0.54.10

Merged PRs

doltgresql

• 2199: Support multiple statements
  This allows for handling multiple statements within a single Query message. This fixes:
  • #2175
    This also tests the contents of the messages returned, which were verified using Wireshark on a local Postgres instance as the source-of-truth.
• 2190: Initial support for UDFs with RETURNS SETOF
  Adds support for user-defined functions that use RETURNS SETOF to return values of a composite type.
  Support for UDFs that use RETURNS TABLE to return an anonymous composite type will be added as a follow up.

Closed Issues

• 2174: Error: DISTINCT ON is not yet supported

0.54.9

Merged PRs

doltgresql

• 2181: Added DISTINCT ON handling
  This adds support for:
  • #2174
    Related PRs:
  • dolthub/vitess#446
  • dolthub/go-mysql-server#3369

Closed Issues

0.54.8

Merged PRs

doltgresql

• 2172: drop schema support
  Fixes #2159
• 2158: Added bit and bit varying types
  Fixes #2149
• 2152: Bug fix: allow bpchar(N) casting
  Fixes: #2145
• 2151: Correctly support empty messages
  Fixes #2147
• 2139: #1863: validate database exists before CREATE SCHEMA

  Summary

  Adds a ValidateCreateSchema analyzer rule to ensure CREATE SCHEMA fails that appropriately when executed against a non-existent or invalid database context.
  • Adds analyzer rule that checks if the current database has a valid root before allowing CREATE SCHEMA to proceed
  • Ensures PostgreSQL-compliant behavior where schemas must be created within an existing database
  • Adds integration and bats tests

  Note

  I wasn't 100% sure if adding an analyzer rule was the right approach here, but it seemed like the cleanest way to validate database context before schema creation. The rule follows the same pattern as existing validation rules like ValidateCreateTable.
  The connection-level validation (rejecting connections to non-existent databases) already handles most cases, but this analyzer rule acts as an additional safety net for edge cases where a query might run with an invalid database context (also not sure how realistic this is 😕 ).
  Open to feedback on whether there's a better approach!

  Testing

  • Integrationt: TestCreateSchemaWithNonExistentDatabase
  • Bats: bats --filter "non-existent database|CREATE SCHEMA works" testing/bats/doltgres.bats
  • Existing schema tests pass: go test ./testing/go/... -run TestSchemas -v
    Closes: #1863
• 2138: #1361: unskip dolt_clean zero-argument tests

  Summary

  • Unskips the SELECT DOLT_CLEAN() zero-argument smoke tests that were added in #1373
  • I believe the underlying issue was resolved in #1763 which added Function0 registration for all Dolt procedures
  • Reorders test assertions so DOLT_CLEAN('t1') runs before DOLT_CLEAN() to properly test both
    Note: #1373 also added skipped tests in prepared_statement_test.go, but those were already unskipped in commit 62569c7.

  Tests to verify

  • Run go test ./testing/go/... -run "TestDoltClean" to verify existing dolt_clean tests pass
  • Run go test ./testing/go/... -run "TestDoltFunctionSmokeTests/smoke_test_select_dolt_clean" to verify unskipped smoke tests pass
    Fixes #1361
• 2137: #1868: remove outdated skip list entries for OR index tests

  Summary

  Remove outdated skip list entries for tests that were previously panicking due to an OR condition index lookup bug. The underlying issue was fixed in PR #2123.

  What was the problem?

  Issue #1868 reported a panic when executing queries like:

  SELECT * FROM test WHERE pk1 = 1 OR i = 1;

  where the first condition matches an index (primary key) and the second doesn't.

  What fixed it?

  I believe PR #2123 fixed this issue. In the PR it implemented a LogicTreeWalker for DoltgreSQL that properly handles doltgres-specific expression nodes when analyzing filters. This allows the query optimizer to correctly handle OR expressions with mixed indexed/non-indexed columns.

  Changes in this PR

  Removed 3 entries from the engine test skip list:
  • Complex Filter Index Scan #2
  • Complex Filter Index Scan #3
  • complicated range tree
    All three tests now pass.

