(ns metabase.query-analysis.native-query-analyzer.impl (:require [metabase.driver :as driver])) | |
NOTE: In the future we should replace [[considered-drivers]] and [[macaw-options]] with (a) driver method(s), but given that the interface with Macaw is still in a state of flux, and how simple the current configuration is, we defer extending the public interface and define the logic locally instead. Macaw itself is not battle tested outside this same narrow range of databases in any case. | |
Since we are unable to ask basic questions of the driver hierarchy outside of that module, we need to explicitly mention all sub-types. This is probably not a bad thing. | (def ^:private considered-drivers
#{:h2 :mysql :postgres :redshift :sqlite :sqlserver}) |
Do we trust that Macaw will not give us false negatives for tables referenced by a given query? At some point, we may want a way for 3rd party drivers to opt in, but a public API deserves some hammock time. Using a separate list to the above, as we may want this to be more restrictive. | (def trusted-for-table-permissions?
#{:h2 :mysql :postgres}) |
Generate the options expected by Macaw based on the nature of the given driver. | (defn macaw-options
[driver]
;; If this isn't a driver we've considered, fallback to Macaw's conservative defaults.
(when (contains? considered-drivers driver)
;; According to the SQL-92 specification, non-quoted identifiers should be case-insensitive, and the majority of
;; engines are implemented this way.
;;
;; In practice there are exceptions, notably MySQL and SQL Server, where case sensitivity is a property of the
;; underlying resource referenced by the identifier, and the case-sensitivity does not depend on whether the
;; reference is quoted.
;;
;; For MySQL the case sensitivity of databases and tables depends on both the underlying file system, and a system
;; variable used to initialize the database. For SQL Server it depends on the collation settings of the collection
;; where the corresponding schema element is defined.
;;
;; For MySQL, columns and aliases can never be case-sensitive, and for SQL Server the default collation is case-
;; insensitive too, so it makes sense to just treat all databases as case-insensitive as a whole.
;;
;; In future, Macaw may support discriminating on the identifier type, in which case we could be more precise for
;; these databases. Being 100% correct would require querying system variables and schema configuration however,
;; which is likely a step too far in complexity.
;;
;; Currently, we go with :agnostic, as it is the most relaxed semantics (the case of both the identifiers and the
;; underlying schema is totally ignored, and correspondence is non-deterministic), but Macaw supports more nuanced
;; :lower and :upper configuration values which coerce the query identifiers to a given case then do an exact
;; comparison with the schema.
{:case-insensitive :agnostic
;; For both MySQL and SQL Server, whether identifiers are case-sensitive depends on database configuration only,
;; and quoting has no effect on this, so we disable this option for consistency with `:case-insensitive`.
:quotes-preserve-case? (not (contains? #{:mysql :sqlserver} driver))
:features {:postgres-syntax (isa? driver/hierarchy driver :postgres)
:square-bracket-quotes (= :sqlserver driver)
:unsupported-statements false
:backslash-escape-char true
;; This will slow things down, but until we measure the difference, opt for correctness.
:complex-parsing true}
;; 10 seconds
:timout 10000
;; There is no plan to be exhaustive yet.
;; Note that while an allowed list would be more conservative, at the time of writing only 2 of the bundled
;; drivers use FINAL as a reserved word, and mentioning them all would be prohibitive.
;; In the future, we will use multimethods to define this explicitly per driver, or even discover it automatically
;; through the JDBC connection, where possible.
:non-reserved-words (vec (remove nil? [(when-not (contains? #{:snowflake :oracle} driver)
:final)]))})) |