adlio / schema Goto Github PK

Hi!

I was working on adding SQLite support. It is all pretty straightforward except for locking. While I can work to add a lock scheme, it's necessarily going to involve polling to try to write the sentinel value, checking I got the lock, handling stale locks, etc. My thought: is it worth it for SQLite? Given the use cases of SQLite, that there can only be one writer, and the migrations are done in a transaction, I think anything beyond a no-op ("") lock/unlock is of questionable value. What do you think?

(v1.1.7) file.go uses "path" instead of "path/filepath" for a few path-related operations.

Because Windows uses a backslash for the path separator, the file path is not properly interpreted. This results in the migration ID to include the entire absolute path instead of just the filename.

Two functions are involved: MigrationIDFromFilename and MigrationsFromDirectoryPath.

Changing the following appears to correct the problem (although only tested under Windows):

path.Base() to filepath.Base()
path.Ext() to filepath.Ext()
path.Join() to filepath.Join()

Commit a7bc68f appears to have changed the methods of the Migrator type to accept a pointer rather than a value. The NewMigrator constructor, however, still returns a Migrator by value, which means that a call like:

err := schema.NewMigrator().Apply(x, y)

(which used to work) will now fail because the Migrator is in a temporary and the compiler will not desugar the address.

For now I can work around this by rewriting expressions like the above as:

m := schema.NewMigrator()
err := m.Apply(x, y)

but I think the value returned by the constructor should have the correct type for use by the methods.