Skip to content

Latest commit

 

History

History
146 lines (123 loc) · 7.64 KB

CHANGELOG.md

File metadata and controls

146 lines (123 loc) · 7.64 KB

Changelog

Notable changes to this fork, newest first. An entry that imports an upstream pull request names its number and author. Releases up to v1.4.0 are releases of upstream jmoiron/sqlx.

v1.4.3 (2026-10-20)

No API change. Batch inserts report an unclosed VALUES ( group instead of producing SQL the database rejects, and expand correctly when the group holds a bracket inside a string literal or comment.

Fixes

  • A batch insert (a slice or array passed to NamedExec, BindNamed and the like) whose VALUES ( group never closes now returns the error missing closing bracket in VALUES, whatever the number of rows. Before, the query came back unexpanded with every row's arguments, and the mistake only surfaced when the database rejected the argument count. Single-struct and map arguments are unchanged (#36).
  • The batch insert expansion skips brackets inside string literals, quoted identifiers and comments. VALUES (:a, ')') or VALUES (:a, /* ) */ :b) with several rows used to be cut at the bracket inside the literal or comment and expanded into invalid SQL; VALUES (:a, '(') was not expanded at all (#36).

v1.4.2 (2026-10-19)

No API change. sqlx's go.mod now requires no other module, so importing sqlx no longer changes which database driver versions a program builds with, and the upstream tags that came with the fork's history are retracted.

Module and toolchain

  • sqlx's go.mod requires no other module. The tests that run against Postgres, MySQL and SQLite move to the nested module dbtest/, which holds the three drivers, so go-sql-driver/mysql, lib/pq and go-sqlite3 no longer enter the module graph of a program that imports sqlx. Before, Go's minimum version selection raised a consumer's own MySQL driver to v1.10.1 even though sqlx only used it in tests (upstream #642, #569, #849). The checks in three tests that read the unexported unsafe field now test its effect instead: scanning into a struct that lacks some selected columns (#34).
  • v1.2.0 to v1.4.0 are retracted under this module path. They are upstream's tags, pushed with the history; their go.mod declares github.com/jmoiron/sqlx, so go get github.com/secondwindoss/sqlx@v1.4.0 failed with a module path error. go list -m -versions now lists only the fork's own releases (#35).

Project

  • Release tags are signed; GitHub shows them as verified.
  • An OpenSSF Scorecard workflow publishes the repository's score, shown as a README badge; CI actions are pinned by commit SHA; Go fuzz tests cover the SQL scanner, Rebind, compileNamedQuery and In, and their seed corpus runs with the unit tests (#32, #33). CodeQL scans every pull request, and merging into master requires the CI checks to pass.
  • The README is rewritten around the fork, with a migration section that explains why a replace directive cannot stand in for the import rewrite (#28, #29, #30).

v1.4.1 (2026-10-10)

The first fork release. It keeps every exported identifier of v1.4.0 (CI runs apidiff against the v1.4.0 tag on every change), requires Go 1.24, adds one method and fixes the bugs that sat in upstream pull requests. The documented behaviours that change are marked changed.

Module and toolchain

  • The module path is github.com/secondwindoss/sqlx. The package name stays sqlx; only the import path changes (#2).
  • Changed: the go directive is 1.24.0, up from 1.10, as upstream #1001 (AshSgDe29071999) proposed. The test drivers move to mysql 1.10.1, pq 1.12.3 and sqlite3 1.14.52, and the deprecated io/ioutil and reflect.PtrTo calls are replaced (#11). filippo.io/edwards25519 is no longer pinned at the version that module scanners flag (#9, #11; upstream #994).

Fixes

  • Named queries, In and Rebind scan SQL instead of counting bytes (#23). A ? or : inside a string literal, a quoted identifier, a -- or nested /* */ comment or a Postgres $tag$ string is no longer a placeholder or a parameter. Fixes upstream #487, #608 and #961 (? in literals and comments, as #984 by Charlie Tonneslan and #768 by Daniel Morell attempted), #206, #253, #302, #368, #394, #441, #443, #872, #894 and #947 (: in literals and comments, as #151 by Mark Beznos, #474 by detaoin and #775 by David Sharnoff attempted) and #983 (a cast attached to a parameter, :id::uuid, as #1002 by Burak Keskin and #985 by Charlie Tonneslan attempted). Parameter names may contain non-ASCII letters.
  • Changed: :: is a cast and passes through untouched, which fixes now()::date (upstream #428, #650, #956). The old escape that turned :: into : is gone, so a query that wrote :::: to get a cast now writes ::.
  • Changed: ?? is a literal ? in In and Rebind, the pgJDBC and squirrel convention, so the Postgres JSONB operators can be written ??, ??| and ??& (upstream #628). Rebind(QUESTION, q) rewrites ?? to ?.
  • Changed: a parameter name can no longer start with a digit; :1 is text.
  • Changed: In expands an empty or nil slice to NULL instead of returning an error, so x IN (?) becomes x IN (NULL) and matches no row. x NOT IN (NULL) also matches no row. Callers that used the error as an empty-list guard must check the length themselves (#22; upstream #945 by Markus, markus-wa).
  • In no longer panics on a nil pointer whose type implements driver.Valuer (#12; upstream #953 by Alexandre Thomas, issue #952).
  • Batch inserts expand a VALUES list that has no column list before it, sits in a FROM or JOIN, or starts on its own line (#13; upstream #962 by Caleb Spare, issues #746, #753, #772, #898; cases from #754 by MishaNiki, #811 by xyz1hang and #948 by stephennancekivell). The group repeated is the first one that contains a bind variable, so a table named values or a MySQL VALUES(col) is not mistaken for it.
  • Named queries accept a pointer to a slice, to a map and to a slice of maps (#20; upstream #805 by Xudong Cai, issues #577, #729).
  • A statement prepared on a Conn keeps the Conn's Mapper and unsafe flag (#15; upstream #993 by itxaiohanglover, issue #975).
  • Tx.Stmtx and Tx.StmtxContext keep the transaction's unsafe flag and the wrapped statement's own (#18; upstream #979 by Gianmarco Fantinuoli).
  • Rows.NextResultSet resets the field cache, so a second result set with a different column layout scans instead of panicking (#19; upstream #858 by Kamil Akhmetov, issues #380, #857; test from #747 by Peter Magnusson).
  • ConnectContext closes the DB and returns nil when the ping fails, as Connect does (#16; upstream #561 by Timo Reitz, issue #910).
  • A colon followed by nothing usable is text, not an empty parameter. The clearer error for that case from upstream #920 by Lee Fernandes went in first (#17); the scanner then made the case unreachable.

Additions

  • Tx.NamedQueryContext (#21; upstream #724 by Dave Worth, issue #447). The prepared-statement queryer also gains QueryRow and QueryRowContext, from #941 by Jon Senchyna.
  • The README and the In doc comment say that In always takes and returns ? placeholders (#14; upstream #1003 by TastyHeadphones, issue #959).

Project

  • CI runs on GitHub Actions: go vet, staticcheck, goimports, govulncheck and apidiff, plus the race tests on Go 1.24 to stable against Postgres, MySQL and SQLite (#1, #11, #24).
  • README fork notice and migration section, LICENSE line, CONTRIBUTING.md, SECURITY.md, a bug report template and Dependabot (#3, #4, #5, #10).

v1.4.0 (2024-04-15, upstream)

The last upstream release; see its release notes.