<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
We pass around a lot of `BTreeMap` and `IndexMap` types. This has two
problems:
a) everytime we change the items inside, we have to manually update lots
of call sites
b) we can't attach useful behaviour to it
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
This adds some wrappers around `object_types`, and adds a `.get()`
function with a useful default error. This error only contains the
missing `type_name`, so the idea is that it would be wrapped in another
error that would provide more context. The hope is that we can get away
from one giant error enum, and instead have a set of smaller error types
that live along each resolving stage.
In isolation this PR isn't very interesting, as it's a tiny drop in the
ocean, so really it's here to say, "is this a thing we vaguely like?"
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: 804a703d7155ce5b929937689d5649c6571357b0
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Put all the errors that happen in the `data_connectors` stage into a new
type `DataConnectorError`.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Moving enum members to the new type. Functional no-op.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: 2e50eb561ce6b7c7fac4de4b31c9957f7ee4696c
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Much like https://github.com/hasura/v3-engine/pull/808, move an error to
the place it is thrown.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Move files around. Functional no-op.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: a61527b57662efada2c7d2049e12f103deec1e4e
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
We have a giant error file in `metadata-resolve`, and it's really
unclear what can go wrong where. Let's improve it in some small way.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Move `GraphqlConfigError` to the `graphql_config` stage, and make that
stage only return that kind of error.
Functional no-op.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
---------
Co-authored-by: Samir Talwar <samir.talwar@hasura.io>
V3_GIT_ORIGIN_REV_ID: 9aa58875a24d66e6945c9ead6434fedb124f1dd8
This PR removes the usage of `ndc_models` `Expression` types from the
IR. It then adds logic to map from the new IR `FilterExpression` types
to `ndc_models` types in the plan part of the code where IR ->
ndc_models mapping is performed. The new IR types and the change to the
IR creation logic is mostly in `crates/execute/src/ir/filter.rs`.
The new IR types have some optimisations applied to them. In particular,
some basic boolean expression simplification logic is applied when
`FilterExpression::mk_and` `mk_or` and `mk_not` are used. This strips
out redundant and/ors/nots resulting in simpler boolexps for connectors
to process. (This logic is similar to what GDC did in Hasura v2).
Unfortunately it turned out that arguments had JSON-serialized
`ndc_models` types embedded in them, in particular, where argument
presets had been used to set a BooleanExpression as an argument value.
The old code was simply creating an ndc_models::Expression and
serializing it directly to JSON. This has now been refactored to retain
the new IR `FilterExpression` until `plan` time, at which point it will
be mapped into the ndc type and serialized to JSON and embedded in the
argument value. The types change for this can be seen in
`crates/execute/src/ir/arguments.rs`, but the real logic is actually in
`crates/execute/src/ir/permissions.rs`, with the
`make_value_from_value_expression` and
`make_argument_from_value_expression` functions.
The mapping of expression and argument IR into `ndc_models` can be found
in `crates/execute/src/plan/common.rs`.
In `metadata_resolve`, some ndc_models types were removed and
non-ndc_model types were used instead. In particular
`ndc_models::ComparisonOperatorName` (swapped with
`DataConnectorOperatorName`) and `ndc_models::UnaryComparisonOperator`
(swapped with a new `UnaryComparisonOperator` enum type).
This PR is a functional no-op, other than the boolean expression
simplification.
V3_GIT_ORIGIN_REV_ID: 13805430fb2e2ca5653406a094d719138f0d5b51
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Since these code comments end up in the docs, let's make sure we point
out these old types are no longer in favour.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Updating doc comments.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: a21190fbf2327e1ae265c76d1f467af063f1dd64
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
If the target data connector for a relationship does not have the
`foreach` capability, we threw an error, however we should allow this.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
Check data connector names for relationship target before throwing
capability error.
V3_GIT_ORIGIN_REV_ID: 157908f0077b7c3af67a77600e5f8c9fc65df7f2
Note: This PR is stacked on #797 and should be merged after it.
This PR removes the use of ndc_models types from the arguments and order
by IR. It then shifts the logic that maps the IR to the ndc_models into
the plan part of the code where IR -> ndc_model mapping is performed.
This will help isolate the IR from ndc_models and move us towards being
able to have multiple IR -> ndc_models mapping codes, one per supported
ndc version.
This is a functional no-op PR.
