Enter on File Path inputs opens a file browser (load/save, extension
filter, hidden toggle, overwrite confirm). Enter on Connection String
inputs opens a builder for PostgreSQL, MSSQL and SQLite with masked
password/preview, parsing and optional connection test.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Adds WriterOptions.TypeMappings and a repeatable --type-map sqltype=gotype
flag. Defaults are unchanged when no mapping is given. Closes#36 (bun/gorm).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- bun: scanonly+generated for GENERATED columns; scanonly+identity for ALWAYS identity (PK gets identity only)
- gorm: <-:false+generated / identity, same primary key rule
- dbml: write GENERATED/IDENTITY column notes and read them back into the model
- pgsql: emit generation/identity clauses in migration create-table and add-column, no DEFAULT
- readmes updated
Direct pgsql output (job output.conn_env, merge --output-conn) now reads the live
schema and executes only the differences. full_ddl: true (job option) or --full-ddl
(merge) restores the full idempotent DDL; file output is unchanged.
- compare PKs by columns, skip constraint-backed indexes, normalize index method,
FK actions, serial/numeric types, default literals/casts and truncated names
- diff table and column comments instead of re-emitting them
- remove leftover ZZDUMP debug code
key_column_usage and constraint_column_usage were joined on constraint name only,
so an N-column foreign key returned N*N column pairs. Read from pg_constraint with
unnest(conkey, confkey) instead.
- derive the primary key sequence from the column's nextval() default instead of an unused identity_<table>_<pk> sequence
- setval after table creation (full and diff paths), forward-only, MAX+1 with is_called=false
- do not DROP DEFAULT on serial/bigserial columns without a model default
Generated files carried the RelSpec version and build date, so every
rebuild produced a diff even when the schema was unchanged.
- bun, gorm, drizzle: remove the GeneratedBy template field and line
- pgsql, mssql, sqlite: emit a constant "-- Generated by RelSpec"
- bun: replace the version header test with a version-free assertion
pkg/buildinfo is unchanged and still backs the CLI version.
Commented refs are collected per file and resolved against the combined
model after all inputs are loaded (directory, --from-list, merge, jobs).
Matched refs become FKs and relationships; duplicates of existing FKs are
skipped; missing targets are skipped with a warning; column type
mismatches warn.
Also keep reused index names within a DBML table instead of overwriting,
give a second FK to the same table a distinct relationship name, and make
the pgsql writer match relationships to FKs by name first.
PostgreSQL index names share the schema-wide relation namespace, so a
reused name makes CREATE INDEX IF NOT EXISTS silently skip the duplicate.
Add duplicate_index_name rule (enforce by default) covering indexes and
PK/unique constraint backing indexes.
* extend Column struct with Generated, GenerationExpression, Identity, and IdentityGeneration fields
* update queryColumns to handle new column properties
* modify migration writer to avoid altering generated and identity columns
* add tests to ensure correct handling of generated and identity columns
- Add pkg/buildinfo with Version/BuildDate (set via ldflags, VCS fallback);
cmd/relspec now sources version info from it
- Emit "RelSpec <version> (built: <date>)" in generated file headers for
bun, gorm, drizzle, pgsql (incl. migration), mssql, sqlite writers
- pgsql writer: getSortedColumns now sorts by Sequence then Name so the
streaming WriteSchema/WriteDatabase path preserves source column order
- dbml reader: mergeTable re-bases merged-in column Sequence values past the
existing max, fixing colliding sequences (and alphabetical fallback) when a
table is split across multiple DBML files
- Tests for bun/gorm header + column order, and dbml multi-file merge ordering
* Add parsing for triple-quoted table notes in DBML
* Update writer to format multiline notes correctly
* Enhance tests for multiline table notes handling
* implement --silent flag to control output verbosity
* update error handling to restore stderr after silent mode
* enhance progress reporting in PostgreSQL reader
Add parseable `@<namespace>[(<target>)]: <args>` directives embedded in DBML.
They are stored losslessly on each object's Metadata, round-trip unchanged
through the DBML writer, and are translated to SQL only by the writer for the
matching dialect.
- models: Directive type + catalog; Metadata map added to Column and Index
- dbml reader: parse and attach directives at database/table/column/index
level; line-numbered errors; repeatable by default with singleton duplicate
detection. Fixes a preexisting bug where an `indexes {}` closing brace ended
the table early, dropping trailing Note: and directive lines.
- dbml writer: re-emit directives at their location; idempotent output
- pgsql writer: PARTITION BY / INHERITS / WITH / TABLESPACE (table),
STORAGE / COMPRESSION / identity (column), WITH / TABLESPACE (index)
- sqlite writer: WITHOUT ROWID / STRICT (table), COLLATE (column)
- --strict-directives flag on ReaderOptions and WriterOptions
- docs/DBML_DIRECTIVES.md + reader/writer READMEs
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ss2MY5J11cRGwEz86ZXk7d
Implements the remaining items from issue #20:
- version is now forward-permissive: any value >= 1 is accepted; a
newer-than-known version loads best-effort (unknown fields ignored,
warning printed) instead of hard-failing on "must be 1"
- from_job input reference: `inputs: [{ from_job: <job> }]` resolves to
that job's single-file output + format and implies a dependency edge;
combined depends_on + from_job graph gets topological ordering and
cycle detection
- logfile size-rotation, on by default (5MB, keep 3), overridable per
job (log_max_size / log_keep) or file-wide via a top-level defaults block
- new commands: split (schema/table subsetting via select:), inspect
(rule validation -> markdown/json report, fails job on enforced-rule
errors), diff (compare exactly two schemas, never fails), scripts-exec
(run SQL script dirs against a live PostgreSQL database)
- atomic single-file output/report writes (temp file + rename)
- symlink-escape hardening in SafeJoin via EvalSymlinks preflight
Updates docs/JOB_FILES.md and examples/jobs/relspec.yml accordingly.
