Skip to content

Commit

Permalink
bundle: merge cleanup etc
Browse files Browse the repository at this point in the history
  • Loading branch information
@erichanson
erichanson committed Jan 20, 2021
1 parent 5b6d678 commit 0dadab3
Showing 1 changed file with 64 additions and 41 deletions.
105 changes: 64 additions & 41 deletions extensions/bundle/001-functions.sql
View file
Original file line number Diff line number Diff line change
Expand Up @@ -411,15 +411,8 @@ $$ language sql;


------------------------------------------------------------------------------
-- CHECKOUT FUNCTIONS
-- user stories:
--
-- 1. user downloads a new bundle, checking out where everything is fresh and
-- new. we don't run into any collisions and just plop it all into place.
--
-- 2. user tries to check out a bundle when his working copy is different from
-- previous commit. this would be indicated by rows in offstage_row_deleted and
-- offstage_field_change, or stage_row_*.
-- CHECKOUT FUNCTIONS
--
------------------------------------------------------------------------------

Expand All @@ -430,17 +423,15 @@ create or replace function checkout_row (in row_id meta.row_id, in fields checko
begin
-- raise log '------------ checkout_row % ----------',
-- (row_id::meta.schema_id).name || '.' || (row_id::meta.relation_id).name ;
-- set search_path=bundle,meta,public;
set local search_path=something_that_must_not_be;

if meta.row_exists(row_id) then
-- raise log '---------------------- row % already exists.... overwriting.',
-- (row_id::meta.schema_id).name || '.' || (row_id::meta.relation_id).name ;

-- check to see if this row which is being merged is going to overwrite a row that is
-- TODO: check to see if this row which is being merged is going to overwrite a row that is
-- different from the head commit


-- overwrite existing values with new values.
/*
execute 'update ' || quote_ident((row_id::meta.schema_id).name) || '.' || quote_ident((row_id::meta.relation_id).name)
Expand Down Expand Up @@ -531,12 +522,6 @@ create or replace function checkout_row (in row_id meta.row_id, in fields checko
-- raise log 'query_str: %', query_str;

execute query_str;

/*
(select string_agg (quote_ident((f::bundle.checkout_field).name), ',') from unnest(fields) as f) || ')'
|| ' values '
|| ' (' || (select string_agg (quote_literal(f.value) || '::' || (f::bundle.checkout_field).type_name, ',') from unnest(fields) as f) || ')';
*/
end if;
end;

Expand Down Expand Up @@ -576,7 +561,6 @@ create or replace function checkout (in commit_id uuid, in comment text default
end if;

raise notice 'bundle.checkout(): % / % @ % by %: "%"', bundle_name, commit_id, commit_time, commit_role, commit_message;
-- raise notice 'bundle: Checking out bundle %', commit_id;
-- insert the meta-rows in this commit to the database
for commit_row in
select
Expand Down Expand Up @@ -625,8 +609,11 @@ create or replace function checkout (in commit_id uuid, in comment text default



-- raise notice '################################################## DISABLING TRIGGERS % ###############################', commit_id;
-- raise notice '### DISABLING TRIGGERS % ###', commit_id;
-- turn off constraints
--
-- TODO: stop doing this. row keys must be analyzed so that they can be inserted in sensible order.
-- in the case of circular dependencies we might still need to briefly and microscopically disable them.
for commit_row in
select distinct
(rr.row_id).pk_column_id.relation_id.name as relation_name,
Expand Down Expand Up @@ -678,7 +665,7 @@ create or replace function checkout (in commit_id uuid, in comment text default


-- turn constraints back on
-- raise notice '################################################## ENABLING TRIGGERS % ###############################', commit_id;
-- raise notice '### ENABLING TRIGGERS % ###', commit_id;
for commit_row in
select distinct
(rr.row_id).pk_column_id.relation_id.name as relation_name,
Expand Down Expand Up @@ -770,8 +757,12 @@ create or replace function commits_common_ancestor(commit1_id uuid, commit2_id u
same_branch_2 integer;
ancestor uuid;
begin
select count(*) from unnest(bundle.commit_ancestry(commit1_id)) c(id) where id = commit2_id into same_branch_1;
select count(*) from unnest(bundle.commit_ancestry(commit2_id)) c(id) where id = commit1_id into same_branch_2;
if commit1_id = commit2_id then return null; end if;

select count(*) from unnest(bundle.commit_ancestry(commit1_id)) c(id)
where id = commit2_id into same_branch_1;
select count(*) from unnest(bundle.commit_ancestry(commit2_id)) c(id)
where id = commit1_id into same_branch_2;

if same_branch_1 > 0 or same_branch_2 > 0 or same_branch_1 is null or same_branch_2 is null then
-- raise notice 'Commits are on the same branch.';
Expand Down Expand Up @@ -827,6 +818,8 @@ create or replace function rows_created_between_commits( new_commit_id uuid, anc
$$ language sql;