V3_GIT_ORIGIN_REV_ID: 66be868ff4c8185c6190537d570d88813cb7f410
This PR replaces most of the string newtypes in the open-dds crate with
new implementations based on `SmolStr`, using a macro adapted from the
macro used in ndc-models. `SmolStr` usage should result in less heap
allocations and the macro ensures all the newtypes get all the same
trait implementations (such as various `From` and `Borrow` instances)
and helper impl functions (such as `as_str`).
The new macro, `str_newtype`, creates a newtype wrapper around `SmolStr`
or another type (typically `Identifier`), and writes out all the various
trait implementations. It also takes a documentation string which
ensures that every newtype has a description in JSON Schema.
The changes in the downstream crates are just adjusting to use the new
newtypes.
Also:
* `QueryGraphqlConfig` used a lot of String typed-properties; these have
been changed for the appropriate `GraphQlTypeName` and
`GraphQlFieldName` types.
* metadata-resolve had a duplicate newtype called
`ConnectorArgumentName`, which duplicated open-dds's
`DataConnectorArgumentName`. This has been removed and replaced.
* A new newtype called `CollectionName` has been added in open-dds to
represent the name of the ndc collection that a model points to. This
was previously just a string.
V3_GIT_ORIGIN_REV_ID: d5a33756ef0e110b2255478358a38e42e6ff89a2
This PR just sorts the definitions in the Open DD JSON schema file, so
that when it changes, we can get better diffs because the types won't
move around inside the file.
V3_GIT_ORIGIN_REV_ID: 80221fae1b4d6e4b98b658f9ba2f19413abf7389
### What
This PR fixes a bug with variable nullability coercion. Specifically,
providing a non-null variable for a nullable field should work, as all
non-nullable variables can be used as nullable variables via "coercion".
Related issues:
* https://hasurahq.atlassian.net/browse/V3ENGINE-243
* https://github.com/hasura/v3-e2e-testing/pull/224 (the test for this
change)
### How
I think someone did something clever with macros, so I tried to unpack
the macro to make sense of it, and in the process, spotted the bug: it's
not that both the location must be nullable AND the variable not, it's
that _if_ the variable is nullable, the location must not be nullable.
V3_GIT_ORIGIN_REV_ID: 4f4ba3fe6898220bbba9ae2867233cd762bef1cb
Update the changelog for new version
---------
Co-authored-by: Anon Ray <ecthiender@users.noreply.github.com>
V3_GIT_ORIGIN_REV_ID: b87f16ebaa1471f010ec461be097bcfd6648c99a
This PR updated ndc-models to the latest version on main. This version
is still a 0.1.x version, but it now includes all the [new
newtypes](https://github.com/hasura/ndc-spec/pull/156) that wrap
previously stringly-typed things. For example, `ArgumentName`,
`FieldName`, etc.
This pervades across the entire engine, but thankfully the changes are
mostly mechanical repetitive changes. Usually you will see conversions
from `String`-typed variables into the newtypes using this sort of form:
`FieldName::from(string.as_str())`, which is the most efficient way
copying the value (the str slice is copied). Or you will see usages of
the newtype as a raw string by `.as_str()`-ing it. Converting the
newtypes into a String can be done with `.into()` if owned, but if
referenced `.as_str().to_owned()` performs the clone and type
conversion.
Other changes:
* A few minor instances of `ok_or()` usages (or similar) have been
converted into lazy error construction variants (eg `ok_or_else()`)
V3_GIT_ORIGIN_REV_ID: 64a371ae6197ef3be98a6f7cdc4052d654a43da0
The PR adds code that handles multiple versions of the ndc-models during
the execution pipeline. Depending on whether a connector supports `v01`
or `v02` models, a different set of types is used to send and receive
the http requests.
However, the engine internally still uses the latest (v02) models inside
its IR. Unfortunately, it was going to be quite traumatic to prevent the
engine from using ndc models inside the IR and during response
processing and remote joins. This means that the engine generates v02
requests, and these are downgraded into v01 requests. v01 responses are
upgraded to v02 responses before being processed by the engine.
The ndc client (`execute::ndc::client`) now only takes new wrapper enum
types (`execute::ndc::types`) that split between v01 or v02
requests/responses. Every place that carries an ndc request now carries
this type instead, which allows it to carry either a v01 or a v02
request.
When ndc requests are created during planning, all creation goes via the
new `execute::plan::ndc_request` module. This inspects the connector's
supported version, creates the necessary request, and if needed,
downgrades it to v01.
