Introduce `TypelessStructLiteralArg` and `NewConstructorArg` interfaces #183

makenowjust · 2024-10-27T14:33:38Z

Close #119
Close #177

TypelessStructValue (#177) and NewConstructorValue (#119) share their struct shapes. Then, this PR introduces a new struct named AsExpr for expression AS name-kind values and replaces them with it.

I believe this is not a hasty generalization because these struct shapes are exactly the same, and they are used for common semantics. However, I'm not sure AsExpr is the best name. If you have a better name, please tell me it!

~~Thank you.~~

See #183 (comment).

apstndb · 2024-10-27T14:54:54Z

TypelessStructValue (#177) and NewConstructorValue (#119) share their struct shapes. Then, this PR introduces a new struct named AsExpr for expression AS name-kind values and replaces them with it.

I believe this is not a hasty generalization because these struct shapes are exactly the same, and they are used for common semantics.

I have concerned about there are undocumented differences in NEW constructors and typeless STRUCT literals.

https://github.com/google/zetasql/blob/194cd32b5d766d60e3ca442651d792c7fe54ea74/zetasql/parser/bison_parser.y#L8927-L8942

    | expression "AS" "(" path_expression ")"
      {
        // Do not parenthesize $4 because it is not really a parenthesized
        // path expression. The parentheses are just part of the syntax here.
        $$ = MAKE_NODE(ASTNewConstructorArg, @$, {$1, $4});
      }

I believe it is a syntax for proto extensions and it is not yet available in Cloud Spanner, but I think these semantics are not same. It may be a reason to distinguish between TypelessStructValue and NewConstructorValue .

spanner> SELECT NEW examples.shipping.`Order`("Japan" AS (shipping_address.country));
ERROR: spanner: code = "InvalidArgument", desc = "NEW constructor does not support proto extensions [at 1:50]\\nSELECT NEW examples.shipping.`Order`(\\\"Japan\\\" AS (shipping_address.country))\\n

apstndb · 2024-10-27T15:34:56Z

This difference is natural because:

NewConstructor is a syntax for PROTO type.
TypelessStructLiteral is a syntax for STRUCT type.

makenowjust · 2024-10-27T16:35:05Z

IMHO, when expression AS (path) is needed, we can introduce a new interface NewConstructorArg, and a new struct ExprAsPath, so we can go this way even if STRUCT and NEW syntax are different.

apstndb · 2024-10-27T17:22:16Z

Personally I think it is acceptable to generalize them temporarily, even if their semantics are not same,
But I am wondering if this generalization really have benefit.

we can introduce a new interface NewConstructorArg, and a new struct ExprAsPath

I think it can't be a general name because it is expression AS (path_expression), not expression AS path_expression.
It seems that (path_expression) is restricted form of generalized_extension_path.
In ZetaSQL, new_constructor_arg is really special structure with no other similar syntax, because its syntax is specialized to Protocol Buffers.

apstndb · 2024-10-27T17:46:01Z

I understood. We can add another interface when it is needed.
NewConstructorArg won't be concrete struct so it is not needed to define.

apstndb · 2024-10-27T18:01:54Z

However, I'm not sure AsExpr is the best name. If you have a better name, please tell me it!
Rename AsExpr to ExprAsName, and use AsAlias

IMHO, ExprAsName doesn't imply the name is optional, and there are other expr AS name structure(e.g. in * REPLACE(expr AS name, ...)).
I think it can be confusing so more explicit name, like ExprOptName, may be prefered.

syntax	AS	ident	Rule in ZetaSQL	ast node in ZetaSQL	Parser func in memefish	Name in memefish
AS ident	required		opt_as_alias_with_required_as	ASTAlias	tryParseAsAlias()	AsAlias
[AS] ident	optional		opt_as_alias	ASTAlias	tryParseAsAlias()	AsAlias
expr	none	none		ASTExpression	ExprArg or Expr
expr AS ident	required	required	(star_replace_item, select_column_expr_with_as_alias, ...)	ASTExpressionWithAlias (not used externally?)
expr [AS ident]	required	optional	expression_with_opt_alias, (expression_using_argument, ...)	ASTExpressionWithOptAlias		ExprAsName?
expr [AS] ident	optional	required				Alias
expr [[AS] ident]	optional	optional	(select_column_expr, ...)	concrete nodes

(There are many variants.)

parser.go

apstndb · 2024-10-28T03:20:44Z

ast/ast.go

+	Values []*ExprAsName
+}
+
+// ExprAsName is value with optional name in typeless struct literal or new constructor.


I think it is not only for typeless struct literal and new constructor.
We can encourage to use this struct for any expr [AS ident] syntax, without semantic implications.

