supabase / supabase-swift / 37920151054

93%

push

github

feat(postgrest): generate @Table structs with a Swift typegen (#1488)

* feat(typegen): add the supabase-typegen executable and its GeneratorMetadata input

Decodes postgrest-typegen's GeneratorMetadata (version 1) from standard
input, indexes it into a deterministic model, and parses --schema,
--output and --access-control. Exit codes follow sysexits: 64 for usage,
65 for data errors. Output is a placeholder until the emitter lands.

Fixtures come from postgrest-typegen's introspect against the integration
schema and against postgrest-typegen's own fixture SQL, refreshed with
scripts/refresh-typegen-fixtures.sh.

Co-Authored-By: Claude <noreply@anthropic.com>

* feat(typegen): emit @Table structs from the GeneratorMetadata model

The generator now writes one struct per relation, with properties typed
from the Postgres column types, and a PostgrestSchema enum for every
selected schema other than public. Declarations are built with
SwiftSyntaxBuilder and printed with swift-format under a built-in
configuration, so the output is the same on every machine.

Names that @Table cannot take are renamed and reported on standard error:
members the expansion adds, keywords (escaped), and Postgres names that
convert to the same Swift name. A type name that clashes across schemas
is a data error. Enums and unmapped types decode as JSONValue for now.

Golden files under Tests/SupabaseTypegenTests/__Goldens__ pin the output;
SUPABASE_TYPEGEN_RECORD=1 rewrites them.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(postgrest): escape quotes and backslashes in names the macros splice

@Table, @Column and @SelectionOf read their string arguments with
representedLiteralValue, which undoes the author's escapes, then spliced
the value between quotes in the expansion. A relation or column name
holding " or \ produced an expansion that did not parse. The name is now
escaped again before it lands in a generated literal.

Co-Authored-By: Claude <noreply@anthropic.com>

... (continued)

10 of 11 new or added lines in 1 file covered. (90.91%)

12358 of 13539 relevant lines covered (91.28%)

140.98 hits per line