When ndc responses are read during planning or during remote joins, they
are upgraded to v02 via helper functions defined on the types in
`execute::ndc::types`.
The upgrade/downgrade code is located in `execute::ndc::migration`. Keep
in mind the "v02" types are currently the same as the "v01" types so the
migration code is not doing much. This will change as the v02 types are
modified.
However, this approach has its drawbacks. One is that it prevents
changes to the ndc types [like
this](https://github.com/hasura/ndc-spec/pull/158) without a fair bit of
pain (see
[comment](https://github.com/hasura/ndc-spec/pull/158#issuecomment-2202127094)).
Another is that the downgrade code can fail at runtime and it is not
immediately obvious to developers using new, but unused, v02 features
that their new feature would fail on v01, because that mapping to v01
has already been written. Another is that we're paying some (small,
probably?) performance cost by upgrading/downgrading types because we
need to rebuild data structures.
Also:
* `execute::ndc::response` has been merged into `execute::ndc::client`,
since it was inextricably linked.
V3_GIT_ORIGIN_REV_ID: f3f36736b52058323d789c378fed06af566f39a3
### What
When generating GraphQL schema for relationship fields in model filter,
engine ignores `relation_comparisons` capability of the data connector.
Engine would generate schema for data connectors which don't have this
capability. This PR fixes that.
### How
While generating fields for the filter input type, take the relationship
capabilities into account.
The `ObjectBooleanExpressionType` and `BooleanExpressionType` objects
are quite different, hence their schema generation part is also
different, and is split in two different functions
(`build_comparable_relationships_schema`, and
`build_new_comparable_relationships_schema`). Added checking of
relationship comparison capability in both the functions.
V3_GIT_ORIGIN_REV_ID: dce2b88f7792e01e5bb390ecdb580e223ec80f01
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
# BooleanExpressionType
A new metadata kind `BooleanExpressionType` can now be defined. These
can be used in place of `ObjectBooleanExpressionType` and
`DataConnectorScalarRepresentation`, and allow more granular control of
comparison operators and how they are used.
The old metadata types still work, but will eventually be deprecated.
```yaml
kind: BooleanExpressionType
version: v1
definition:
name: album_bool_exp
operand:
object:
type: Album
comparableFields:
- fieldName: AlbumId
booleanExpressionType: pg_int_comparison_exp
- fieldName: ArtistId
booleanExpressionType: pg_int_comparison_exp_with_is_null
- field: Address
booleanExpressionType: address_bool_exp
comparableRelationships:
- relationshipName: Artist
booleanExpressionType: artist_bool_exp
logicalOperators:
enable: true
isNull:
enable: true
graphql:
typeName: app_album_bool_exp
```
```yaml
kind: BooleanExpressionType
version: v1
definition:
name: pg_int_comparison_exp
operand:
scalar:
type: Int
comparisonOperators:
- name: equals
argumentType: String!
- name: _in
argumentType: [String!]!
dataConnectorOperatorMapping:
- dataConnectorName: postgres_db
dataConnectorScalarType: String
operatorMapping:
equals: _eq
logicalOperators:
enable: true
isNull:
enable: true
graphql:
typeName: app_postgres_int_bool_exp
```
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Remove feature flag, unhide JsonSchema items, fix a few missing bits of
JsonSchema the tests didn't warn us about before.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: dd3055d926fdeb7446cd57085679f2492a4358a1
### What
This is a no-op refactor that involves unifying `get_base_type` and
`get_underlying_type_name` functions, whose motives are the same.
### How
Replace the name of erstwhile `get_base_type` fn with
`get_underlying_type_name`. Remove the latter. Update the rest of the
code to use the new function.
V3_GIT_ORIGIN_REV_ID: 0b0d999670641ba265fde153af9b43b4d865e215
### What
Update ndc-postgres configuration to v4, including the new
`mutationsVersion: "v4"`.
### How
- sed `s/experimental_/v2_/g`
- `ndc-postgres-cli upgrade`
- `ndc-postgres-cli update`
V3_GIT_ORIGIN_REV_ID: 67c608c2a3f1e232d5e0725fe8817672aa3dd627
This PR introduces support for multiple versions of the ndc-spec by
adding a new `VersionedSchemaAndCapabilities` enum variant under the
`DataConnectorLink` in OpenDD. This allows the capture of both ndc
v0.1.* and v0.2.* schema and capabilities.
