[MASSMAIL]❓ JSON Path Dot Precedence - Mailing list pgsql-hackers

From David E. Wheeler
Subject [MASSMAIL]❓ JSON Path Dot Precedence
Date
Msg-id 2F757EB8-AEB9-49E8-A2C6-613E06BA05D4@justatheory.com
Whole thread Raw
Responses Re: ❓ JSON Path Dot Precedence
Re: ❓ JSON Path Dot Precedence
List pgsql-hackers
Hello Hackers,

A question about the behavior of the JSON Path parser. The docs[1] have this to say about numbers:

>  Numeric literals in SQL/JSON path expressions follow JavaScript rules, which are different from both SQL and JSON in
someminor details. For example, SQL/JSON path allows .1 and 1., which are invalid in JSON. 

In other words, this is valid:

david=# select '2.'::jsonpath;
 jsonpath
----------
 2

But this feature creates a bit of a conflict with the use of a dot for path expressions. Consider `0x2.p10`. How should
thatbe parsed? As an invalid decimal expression ("trailing junk after numeric literal”), or as a valid integer 2
followedby the path segment “p10”? Here’s the parser’s answer: 

david=# select '0x2.p10'::jsonpath;
 jsonpath
-----------
 (2)."p10"

So it would seem that, other things being equal, a path key expression (`.foo`) is slightly higher precedence than a
decimalexpression. Is that intentional/correct? 

Discovered while writing my Go lexer and throwing all of Go’s floating point literal examples[2] at it and comparing to
thePostgres path parser. Curiously, this is only an issue for 0x/0o/0b numeric expressions; a decimal expression does
notbehave in the same way: 

david=# select '2.p10'::jsonpath;
ERROR:  trailing junk after numeric literal at or near "2.p" of jsonpath input
LINE 1: select '2.p10'::jsonpath;

Which maybe seems a bit inconsistent.

Thoughts on what the “correct” behavior should be?

Best,

David

  [1]: https://www.postgresql.org/docs/devel/datatype-json.html#DATATYPE-JSONPATH
  [2]: https://tip.golang.org/ref/spec#Floating-point_literals


pgsql-hackers by date:

Previous
From: Erik Wienhold
Date:
Subject: Re: CASE control block broken by a single line comment
Next
From: Andres Freund
Date:
Subject: Re: Cluster::restart dumping logs when stop fails