Support index.length for MySQL 8.0.0-dmr

[yahonda]

Summary

This pull request addresses #26774 since MySQL 8.0.0-dmr SUB_PART column of information_schema.statistics changed to varbinary(12), which is bigint(3) in MySQL 5.6.

This pull request has been tested with both versions of MySQL.

Server version: 8.0.0-dmr MySQL Community Server (GPL)
Server version: 5.6.33 MySQL Community Server (GPL)

[rails-bot]

r? @kaspth

(@rails-bot has picked a reviewer for you, use r? to override)

[rafaelfranca]

Backported in 4cf4c86

[matthewd]

The backport broke the build: https://travis-ci.org/rails/rails/jobs/168139524

(I haven't looked into why master doesn't have the same problem.)

[rafaelfranca]

💣 I'll revert the backport :(. @yahonda could you investigate?

[kamipo]

If row[:Sub_part] is nil, row[:Sub_part].to_i is 0.

[kamipo]

Originally (index.lengths || []).compact have a bug.
https://github.com/rails/rails/blob/5-0-stable/activerecord/lib/active_record/schema_dumper.rb#L216
If #26785 will backport, it is better to backport #26745 as well.

[yahonda]

let me check.

[yahonda]

I have attempted to backport #26745 to 5-0-stable by cherry-pick 5025fd3 then it conflicts with #26155 commit a57cd38. It also conflicts with rubocop format 55f9b81

Then simply add another if condition to add length only if row[:Sub_part] is not nil. Let me open another pull request to 5-0-stable.

[yahonda]

Opened #26806 for 5-0-stable branch.

[yahonda]

Closed #26806 which would cause another regression. So far I'm leaving 5-0-stable current implementation as they are. Let me tackle it when I have some more time to backport commits described in #26785 (comment) without external spec change and conflicts.