-- checks to see if it structurally makes sense that this commit be merged.
-- does not check for changes in the working copy.
create or replace function commit_is_mergable(merge_commit_id uuid) returns boolean as $$
declare
_bundle_id uuid;
Expand All @@ -848,29 +841,37 @@ create or replace function commit_is_mergable(merge_commit_id uuid) returns bool
return false;
end if;

/*
-- assert that the working copy does not contain uncommitted changes (TODO: allow this?)
select count(*) from bundle.head_db_stage_changed where bundle_id=_bundle_id into changes_count;
if changes_count > 0 then
raise exception 'Merge not permitted when there are uncommitted changes';
end if;
*/

-- assert that the two commits share a common ancestor
select bundle.commits_common_ancestor(head_commit_id, merge_commit_id) into common_ancestor_id;
if common_ancestor_id is null then
-- raise notice 'Head commit and merge commit do not share a common ancestor.';
return false;
end if;

-- assert that this commit is not already merged
-- TODO


return true;
end;
$$ language plpgsql;



-- merge
-- the big kahuna.
/*
Merge
- assert that this commit is mergable
- head_commit_id equals checkout_commit_id (no detached head)
- working copy does not contain any changes
- the two commits share a common ancestor
- set bundle in merge mode: bundle.merge_commit_id is not null and references the commit being merged
- update and stage non-conflicting field changes
- update but do not stage conflicting field changes
- insert and stage new row creates
- delete and stage new row deletes
*/


create or replace function merge(merge_commit_id uuid) returns void as $$
declare
Expand Down Expand Up @@ -899,7 +900,7 @@ create or replace function merge(merge_commit_id uuid) returns void as $$
-- assert that the working copy does not contain uncommitted changes (TODO: allow this?)
select count(*) from bundle.head_db_stage_changed where bundle_id=_bundle_id into changes_count;
if changes_count > 0 then
raise exception 'Merge not permitted when there are uncommitted changes';
raise exception 'Merge not permitted when this bundle has uncommitted changes';
end if;

-- assert that the two commits share a common ancestor
Expand All @@ -908,6 +909,9 @@ create or replace function merge(merge_commit_id uuid) returns void as $$
raise exception 'Head commit and merge commit do not share a common ancestor.';
end if;

-- assert that this commit is not already merged
-- TODO

raise notice 'Merging commit % into head (%), with common ancestor commit %', merge_commit_id, head_commit_id, common_ancestor_id;

/*
Expand Down Expand Up @@ -1003,24 +1007,43 @@ create or replace function merge(merge_commit_id uuid) returns void as $$
);
-- raise notice 'STMT: %', update_stmt;
execute update_stmt;
end loop;

/*
if conflicted_merge then
update bundle set merge_commit_id = merge_commit_id where id=_bundle_id;
else
select bundle.commit();
end if;
*/
-- insert record of the conflicting field into bundle.merge_conflict
update_stmt := format('
insert into bundle.merge_conflict
( bundle_id, rowset_row_field_id, field_id, conflict_value )
select _bundle_id, rrf.id, rrf.field_id, b.value
from bundle.commit c
join bundle.rowset r on c.rowset_id = r.id
join bundle.rowset_row rr on rr.rowset_id = r.id
join bundle.rowset_row_field rrf on rrf.rowset_row_id = rr.id
join bundle.blob b on rrf.value_hash = b.hash
where c.id = %L
and rrf.field_id::text = %L',
merge_commit_id, f.field_id::text,
);
-- raise notice 'STMT: %', update_stmt;
execute update_stmt;

end loop;

/*
TODO:
- row deletes
*/

-- set merge state
update bundle.bundle set merge_commit_id = merge_commit_id where id=_bundle_id;

end;
$$ language plpgsql;

create or replace function merge_finish(_bundle_id uuid) returns void as $$
begin
end;
$$ language plpgsql;


create or replace function merge_cancel(_bundle_id uuid) returns void as $$
begin
-- assert that bundle is in merge mode
Expand Down

0 comments on commit 0dadab3

Please sign in to comment.