graphql-engine/server/src-lib/Data/SqlCommenter.hs

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

113 lines
5.2 KiB
Haskell
Raw Normal View History

-- | Add metadata tags as comments to SQL queries.
module Data.SqlCommenter
( sqlCommenterGoogle,
sqlCommenterStandard,
Attribute,
)
where
import Data.List.NonEmpty qualified as NE
import Data.Text.Extended (commaSeparated)
import Hasura.Prelude
import Hasura.QueryTags
import Network.URI.Encode qualified as URI
-- | query-tags format as defined in the Spec <https://google.github.io/sqlcommenter/spec/#sql-commenter>
sqlCommenterGoogle :: QueryTagsAttributes -> QueryTagsComment
sqlCommenterGoogle qtAttributes
| null attributes = emptyQueryTagsComment
| otherwise = QueryTagsComment $ createSQLComment $ generateCommentTags (NE.fromList attributes)
where
createSQLComment comment = " /* " <> comment <> " */"
attributes = _unQueryTagsAttributes qtAttributes
-- | Default 'Query Tags' format in the Hasura GraphQL Engine
-- Creates simple 'key=value' pairs of query tags. No sorting, No URL encoding.
-- If the format of query tags is not mentioned in the metadata then, query-tags
-- are formatted using hte below format
sqlCommenterStandard :: QueryTagsAttributes -> QueryTagsComment
sqlCommenterStandard qtAttributes
| null attributes = emptyQueryTagsComment
| otherwise = QueryTagsComment $ createSQLComment $ generateComment (NE.fromList attributes)
where
generateComment attr = commaSeparated [k <> "=" <> v | (k, v) <- NE.toList attr]
createSQLComment comment = " /* " <> comment <> " */"
attributes = _unQueryTagsAttributes qtAttributes
-- | Top-level algorithm to generate the string comment from list of
-- 'Attribute's
-- Spec <https://google.github.io/sqlcommenter/spec/#sql-commenter>
-- See Note [Ambiguous SQLCommenterGoogle Specification]
generateCommentTags :: NE.NonEmpty Attribute -> Text
generateCommentTags attributes =
let -- 1. URL encode the key,value pairs. What the spec calls serialization.
-- https://google.github.io/sqlcommenter/spec/#key-serialization-algorithm
-- https://google.github.io/sqlcommenter/spec/#value-serialization-algorithm
encoded = NE.map urlEncodePair attributes
-- 2. Sort the pairs. Spec: https://google.github.io/sqlcommenter/spec/#sorting
-- Note the 'sort' from 'Data.List' works by sorting the first element in
-- the pair. And sorting on 'Text' is lexicographic.
sorted = NE.sort encoded
-- 3. Finally, serialize the pairs into a CSV. What the spec calls concatenation
-- https://google.github.io/sqlcommenter/spec/#concatenation
serialized = serializePairs sorted
in serialized
where
urlEncodePair (k, v) = (URI.encodeText k, URI.encodeText v)
serializePairs pairs = commaSeparated [k <> "='" <> v <> "'" | (k, v) <- NE.toList pairs]
-- | NOTE: [Ambiguous SQLCommenterGoogle Specification]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-- The specification is ambiguos/unclear about the following two steps:
--
-- 1. [Comment Escaping](https://google.github.io/sqlcommenter/spec/#comment-escaping)
-- ~~~~~~~~~~~~~~~~~~~~
-- The Spec states that:
--
-- > If a comment already exists within a SQL statement, we MUST NOT mutate that statement.
--
-- Oddly, the implementation of above rule/statement is not uniform among it's various language
-- libraries. That along with the fact that the above statement is clear for the scenarios when
-- we have properly formed comment, but does not state anything about what to do when the comments
-- are malformed is why it was decided to skip this step.
--
-- I have created a issue regarding this at https://github.com/google/sqlcommenter/issues/57
--
-- When discussed the above problem with Tiru, it was decided that since in our case we do not have
-- any other sql comments in the prepared statement, this step won't affect us thus it's okay to
-- skip this.
--
-- 2. [Escaping Meta Characters](https://google.github.io/sqlcommenter/spec/#meta-characters)
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-- The Spec states that:
--
-- > Meta characters such as `'` should be escaped with a slash `\`.
--
-- And the algorithm in brief is:
-- 1. URL encode the Key/Value
-- 2. Escape the meta-characters withing the raw value; a single quote `'` becomes `\'`
--
-- From the above statement, it could be understood that the meta-characters in context of
-- sqlcommenter is `'`. And 'Network.URI.Encode.encodeText' is capable of encoding `'`
-- without any need of escaping.
--
-- ```
-- Prelude Network.URI.Encode> encodeText"'"
-- "%27"
--
-- Prelude Network.URI.Encode> encodeText "\'"
-- "%27"
-- ```
--
-- During the investigation, we also found that the reference implementation of the sqlcommenter
-- in other languages did not even do this step:
-- 1. https://github.com/google/sqlcommenter/blob/master/python/sqlcommenter-python/google/cloud/sqlcommenter/__init__.py#L29
-- 2. https://github.com/google/sqlcommenter/blob/master/java/sqlcommenter-java/src/main/java/com/google/cloud/sqlcommenter/threadlocalstorage/State.java#L179
--
-- The above can be summarized as:
-- 1. The meta-characters in the context of sqlcommenter is `'`
-- 2. Reference implementation of sqlcommenter does not include the step
-- 3. 'Network.URI.Encode.encodeText' can encode `'` without any need of escaping
--
-- Thus, on the basis of the above three facts we decided to ignore the 'Escape Meta Characters' step.