Merge pull request #8206 from denschub/lowercase-tags

Enforce lowercase tags and migrate existing uppercase tags for Postgres.
This commit is contained in:
Dennis Schubert 2021-02-13 04:43:03 +01:00
commit ecd2b87475
No known key found for this signature in database
GPG key ID: 5A0304BEA7966D7E
2 changed files with 82 additions and 0 deletions

View file

@ -31,6 +31,7 @@ Although the chat was never enabled per default and was marked as experimental,
* Add client-side cropping of profile image uploads [#7581](https://github.com/diaspora/diaspora/pull/7581)
* Add client-site rescaling of post images if they exceed the maximum possible size [#7734](https://github.com/diaspora/diaspora/pull/7734)
* Add backend for archive import [#7660](https://github.com/diaspora/diaspora/pull/7660)
* For pods running PostgreSQL, make sure that no upper-case/mixed-case tags exist, and create a `lower(name)` index on tags to speed up ActsAsTaggableOn [#8206](https://github.com/diaspora/diaspora/pull/8206)
# 0.7.15.0

View file

@ -0,0 +1,81 @@
# frozen_string_literal: true
class LowercaseifyTags < ActiveRecord::Migration[5.2]
# We have to disable the global transaction for migrations, otherwise we
# can't rescue from duplicates below. Checking if there are duplicates before
# running the update would make this migration even more complicated, so...
disable_ddl_transaction!
# rubocop:disable Metrics/AbcSize, Metrics/MethodLength
def up
# MySQL is case-insensitive by default. However, using utf8mb4_bin, it is
# not. So, in theory, MySQL is affected by the same issue. Unfortunately,
# MySQL does not support function indexes, so we can't simply run an index
# on lower(name). The "solution" for MySQL is to add a generated column
# and triggers to update that column.
#
# Since MySQL is deprecated from our POV anyway, and no new pods should
# run MySQL - let's ignore this for now.
return unless AppConfig.postgres?
# Get all tag names for tags that are currently not completely lowercase
affected_tags = query_first_field("SELECT DISTINCT lower(name) FROM tags WHERE lower(name) != name")
Rails.logger.info "Merging duplicate tag entities for #{affected_tags.length} tags..."
affected_tags.each do |tag_name|
duplicate_tag_ids = query_first_field("SELECT id FROM tags WHERE lower(name) = '#{tag_name}' ORDER BY id ASC")
# Take the lowest ID tag as the target
target_tag_id = duplicate_tag_ids.shift
duplicate_tag_ids.each do |duplicate_tag|
# Note: Because things could be tagged with multiple cases of the same
# tag, we can't simply update all taggings - it might create a
# duplicate tagging violation...
# Instead, query each tagging, try to update, and if it's a duplicate,
# simply drop the tagging.
# Do the aforementioned for taggings (posts, profiles, ...)
query_first_field("SELECT id FROM taggings WHERE tag_id = #{duplicate_tag}").each do |tagging_id|
begin
execute("UPDATE taggings SET tag_id = #{target_tag_id} WHERE id = #{tagging_id}")
rescue ActiveRecord::RecordNotUnique
execute("DELETE FROM taggings WHERE id = #{tagging_id}")
end
end
# Do the aforementioned for tag_followings (i.e. tag streams)
query_first_field("SELECT id FROM tag_followings WHERE tag_id = #{duplicate_tag}").each do |tag_following_id|
begin
execute("UPDATE tag_followings SET tag_id = #{target_tag_id} WHERE id = #{tag_following_id}")
rescue ActiveRecord::RecordNotUnique
execute("DELETE FROM tag_followings WHERE id = #{tag_following_id}")
end
end
# Delete the duplicate tag
execute("DELETE FROM tags WHERE id = #{duplicate_tag}")
end
# Make sure the target tag is actually lowercase
execute("UPDATE tags SET name = '#{tag_name}' WHERE id = #{target_tag_id}")
# Re-count the taggings. I don't think this matters for us, but...
ActsAsTaggableOn::Tag.reset_counters(target_tag_id, :taggings)
end
# Finally, create the new, now unique, index on the lowercase'd tag name
add_index :tags, "lower(name)", name: "index_tags_on_lower_name", unique: true
end
# rubocop:enable Metrics/AbcSize, Metrics/MethodLength
def down
remove_index :tags, name: "index_tags_on_lower_name"
end
private
def query_first_field(query)
execute(query).values.map {|row| row[0] }
end
end