Marshal JSON/YAML/XML and driver Value now emit time.RFC3339 instead of
the offset-less "2006-01-02T15:04:05" layout; zero-time guards compare
against the RFC3339 zero value. Parsing is unchanged and still accepts
bare datetimes.
Add `relspec job list` and `relspec job run <name>` driven by YAML job
manifests (relspec.yml / relspec.<name>.yml), so multi-file merge and
conversion workflows can be expressed declaratively instead of as long
shell command lines.
v1 contract (see docs/JOB_FILES.md):
- `command` is a closed allow-list (convert, merge, scripts-list); no
field accepts a shell string or executable path.
- Deterministic discovery: default file first, then named files sorted
lexically; all files merged into one namespace; duplicate job names
across files are a hard error.
- Every path resolves relative to the job file's directory; absolute,
home-relative and directory-escaping paths are rejected at validation.
- Database credentials referenced by env-var name via `conn_env:`;
connection strings are never stored and are redacted from logs/plan.
- Full validation (version, unknown fields, command/format, per-command
input/output shape, path traversal, depends_on targets, dependency
cycles) runs before anything is read, written or executed; per-job
pre-flight then checks input existence, script dirs, env vars and the
output overwrite policy for the whole plan.
- `depends_on` closure runs in deterministic topological order;
`--no-deps` runs only the named job.
- `--dry-run` (alias `--plan`) prints the resolved plan and exits 0
without touching inputs, outputs or databases.
- A failing job propagates the underlying non-zero exit status, logs
FAILED (never OK), and writes no success marker.
pkg/jobs is side-effect free (discovery/parse/validate/plan only);
execution adapters live in cmd/relspec/job.go. Includes unit tests for
discovery, validation, planning and path safety, plus CLI tests for
end-to-end convert/merge, scripts-list across multiple directories,
dry-run, dependency chains, exit-code propagation and log redaction.
Deferred: live `scripts execute` from jobs, split/inspect/diff/templ
commands, job-to-job output wiring, log rotation/retention.
Refs #20
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* Add handling for pgvector and PostGIS extensions in migration scripts
* Implement operator class and storage parameters for vector indexes
* Update tests to validate new index behaviors and extension creation
A composite [pk] entry inside an Indexes block (e.g. (a, b) [pk]) was
silently dropped: models.Index has no way to represent a primary key,
so the attribute was parsed and ignored, producing neither a PK nor a
meaningful index. It's now converted into a PrimaryKeyConstraint.
Also, Column.Sequence was never set by the DBML reader, so composite
PKs assembled from column-level [pk] attributes fell back to
alphabetical Name sorting instead of declaration order. Columns now
get a per-table sequence counter reflecting the order they were
declared.
* Implemented `Concurrent` field in index model
* Updated index creation template to support `CREATE INDEX CONCURRENTLY`
* Added tests for concurrent index creation in migration writer
SQLite can't ALTER TABLE ADD CONSTRAINT, so foreign keys are now written
as inline FOREIGN KEY clauses in CREATE TABLE instead of commented-out
ALTER statements. The default schema (public/main) now produces bare
table names instead of a "public_" prefix; other schemas are still
prefixed to avoid collisions. Also adds direct-to-file execution: the
sqlite writer can now apply generated DDL straight to a .db file via
Metadata["connection_string"], wired into `relspec merge --output-conn`.
Merge conflict detection compared bigserial (DBML) against bigint (live
PostgreSQL read of an existing serial column) as incompatible types, since
serial is sugar over an integer column plus a sequence default and
PostgreSQL always reports back the underlying integer type. Add
SerialUnderlyingType and use it when comparing column types for conflicts.
QuoteDefaultValue also wrapped bare keyword expressions like CURRENT_DATE
in string quotes because they contain no parentheses, unlike function-call
defaults such as now(). Recognize known bare keyword defaults and leave
them unquoted across CREATE TABLE, ALTER TABLE ADD COLUMN, and
ALTER COLUMN SET DEFAULT generation.
Backtick-wrapped defaults (e.g. from GORM tags like `now()`) were only
stripped in the CREATE TABLE column-definition path, leaving raw
backticks in the ALTER COLUMN ... SET DEFAULT migration statement and
in the migration-generated CREATE TABLE template, producing invalid
SQL. Default-drift comparisons also compared raw values, so a
backtick-wrapped model default never matched the live DB default and
kept re-emitting redundant ALTER statements.
Existing databases that already ran an old migration kept stale NOT
NULL constraints and mismatched column types/defaults, because the
schema writer only emitted idempotent ADD COLUMN IF NOT EXISTS guards
and never altered columns that already existed.
- Emit guarded ALTER COLUMN ... SET/DROP NOT NULL when a column's
nullability differs from the model.
- Emit guarded ALTER COLUMN ... TYPE, falling back to renaming the old
column and adding a fresh one when the in-place conversion fails.
- Emit guarded ALTER COLUMN ... SET/DROP DEFAULT for default drift.
- Collapse the previously duplicated plain/guarded templates so
WriteSchema (full-schema, live-state-checking) and WriteMigration
(diff-based) share the same guarded SQL templates and Go helpers
instead of maintaining the logic twice.
warkanum
SG Command
Hermes AgentandClaude Sonnet 5.5
sgcommand
HeinandClaude Sonnet 5