We can update the document anytime.

It means that this struct is added only for NewConstructorArg and TypelessStructValue in this PR context and we need to discuss to change this struct semantics in another PR.

From this conversation, it sounds like you don't intend to widely recommend the use of this struct.

My thought:

TypelessStructValue and NewConstructorArg is not same abstraction, but have common concrete structure.

Concrete structure will able to be reused without modification.

IMO, currently ExprAsName is no strong reason to be introduced.

makenowjust · 2024-10-28T12:12:51Z

Finally, I conclude there is no better name for the struct expression [AS name]. That is, we should not introduce such a struct.

Therefore, we introduce new interfaces TypelessStructLiteralArg and NewConstructorArg. Currently, two interfaces have the same implementations and parsers, but they are separated for future extensibility.

apstndb

I feel it is natural to introduce these interfaces and use ExprArg and Alias.
I have not noticed their SelectItem implementation can be used to implement them.

I agree their implementations are better than my original PRs.

apstndb · 2024-10-28T13:50:20Z

parser.go

@@ -479,22 +479,26 @@ func (p *Parser) parseSelectItem() ast.SelectItem {
 	}
 }

-func (p *Parser) tryParseAsAlias() *ast.AsAlias {
+func (p *Parser) tryParseAsAlias(requiredAs bool) *ast.AsAlias {


I feel like p.tryParseAsAlias( /* requiredAs */ false) is a little difficult to write.
Can't we introduce two method for requiredAs=true and requiredAs=False? (like (tryParseAsAliasAsRequired and tryParseAsAliasAsOptional)

tryParseAsAliasAsRequired seems like a strange name to someone unfamiliar with this project (e.g. twice As looks like typo, and try and required seem contradictory), and it is hard to guess what the method does. By making the explicit argument, it is clear that the method parses AS name as possible, and the comment makes it clear that the argument gives the required AS.

My thought:

twice As looks like typo

I think that the current AsAlias([AS] alias) may be better to be called as Alias, and the current Alias(expr [AS] alias, alias required) can be called as ExprWithAlias or AliasedExpr.
tryParseAliasAsRequired and tryParseAliasAsOptional are felt better.

the comment makes it clear that the argument gives the required AS.

I think maintaining /* requiredAs */ comment is harder because it can't be assisted by ~~go-lsp~~ gopls or IDEs, and it is also strange to someone(me) unfamiliar with this project. At least, it must be noted in doc comment of tryParseAsAlias().

tryParseAliasAsRequired also seems strange to me because As looks like a usual preposition, and try and required have conflicting meanings.

I think that Alias is the identifier after AS, not the AS name itself. In that sense, I think that the current Alias should be changed to AliasedExpr, but I do not agree with naming AsAlias as Alias.

tryParseAliasAsRequired also seems strange to me because As looks like a usual preposition, and try and required have conflicting meanings.

It means they can be better if the subject of "required" or "optional" is unambiguous.
The subject is AS keyword, so they can be called as like tryParseAsAliasRequireKeyword, tryParseAsAliasOptionalKeyword.

It's ambiguous enough that there are multiple methods to parse a particular struct, and I don't think it's a good idea to distinguish them by method names. If it really should be parsed by multiple methods, we should introduce multiple structs, but I don't think in this case.

It's ambiguous enough that there are multiple methods to parse a particular struct

I agree structs and methods should be one by one.
In other languages, it may be a usecase of named parameter which is missing in Go.
In Go, it may be better to introduce typed constant with enum semantics.

Okay, because of code jumps and editor popups, I decided to remove the /* requiredAs */ comment.

We have agreed about tryParseAsAlias(), so it is just my thought.

I think that Alias is the identifier after AS, not the AS name itself. In that sense, I think that the current Alias should be changed to AliasedExpr, but I do not agree with naming AsAlias as Alias.

SQL has common structure "alias, which is optionally prefixed by AS", and they are safe to unparse always with AS in GoogleSQL.
I prefer to call it Alias than AsAlias because someones have coding guideline which prefer to omit AS.
As far as I know, AS of table alias is often omitted. (Probably due to Oracle's past not allowing AS.)

The name Alias has less information than the name AsAlias, so I don't think it should be changed proactively.

apstndb

I believe we have finished discussions and tryParseAsAlias with typed enum resolves our concernings.
Thank you for finishing my PRs.
LGTM!

…TypedStructLiteral

Also, this fixes a `AsAlias.As` position bug.