  Testing

  • Verified all three previously-skipped tests now pass
  • Verified the exact scenario from issue #1868 works without panicking
    Fixes #1868
• 2136: #1057 allow unquoted STATUS keyword in DESCRIBE statements

  IMPORTANT

  Below is the summary of the cause of the bug in #1057 and the simple fix, but I want to clarify this is the desired fix.

  Summary

  • Fixes DESCRIBE dolt.status failing with "at or near 'status': syntax error"
  • Adds STATUS to the simple_ident grammar rule in sql.y, following the pattern established for PUBLIC in PR #828
  • Enables the previously skipped test for this functionality

  Background

  The simple_ident rule is used for parsing table name components in DESCRIBE statements. It only allowed IDENT tokens and the explicitly added PUBLIC keyword. Since STATUS is an unreserved keyword (not a regular identifier), the parser rejected it unless quoted.
  Before: DESCRIBE dolt.status fails, DESCRIBE dolt."status" works
  After: Both work

  Test Plan

  DESCRIBE dolt.status test now passes (TestUserSpaceDoltTables/dolt_status)
  Fixes #1057
• 2126: Add tests for SHOW TABLES bug
  Adds a test for a bug in SHOW TABLES behavior that caused sequences to be included in returned results.
  Depends on: dolthub/dolt#10220
  Fixes: #1743

Closed Issues

• 2149: bit type is not supported
• 2159: Please complete the implementation for dropping a schema in Doltgres
• 2099: [RFC] Exploring feasibility of Dolt or Doltgres as backend fo Matrix homeserver
• 1743: SHOW TABLES should not show sequences
• 2145: PostgreSQL compat: select '1'::bpchar(1) fails
• 2147: PostgreSQL compatibility: status field for empty query should be EMPTY
• 1359: Error adding a column with a check constraint to table with data
• 1361: dolt_clean procedure does not work if no arguments provided
• 1863: Handle PostgreSQL CREATE SCHEMA error when no database is selected
• 1057: describe dolt.status doesn't work
• 1868: Panic on SELECT ... WHERE ... OR
• 2127: JSONB SELECT Panic for large objects ~2KB

0.54.7

Merged PRs

doltgresql

• 2132: Fixed JSONB extended value error
  Fixes the issue:
  • #2127
• 2128: Added some hooks for DROP TABLE
  This adds support for ensuring that dropping a table while using table types doesn't outright put the database in an unusable state. There are many more hooks that need to be implemented across the board, but these are relatively straightforward ones for a common case of dropping a table, function, or procedure that uses them.
  This also adds support for DROP PROCEDURE, which I apparently overlooked when adding procedure support in general. Lastly, this adds support for the serialization and deserialization of values within columns that use table types. While it was previously possible to create tables that use table types, it wasn't possible to read from those tables (and writing to them just wrote useless data that couldn't be removed).
  You'll notice that error messages involving procedures return "function" instead of "procedure". Postgres calls them functions in their error messages, so we're replicating that.
• 2125: Changed configuration file for function passing
  In a nutshell, our internal repository for handling Hosted Doltgres only needs to include the configuration file, however recent changes added more packages that were referenced from our servercfg package. As our internal repository wasn't configured to handle these additional packages (which essentially requires building the entirety of Doltgres), it broke the existing workflow. To workaround this, I've moved the "details" of the configuration to its own package (named cfgdetails), and that will be referenced by our internal repository. Elsewhere in Doltgres, we still refer to servercfg, which has a wrapper struct around our details.
  The wrapper is needed as we cannot store a field that contains the overrides on the struct itself. Our YAML deserialization function does not honor ignored fields (which are supposed to work using yaml:"-") and is therefore panicking. Instead, the wrapper declares its own Overrides() function that overrides the function within cfgdetails.
• 2123: Bug fix: ensure OR expressions on a single table use an index when possible
  This fixes the performance regression in select_random_ranges, which uses BETWEEN .. OR BETWEEN .. in its filters.
  Requires dolthub/go-mysql-server#3347

