*: support AST to SQL text

[zier-one]

1、add Recoverable interface
2、add Recoverable implement example(CreateDatabaseStmt,DropDatabaseStmt,UnaryOperationExpr)
3、add stmt Recoverable test frame
4、replace the tidb dependency to github.com/leoppro/tidb temporarily,
github.com/leoppro/tidb forked from pingcap/tidb and impl the Recoverable interface for ValueExpr and ParamMarkerExpr, when it is stable,i will pull it to pingcap/tidb

[zier-one]

PTAL @GregoryIan @tiancaiamao

[XuHuaiyu]

It'll be nice if you can create an issue in the tidb repo to trace the process of this task.
We can list the sub-tasks in the issue, thus the community can choose their interested tasks to participate.

This can be referred. pingcap/tidb#7623

[zier-one]

@kennytm PTAL again

[kennytm]

Rest LGTM

[tiancaiamao]

LGTM
Don't forget to address comment from kenny @leoppro

[kennytm]

LGTM

[zier-one]

@tiancaiamao PTAL again, I moved IsReadOnly into util.go https://github.com/pingcap/parser/pull/56/files#diff-ed076aa6b0e8db7b5db0c844dcc1cb33R21

[XuHuaiyu]

PTAL @tiancaiamao @zz-jason

needs one more LGTM

[zz-jason]

If there are still some comments not addressed and you think they should be addressed, please do not comment a "LGTM" @tiancaiamao

How about using the visitor model? Implement a visitor visits the AST tree, complement the sql during the traverse? Thus we don't need to introduce a Recoverable interface and change the AST node implementations. @leoppro

[zier-one]

How about using the visitor model? Implement a visitor visits the AST tree, complement the sql during the traverse? Thus we don't need to introduce a Recoverable interface and change the AST node implementations. @leoppro

I have considered this way, it's hard to implement.
Firstly, we should use type assertion to handle different ast.Node, it will be complicated;
Secondly, we can‘t control visiting order for child node.

[tiancaiamao]

I think Restore interface is necessary to implement AST to SQL
Maybe we can add a method to Node, rather than define a Restorable interface.

[zz-jason]

How about using the visitor model? Implement a visitor visits the AST tree, complement the sql during the traverse? Thus we don't need to introduce a Recoverable interface and change the AST node implementations. @leoppro

I have considered this way, it's hard to implement.
Firstly, we should use type assertion to handle different ast.Node, it will be complicated;
Secondly, we can‘t control visiting order for child node.

Got it!

[zier-one]

I think Restore interface if necessary to implement AST to SQL
Maybe we can add a method to Node, rather than define a Restorable interface.

some struct which isn't a Node need implement Restorable too, such as DatabaseOption

[tiancaiamao]

type Node interface {
    Recoverable
    ...
}

Using embed interface , you still need to implement Restore method for DababaseOption @leoppro

So what's the difference if you define Node like this ?

type Node interface {
    Recover(builder) error
    ...
}

[tiancaiamao]

And if you define the Recover() method in the node struct (the conceptual base class), it would save you many codes to define

func (xx XXNode) Recover() error {
   return errors.New("not implemented")
}

@leoppro

[XuHuaiyu]

LGTM

[XuHuaiyu]

PTAL @zz-jason @tiancaiamao

[zz-jason]

LGTM

[kennytm]

(rest LGTM)