Optimize DDL for CHAR(10) to VARCHAR(100)

[dveeden]

Feature Request

Changing a column from a CHAR(n) to a VARCHAR(m) is now going through a full reorg. However the CHAR is guaranteed to fit in the VARCHAR as long as the charset remains the same and if m>=n.

Here the source is a CHAR and the target is a VARCHAR, but this is likely also true for any other combination of these types and might also be true for VARBINARY etc.

The improvement would be to do this as a metadata only change and only change the actual rows when they are written again.

[mjonss]

It looks like the secondary indexes have different format for char and varchar (according to comment here).
We could still do this as metadata change if the column is not included in any index.

[bb7133]

Thanks, @mjonss

Even for columns without a secondary index, we cannot convert VARCHAR to CHAR without data reorganization because for TiDB, the data in CHAR is potentially truncated by removing all trailing spaces.

But for CHAR to VARCHAR without any index, I think this can be optimized. /cc @zimulala @Benjamin2037

[yiwen92]

It looks like the secondary indexes have different format for char and varchar (according to comment here). We could still do this as metadata change if the column is not included in any index.

What is the exactly difference between char and varchar for index encoding?
// 2. char -> varchar: the index value encoding of secondary index on clustered primary key tables is different. // These secondary indexes need to be rewritten.