
	revisions: iterable of dictionaries representing the individual
	revisions to add. Each dict has the following keys:
	- id (sha1_git): id of the revision to add
	- date (datetime.DateTime): date the revision was written
	- date_offset (int): offset from UTC in minutes the revision
	was written
	- date_neg_utc_offset (boolean): whether a null date_offset
	represents a negative UTC offset
	- committer_date (datetime.DateTime): date the revision got
	added to the origin
	- committer_date_offset (int): offset from UTC in minutes the
	revision was added to the origin
	- committer_date_neg_utc_offset (boolean): whether a null
	committer_date_offset represents a negative UTC offset
	- type (one of 'git', 'tar'): type of the revision added
	- directory (sha1_git): the directory the revision points at
	- message (bytes): the message associated with the revision
	- author_name (bytes): the name of the revision author
	- author_email (bytes): the email of the revision author
	- committer_name (bytes): the name of the revision committer
	- committer_email (bytes): the email of the revision committer
	- metadata (jsonb): extra information as dictionary
	- synthetic (bool): revision's nature (tarball, directory
	creates synthetic revision)
	- parents (list of sha1_git): the parents of this revision
	"""
	db = self.db

	revisions_missing = set(self.revision_missing(
	set(revision['id'] for revision in revisions)))

	if not revisions_missing:
	return

	with db.transaction() as cur:
	db.mktemp_revision(cur)

	revisions_filtered = (
	converters.revision_to_db(revision) for revision in revisions
	if revision['id'] in revisions_missing)

	parents_filtered = []

	db.copy_to(
	revisions_filtered, 'tmp_revision',
	['id', 'date', 'date_offset', 'date_neg_utc_offset',
	'committer_date', 'committer_date_offset',
	'committer_date_neg_utc_offset', 'type', 'directory',
	'message', 'author_name', 'author_email', 'committer_name',
	'committer_email', 'metadata', 'synthetic'],
	cur,
	lambda rev: parents_filtered.extend(rev['parents']))

	db.revision_add_from_temp(cur)

	db.copy_to(parents_filtered, 'revision_history',
	['id', 'parent_id', 'parent_rank'], cur)

	@db_transaction_generator
	- def revision_missing(self, revisions, type='git', cur=None):
	+ def revision_missing(self, revisions, cur=None):
	"""List revisions missing from storage

	- Args:
	- - revisions: an iterable of revision ids
	- - type: revision's type (git by default)
	+ Args: an iterable of revision ids

	Returns: a list of missing revision ids
	"""
	db = self.db

	- # Create temporary table for metadata injection
	- db.mktemp('revision', cur)
	-
	- revisions_dicts = ({'id': dir, 'type': type} for dir in revisions)
	-
	- db.copy_to(revisions_dicts, 'tmp_revision', ['id', 'type'], cur)
	+ db.store_tmp_bytea(revisions, cur)

	for obj in db.revision_missing_from_temp(cur):
	yield obj[0]

	@db_transaction_generator
	def revision_get(self, revisions, cur):
	"""Get all revisions from storage
	Args: an iterable of revision ids
	Returns: an iterable of revisions as dictionaries
	(or None if the revision doesn't exist)
	"""

	keys = ('id', 'date', 'date_offset', 'date_neg_utc_offset',
	'committer_date', 'committer_date_offset',
	'committer_date_neg_utc_offset', 'type', 'directory',
	'message', 'author_id', 'author_name', 'author_email',
	'committer_id', 'committer_name', 'committer_email',
	'metadata', 'synthetic', 'parents')

	db = self.db

	- # Create temporary table for metadata injection
	- db.mktemp('revision', cur)
	-
	- revisions_dicts = ({'id': rev, 'type': 'git'} for rev in revisions)
	-
	- db.copy_to(revisions_dicts, 'tmp_revision', ['id', 'type'], cur)
	+ db.store_tmp_bytea(revisions, cur)

	for line in self.db.revision_get_from_temp(cur):
	data = converters.db_to_revision(dict(zip(keys, line)))
	if not data['type']:
	yield None
	continue
	yield data

	@db_transaction_generator
	def revision_log(self, revisions, limit=None, cur=None):
	"""Fetch revision entry from the given root revisions.

