persistent-mysql: truncate foreign key constraint names to MySQL's 64 char limit

[Moferanoluwa]

Description

Fixes #1000. persistent-mysql currently generates foreign key constraint names using the shared default scheme (tableName_columnName_fkey) with no length limit. Unlike Postgres, which silently truncates over-length identifiers (worked around in persistent-postgresql via refName, added in #996), MySQL just raises an error for identifiers over 64 characters.

This adds the MySQL equivalent of that same refName override, reusing the exact same truncation algorithm persistent-postgresql already uses (approximating how the database itself truncates), just with MySQL's 64-character limit instead of Postgres' 63, wired in via the existing setBackendSpecificForeignKeyName extension point.

Evidence this is the intended fix

persistent-test/src/LongIdentifierTest.hs and its hookup in persistent-mysql/test/main.hs were already sitting there, explicitly disabled and waiting on exactly this:

-- TODO: implement automatic truncation for too long foreign keys, so we can run this test.
xdescribe
    "The migration for this test currently fails because of MySQL's 64 character limit for identifiers. See https://github.com/yesodweb/persistent/issues/1000 for details"
    $ LongIdentifierTest.specsWith db

I enabled that test (both the xdescribe wrapper and the commented-out migration registration a few lines up) rather than leaving it disabled.

Changes

• persistent-mysql/Database/Persist/MySQL.hs — add refName, mirroring persistent-postgresql's function of the same name/signature almost exactly (maximumIdentifierLength = 64 instead of 63), and wire it into mysqlMkColumns via setBackendSpecificForeignKeyName.
• persistent-mysql/test/main.hs — enable LongIdentifierTest, which was previously disabled specifically because this was missing.
• persistent-test/src/LongIdentifierTest.hs — update the comment that said MySQL didn't run this test yet.
• persistent-mysql/persistent-mysql.cabal — bump patch version (2.13.1.6 → 2.13.1.7) per CONTRIBUTING.md's versioning guide (bug fix = patch bump).

Per CONTRIBUTING.md, I'll follow up with the Changelog.md entry once I have a PR number to link.

Testing/Review Recommendations

I don't have a Haskell toolchain (no ghc/stack/cabal) available in this environment, so I was not able to compile or run this locally — please treat that as a real gap, not just a formality, and lean on CI here. What gives me confidence despite that:

• refName's type signature and every symbol it uses (EntityNameDB, FieldNameDB, ConstraintNameDB, setBackendSpecificForeignKeyName, emptyBackendSpecificOverrides) come from Database.Persist.Sql, which MySQL.hs already imports unqualified — same import the Postgres file uses for the same symbols.
• The function body is the Postgres implementation with only the constant changed (63 → 64); I traced the shortenNames recursion by hand rather than just assuming it carries over.
• The now-enabled LongIdentifierTest is exactly the test the maintainers left in place for this fix (it references issue MySQL backend generates foreign keys over length limit #1000 by number in its own skip message), so CI running it is a real, existing check of correctness, not something I added ad hoc.

AI Usage

• Tool: Claude Code (Claude Sonnet)
• Type: Locating the existing Postgres refName implementation and the pre-wired-but-disabled LongIdentifierTest, porting the former to MySQL, and enabling the latter.
• Scope: The four files listed above.