Closed Issues

• 2097: auth.db, unable to configure path for "authFileName" file
• 2100: DOLTGRES_DB environment variable isn't respected

0.54.6

Merged PRs

doltgresql

• 2108: typo preventing release version from getting to docker push workflow
• 2107: added another function stub
• 2106: Fixed issue #2097
  This fixes
  • #2097
• 2093: remove doltgres index implementation, various bug fixes
  Companion PRs:
  • dolthub/go-mysql-server#3334
  • dolthub/dolt#10184
    This removes the custom index logic Doltgres was using, and fixes many bugs and incompatibilities that were causing the Dolt index logic to not work correctly in the first place.
• 2009: Tables as types
  Related PRs:
  • dolthub/dolt#10140
  • dolthub/go-mysql-server#3303

Closed Issues

• 2097: auth.db, unable to configure path for "authFileName" file
• 2100: DOLTGRES_DB environment variable isn't respected

0.54.5

Merged PRs

doltgresql

• 2102: Set the default DB from the DOLTGRES_DB user var
  Fixes #2100
• 2091: added MetricsJwksConfig() stub
• 2089: updated readme
• 2068: Fix Doltgres and Dolt authentication and system table compatibility
  Companion dolthub/dolt#10159
  • Add SUPERUSER authentication for generated Dolt procedure callables for SELECT statements.
  • Remove Dolt procedures from CALL statements.
  • Add adapters.TableAdapter for dolt_status system table to handle type change in Dolt due to a limitation with the MySQL protocol in distinguishing Booleans.
  • Add params column in the pgcatalog table for dolt_backups recent changes.
  • Add extensive suite of _go.ScriptTest for Dolt procedure authentication testing in Doltgres.
  • Add support to run local file system tests by setting flag in _go.ScriptTest for new tests. Some procedures require this because some Noms functions use os package calls for directory writes instead of the filesys package.

Closed Issues

• 2100: DOLTGRES_DB environment variable isn't respected

0.54.4

Merged PRs

doltgresql

• 2076: push to docker hub
• 2061: Dockerfile
  Also: configure the super user and password with new env vars:

  DOLTGRES_USER
  DOLTGRES_PASSWORD

Closed Issues

• 2060: Error parsing time "America/Los_Angeles"

0.53.6

Merged PRs

doltgresql

• 2069: /{Dockerfile,.dockerignore}: some tweaks
• 2067: Fixed timestamp replication
  This fixes the two issues found in
  • #2060
    The first was that replication was incorrectly formatting timestamp values, so now we bypass it and use the returned time strings directly. The second was that we had America/Los_Angeles hardcoded as the parameter value, and in some cases we would ignore the local timezone and use the parameter value. Now, we default the parameter value to the local as well.

Closed Issues

• 2049: SyntaxError: Bad control character in string literal in JSON

0.53.5

Merged PRs

doltgresql

• 2055: Fixed Issue #2049
  This fixes the following PR:
  • #2049
    At its core, our JSON parsing libraries (both go-json and the internal json packages) replace binary escape sequences (\n, \t, \r) with their ASCII byte replacements, which doesn't mirror what Postgres clients expect (that the binary representation is preserved). This extends to our testing framework as well, so this also adds raw byte comparisons for testing.
• 2047: updated README
  Also cleaned up some leftover dolt typos in install.sh

Closed Issues

• 2030: Drizzle relational query fails with BigInt error

0.53.4

Merged PRs

doltgresql

• 2041: Fixed a couple issues with prepared statements
  This PR addresses a couple issues discovered by a customer working with the Drizzle ORM
  1. Literals in LIMIT statements were being inappropriately flagged as invalid when they appeared in subqueries
  2. Bindvars in ON CONFLICT clauses were not being handled correctly
     Fixes #2030
• 2038: add metrics tls param to the server config
• 2036: Another attempt at install.sh

Closed Issues

• 2030: Drizzle relational query fails with BigInt error