Co-authored-by: apstndb <[email protected]>

apstndb reviewed Oct 27, 2024

View reviewed changes

parser.go Outdated Show resolved Hide resolved

apstndb reviewed Oct 28, 2024

View reviewed changes

makenowjust changed the title ~~Introduce AsExpr instead of TypelessStructValue and NewConstructorValue~~ Introduce TypelessStructLiteralArg and NewConstructorArg interfaces Oct 28, 2024

apstndb reviewed Oct 28, 2024

View reviewed changes

apstndb approved these changes Oct 29, 2024

View reviewed changes

apstndb and others added 18 commits October 29, 2024 21:57

Implement BracedNewConstructor

4b7c949

Clean up NEW operator implementation

1972ee7

Add testdata for NEW operators

72d2a44

Fix receiver of isExpr

df9755f

Fix to use NamedType

8d2b00a

Update to use p.parseNamedType()

becd674

Refactor SQL() using helper functions

b3141e1

Add NEW constructor section in ast

3db71b8

Fix comments

67ea326

Update parser implementation of NEW constructors

bf06b50

Do go test --update

fc608f3

Fix code placement

8b37f43

Do go test --update

e9d5f86

Separate StructLiteral to TupleStructLiteral, TypelessStructLiteral, …

ed0bd6b

…TypedStructLiteral

Update testdata

432cdf0

Fix comment

15804df

Introduce AsExpr instead of TypelessStructValue and NewConstructorValue

64f8042

Rename AsExpr to ExprAsName, and use AsAlias

6a90127

Also, this fixes a `AsAlias.As` position bug.

makenowjust and others added 4 commits October 29, 2024 22:05

Update parser.go

c256076

Co-authored-by: apstndb <[email protected]>

Introduce TypelessStructLiteralArg and NewConstructorArg

d23d576

Remove requiredAs comments

41ae342

Add constants for the tryParseAsAlias argument

8ea0e53

makenowjust force-pushed the as-expression branch from a2a7a36 to 8ea0e53 Compare October 29, 2024 13:07

makenowjust merged commit db40951 into main Oct 29, 2024
4 checks passed

makenowjust deleted the as-expression branch October 29, 2024 13:17

This was referenced Nov 4, 2024

Support PROTO related syntax #89

Closed

Support typeless STRUCT syntax with field name #143

Closed

apstndb mentioned this pull request Nov 14, 2024

Implement EXCEPT and REPLACE in select item #196

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Introduce `TypelessStructLiteralArg` and `NewConstructorArg` interfaces #183

Introduce `TypelessStructLiteralArg` and `NewConstructorArg` interfaces #183

makenowjust commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

makenowjust commented Oct 27, 2024

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb Oct 28, 2024 •

edited

Loading

makenowjust Oct 28, 2024

apstndb Oct 28, 2024

makenowjust Oct 28, 2024

apstndb Oct 28, 2024

apstndb Oct 28, 2024

makenowjust commented Oct 28, 2024

apstndb left a comment

apstndb Oct 28, 2024

makenowjust Oct 28, 2024

apstndb Oct 28, 2024 •

edited

Loading

makenowjust Oct 28, 2024

apstndb Oct 28, 2024 •

edited

Loading

makenowjust Oct 28, 2024

apstndb Oct 28, 2024 •

edited

Loading

makenowjust Oct 28, 2024

apstndb Oct 29, 2024 •

edited

Loading

makenowjust Oct 29, 2024

apstndb left a comment •

edited

Loading

Introduce TypelessStructLiteralArg and NewConstructorArg interfaces #183

Introduce TypelessStructLiteralArg and NewConstructorArg interfaces #183

Conversation

makenowjust commented Oct 27, 2024 • edited Loading

apstndb commented Oct 27, 2024 • edited Loading

apstndb commented Oct 27, 2024 • edited Loading

makenowjust commented Oct 27, 2024

apstndb commented Oct 27, 2024 • edited Loading

apstndb commented Oct 27, 2024 • edited Loading

apstndb commented Oct 27, 2024 • edited Loading

apstndb Oct 28, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

makenowjust commented Oct 28, 2024

apstndb left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

apstndb Oct 28, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

apstndb Oct 28, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

apstndb Oct 28, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

apstndb Oct 29, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

apstndb left a comment • edited Loading

Choose a reason for hiding this comment

Introduce `TypelessStructLiteralArg` and `NewConstructorArg` interfaces #183

Introduce `TypelessStructLiteralArg` and `NewConstructorArg` interfaces #183

makenowjust commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb commented Oct 27, 2024 •

edited

Loading

apstndb Oct 28, 2024 •

edited

Loading

apstndb Oct 28, 2024 •

edited

Loading

apstndb Oct 28, 2024 •

edited

Loading

apstndb Oct 28, 2024 •

edited

Loading

apstndb Oct 29, 2024 •

edited

Loading

apstndb left a comment •

edited

Loading