	Args:
	- revisions: array of root revision to lookup
	- limit: limitation on the output result. Default to null.

	Yields:
	List of revision log from such revisions root.

	"""
	db = self.db

	keys = ['id', 'date', 'date_offset', 'date_neg_utc_offset',
	'committer_date', 'committer_date_offset',
	'committer_date_neg_utc_offset', 'type', 'directory',
	'message', 'author_id', 'author_name', 'author_email',
	'committer_id', 'committer_name', 'committer_email',
	'metadata', 'synthetic', 'parents']

	for line in db.revision_log(revisions, limit, cur):
	data = converters.db_to_revision(dict(zip(keys, line)))
	if not data['type']:
	yield None
	continue
	yield data

	@db_transaction_generator
	def revision_shortlog(self, revisions, limit=None, cur=None):
	"""Fetch the shortlog for the given revisions

	Args:
	revisions: list of root revisions to lookup
	limit: depth limitation for the output

	Yields:
	a list of (id, parents) tuples.
	"""

	db = self.db

	yield from db.revision_shortlog(revisions, limit, cur)

	@db_transaction_generator
	def revision_log_by(self, origin_id, limit=None, cur=None):
	"""Fetch revision entry from the actual origin_id's latest revision.

	"""
	db = self.db

	keys = ('id', 'date', 'date_offset', 'date_neg_utc_offset',
	'committer_date', 'committer_date_offset',
	'committer_date_neg_utc_offset', 'type', 'directory',
	'message', 'author_id', 'author_name', 'author_email',
	'committer_id', 'committer_name', 'committer_email',
	'metadata', 'synthetic', 'parents')

	for line in db.revision_log_by(origin_id, limit, cur):
	data = converters.db_to_revision(dict(zip(keys, line)))
	if not data['type']:
	yield None
	continue
	yield data

	def release_add(self, releases):
	"""Add releases to the storage

	Args:
	releases: iterable of dictionaries representing the individual
	releases to add. Each dict has the following keys:
	- id (sha1_git): id of the release to add
	- revision (sha1_git): id of the revision the release points
	to
	- date (datetime.DateTime): the date the release was made
	- date_offset (int): offset from UTC in minutes the release was
	made
	- date_neg_utc_offset (boolean): whether a null date_offset
	represents a negative UTC offset
	- name (bytes): the name of the release
	- comment (bytes): the comment associated with the release
	- author_name (bytes): the name of the release author
	- author_email (bytes): the email of the release author
	"""
	db = self.db

	release_ids = set(release['id'] for release in releases)
	releases_missing = set(self.release_missing(release_ids))

	if not releases_missing:
	return

	with db.transaction() as cur:
	db.mktemp_release(cur)

	releases_filtered = (
	converters.release_to_db(release) for release in releases
	if release['id'] in releases_missing
	)

	db.copy_to(releases_filtered, 'tmp_release',
	['id', 'target', 'target_type', 'date', 'date_offset',
	'date_neg_utc_offset', 'name', 'comment',
	'author_name', 'author_email', 'synthetic'],
	cur)

	db.release_add_from_temp(cur)

	@db_transaction_generator
	def release_missing(self, releases, cur=None):
	"""List releases missing from storage

	Args: an iterable of release ids
	Returns: a list of missing release ids
	"""
	db = self.db

	# Create temporary table for metadata injection
	db.store_tmp_bytea(releases, cur)

	for obj in db.release_missing_from_temp(cur):
	yield obj[0]

	@db_transaction_generator
	def release_get(self, releases, cur=None):
	"""Given a list of sha1, return the releases's information

	Args:
	releases: list of sha1s

	Returns:
	Generates the list of releases dict with the following keys:
	- id: origin's id
	- revision: origin's type
	- url: origin's url
	- lister: lister's uuid
	- project: project's uuid (FIXME, retrieve this information)

	Raises:
	ValueError if the keys does not match (url and type) nor id.

	"""
	db = self.db

	keys = ['id', 'target', 'target_type', 'date', 'date_offset',
	'date_neg_utc_offset', 'name', 'comment', 'synthetic',
	'author_id', 'author_name', 'author_email']

	# Create temporary table for metadata injection
	db.store_tmp_bytea(releases, cur)

	for release in db.release_get_from_temp(cur):
	yield converters.db_to_release(dict(zip(keys, release)))

	@db_transaction
	def occurrence_add(self, occurrences, cur=None):
	"""Add occurrences to the storage

	Args:
	occurrences: iterable of dictionaries representing the individual
	occurrences to add. Each dict has the following keys:
	- origin (int): id of the origin corresponding to the
	occurrence
	- branch (str): the reference name of the occurrence
	- target (sha1_git): the id of the object pointed to by
	the occurrence
	- target_type (str): the type of object pointed to by the
	occurrence
	- date (datetime.DateTime): the validity date for the given
	occurrence
	"""
	db = self.db

	processed = []
	for occurrence in occurrences:
	if isinstance(occurrence['date'], str):
	occurrence['date'] = dateutil.parser.parse(occurrence['date'])
	processed.append(occurrence)

	db.mktemp_occurrence_history(cur)
	db.copy_to(processed, 'tmp_occurrence_history',
	['origin', 'branch', 'target', 'target_type', 'date'], cur)

	db.occurrence_history_add_from_temp(cur)

	@db_transaction_generator
	def occurrence_get(self, origin_id, cur=None):
	"""Retrieve occurrence information per origin_id.

	Args:
	origin_id: The occurrence's origin.

	Yields:
	List of occurrences matching criterion.

	"""
	db = self.db
	for line in db.occurrence_get(origin_id, cur):
	yield {
	'origin': line[0],
	'branch': line[1],
	'target': line[2],
	'target_type': line[3],
	}

	@db_transaction_generator
	def revision_get_by(self,
	origin_id,
	branch_name=None,
	timestamp=None,
	limit=None,
	cur=None):
	"""Given an origin_id, retrieve occurrences' list per given criterions.

	Args:
	origin_id: The origin to filter on.
	branch_name: optional branch name.
	timestamp:
	limit:

	Yields:
	List of occurrences matching the criterions or None if nothing is
	found.

	"""
	keys = ('id', 'date', 'date_offset', 'date_neg_utc_offset',
	'committer_date', 'committer_date_offset',
	'committer_date_neg_utc_offset', 'type', 'directory',
	'message', 'author_id', 'author_name', 'author_email',
	'committer_id', 'committer_name', 'committer_email',
	'metadata', 'synthetic', 'parents')

	for line in self.db.revision_get_by(origin_id,
	branch_name,
	timestamp,
	limit=limit,
	cur=cur):
	data = converters.db_to_revision(dict(zip(keys, line)))
	if not data['type']:
	yield None
	continue
	yield data

	def release_get_by(self, origin_id, limit=None):
	"""Given an origin id, return all the tag objects pointing to heads of
	origin_id.

	Args:
	origin_id: the origin to filter on.
	limit: None by default

	Yields:
	List of releases matching the criterions or None if nothing is
	found.

	"""
	keys = ('id', 'target', 'target_type', 'date', 'date_offset',
	'date_neg_utc_offset', 'name', 'comment', 'synthetic',
	'author_id', 'author_name', 'author_email')

	for line in self.db.release_get_by(origin_id, limit=limit):
	data = converters.db_to_release(dict(zip(keys, line)))
	yield data

	@db_transaction
	def object_find_by_sha1_git(self, ids, cur=None):
	"""Return the objects found with the given ids.

	Args:
	ids: a generator of sha1_gits
	Returns:
	a dict mapping the id to the list of objects found. Each object
	found is itself a dict with keys:
	sha1_git: the input id
	type: the type of object found
	id: the id of the object found
	object_id: the numeric id of the object found.
	"""
	db = self.db

	ret = {id: [] for id in ids}

	for retval in db.object_find_by_sha1_git(ids):
	if retval[1]:
	ret[retval[0]].append(dict(zip(db.object_find_by_sha1_git_cols,
	retval)))

	return ret

	@db_transaction
	def origin_get(self, origin, cur=None):
	"""Return the origin either identified by its id or its tuple
	(type, url).

	Args:
	origin: dictionary representing the individual
	origin to find.
	This dict has either the keys type and url:
	- type (FIXME: enum TBD): the origin type ('git', 'wget', ...)
	- url (bytes): the url the origin points to
	either the id:
	- id: the origin id

	Returns:
	the origin dict with the keys:
	- id: origin's id
	- type: origin's type
	- url: origin's url
	- lister: lister's uuid
	- project: project's uuid (FIXME, retrieve this information)

	Raises:
	ValueError if the keys does not match (url and type) nor id.

	"""
	db = self.db

	keys = ['id', 'type', 'url', 'lister', 'project']

	origin_id = origin.get('id')
	if origin_id: # check lookup per id first
	ori = db.origin_get(origin_id, cur)
	elif 'type' in origin and 'url' in origin: # or lookup per type, url
	ori = db.origin_get_with(origin['type'], origin['url'], cur)
	else: # unsupported lookup
	raise ValueError('Origin must have either id or (type and url).')

	if ori:
	return dict(zip(keys, ori))
	return None

	@db_transaction
	def _person_add(self, person, cur=None):
	"""Add a person in storage.

	BEWARE: Internal function for now.
	Do not do anything fancy in case a person already exists.
	Please adapt code if more checks are needed.

	Args:
	person dictionary with keys name and email.

	Returns:
	Id of the new person.

	"""
	db = self.db

	return db.person_add(person['name'], person['email'])

	@db_transaction_generator
	def person_get(self, person, cur=None):
	"""Return the persons identified by their ids.

	Args:
	person: array of ids.

	Returns:
	The array of persons corresponding of the ids.

	"""
	db = self.db

	keys = ['id', 'name', 'email']

	for person in db.person_get(person):
	yield dict(zip(keys, person))

	@db_transaction
	def origin_add_one(self, origin, cur=None):
	"""Add origin to the storage

	Args:
	origin: dictionary representing the individual
	origin to add. This dict has the following keys:
	- type (FIXME: enum TBD): the origin type ('git', 'wget', ...)
	- url (bytes): the url the origin points to

	Returns:
	the id of the added origin, or of the identical one that already
	exists.

	"""
	db = self.db

	data = db.origin_get_with(origin['type'], origin['url'], cur)
	if data:
	return data[0]

	return db.origin_add(origin['type'], origin['url'], cur)

	@db_transaction
	def fetch_history_start(self, origin_id, cur=None):
	"""Add an entry for origin origin_id in fetch_history. Returns the id
	of the added fetch_history entry
	"""
	fetch_history = {
	'origin': origin_id,
	'date': datetime.datetime.now(tz=datetime.timezone.utc),
	}

	return self.db.create_fetch_history(fetch_history, cur)

	@db_transaction
	def fetch_history_end(self, fetch_history_id, data, cur=None):
	"""Close the fetch_history entry with id `fetch_history_id`, replacing
	its data with `data`.
	"""
	now = datetime.datetime.now(tz=datetime.timezone.utc)
	fetch_history = self.db.get_fetch_history(fetch_history_id, cur)

	if not fetch_history:
	raise ValueError('No fetch_history with id %d' % fetch_history_id)

	fetch_history['duration'] = now - fetch_history['date']

	fetch_history.update(data)

	self.db.update_fetch_history(fetch_history, cur)

	@db_transaction
	def fetch_history_get(self, fetch_history_id, cur=None):
	"""Get the fetch_history entry with id `fetch_history_id`.
	"""
	return self.db.get_fetch_history(fetch_history_id, cur)

	@db_transaction
	def entity_add(self, entities, cur=None):
	"""Add the given entitites to the database (in entity_history).

	Args:
	- entities: iterable of dictionaries containing the following keys:
	- uuid (uuid): id of the entity
	- parent (uuid): id of the parent entity
	- name (str): name of the entity
	- type (str): type of entity (one of 'organization',
	'group_of_entities', 'hosting', 'group_of_persons',
	'person', 'project')
	- description (str, optional): description of the entity
	- homepage (str): url of the entity's homepage
	- active (bool): whether the entity is active
	- generated (bool): whether the entity was generated
	- lister_metadata (dict): lister-specific entity metadata
	- metadata (dict): other metadata for the entity
	- validity (datetime.DateTime array): timestamps at which we
	listed the entity.
	"""
	db = self.db

	cols = list(db.entity_history_cols)
	cols.remove('id')

	db.mktemp_entity_history()
	db.copy_to(entities, 'tmp_entity_history', cols, cur)
	db.entity_history_add_from_temp()

	@db_transaction_generator
	def entity_get_from_lister_metadata(self, entities, cur=None):
	"""Fetch entities from the database, matching with the lister and
	associated metadata.

	Args:
	entities: iterable of dictionaries containing the lister metadata
	to look for. Useful keys are 'lister', 'type', 'id', ...
	Returns:
	A generator of fetched entities with all their attributes. If no
	match was found, the returned entity is None.
	"""

	db = self.db

	db.mktemp_entity_lister(cur)

	mapped_entities = []
	for i, entity in enumerate(entities):
	mapped_entity = {
	'id': i,
	'lister_metadata': entity,
	}
	mapped_entities.append(mapped_entity)

	db.copy_to(mapped_entities, 'tmp_entity_lister',
	['id', 'lister_metadata'], cur)

	cur.execute('''select id, %s
	from swh_entity_from_tmp_entity_lister()
	order by id''' %
	','.join(db.entity_cols))

	for id, *entity_vals in cur:
	fetched_entity = dict(zip(db.entity_cols, entity_vals))
	if fetched_entity['uuid']:
	yield fetched_entity
	else:
	yield {
	'uuid': None,
	'lister_metadata': entities[i],
	}

	@db_transaction_generator
	def entity_get(self, uuid, cur=None):
	"""Returns the list of entity per its uuid identifier and also its
	parent hierarchy.

	Args:
	uuid: entity's identifier

	Returns:
	List of entities starting with entity with uuid and the parent
	hierarchy from such entity.

	"""
	db = self.db
	for entity in db.entity_get(uuid, cur):
	yield dict(zip(db.entity_cols, entity))

	@db_transaction
	def entity_get_one(self, uuid, cur=None):
	"""Returns one entity using its uuid identifier.

	Args:
	uuid: entity's identifier

	Returns:
	the object corresponding to the given entity

	"""
	db = self.db
	entity = db.entity_get_one(uuid, cur)
	if entity:
	return dict(zip(db.entity_cols, entity))
	else:
	return None

	@db_transaction
	def stat_counters(self, cur=None):
	"""compute statistics about the number of tuples in various tables

	Returns:
	a dictionary mapping textual labels (e.g., content) to integer
	values (e.g., the number of tuples in table content)

	"""
	return {k: v for (k, v) in self.db.stat_counters()}

File Metadata

Mime Type: text/x-diff
Expires: Sat, Jun 21, 5:39 PM (1 w, 6 d ago)
Storage Engine: blob
Storage Format: Raw Data
Storage Handle: 3251575

No OneTemporaryActions

View Options

File Metadata

Event Timeline

No OneTemporary
Actions