This is achieved by referencing the `ndc-models` crate twice, once for
`v0.1.4` and once for the first commit after `v0.1.4`. That commit was
chosen to avoid actual v0.2.0 breaking changes for now, while we lay in
this multiple version support plumbing. Future PRs will use a newer
commit and adopt the breaking changes where necessary. The
`VersionedSchemaAndCapabilities::V02` variant uses the the v0.2
reference of `ndc-models`.
Then, during metadata resolve, when we resolve the
`DataConnectorContext` from `DataConnectorLink`, we perform a migration
of v0.1 types to v0.2 types and store and use the v0.2 types during
metadata resolve. This migration is performed in the new module
`ndc_migration`. We also record the `NdcVersion` (either `V01` or `V02`)
in the `DataConnectorLink`. The `execute` crate will need to use this to
determine which version to send to the connector at runtime (to be
implemented in a future PR).
The new changes to OpenDD are hidden from the JSON Schema via a new
`UnstableFeatures` flag, and the use of the new variant is gated behind
it in metadata resolve, since we don't yet support it upstream in the
`execute` crate.
V3_GIT_ORIGIN_REV_ID: d6d8a768ea3537c0b5e620799e94d3dd1e529526
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Adds new Open DD type `OrderByExpression` and defines `ModelV2` type, as
described in
https://github.com/hasura/v3-engine/blob/main/rfcs/open-dd-expression-type-changes.md.
### How
- Added new types `OrderByExpression`, `ModelV2` and
`ModelGraphQlDefinitionV2` to the `open-dds` crate.
- Added new `UnstableFeatures` flag `enable_order_by_expressions` to
`metadata-resolve`. This is not yet used, as `metadata-resolve` does not
yet use the new Open DD types.
---------
Co-authored-by: Daniel Harvey <danieljamesharvey@gmail.com>
V3_GIT_ORIGIN_REV_ID: e861e88f293f1b380e6346de8490b96cec6f65bb
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Changed tests to use the new `BooleanExpressionType` until something
broke. Fortunately something broke - we need to add the scalar boolean
expression types to the data connector scalar types for various lookups.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Split scalar boolean expression type resolve into own step so it's
available earlier in the pipeline.
Loop through them and add them to the `data_connector_scalar_types`
outputs.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: dbd8969c3e9e9d8db1d4a34e93aefc34bdf31421
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Allow engine to connect to NDCs via HTTPS.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Add `cacert` to Docker image using Nix.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: 52458920236f3868cc8daf18e140f8536d9bc674
This PR refactors the `DataConnectorContext` type in
`crates/metadata-resolve/src/stages/data_connectors/types.rs` to remove
the extra copy of scalar types it had and simplifies and removes the
nesting in the types.
`DataConnectorContext` used to have its own copy of `scalars` which
contained `ScalarTypeInfo`s. However, it already contains a more
complete copy of scalar types inside `inner.schema.scalar_types`! Turns
out the `scalars` copy (and `ScalarTypeInfo`) is unnecessary. The only
value it added was having the computed `ComparisonOperators` struct,
which is simply copied from there to where it really lives on
`ScalarTypeWithRepresentationInfo` during the
`data_connector_scalar_types` stage.
So I've moved `ComparisonOperators`, and the code that creates it, to
`data_connector_scalar_types` and deleted `ScalarTypeInfo`. This meant
that `DataConnectorContext` only contained `DataConnectorCoreInfo`
(pointless!), so I inlined `DataConnectorCoreInfo` into
`DataConnectorContext` to simplify things. This removed `.inner` calls
all through the code.
These changes help my work with supporting multiple `ndc-models`
versions because it simplifies the number of places we store and deal
with scalar types.
This PR is a functional no-op.
V3_GIT_ORIGIN_REV_ID: 3beb0b07abc7ff5cfaa7e9b60eb46aec94d7ec1a
<!-- Thank you for submitting this PR! :) -->
## Description
Upgrade Rust, as a treat. Functional no-op.
---------
Co-authored-by: Samir Talwar <samir@functional.computer>
V3_GIT_ORIGIN_REV_ID: 1e0014049e89b8658326c8d8f652df800c415526
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
In https://github.com/hasura/v3-engine/pull/750 and
https://github.com/hasura/v3-engine/pull/754 we added a number of checks
for boolean expression types that can be run once we know the data
source they will be used against.
Previously these checks were only used for model `where` clauses, this
pull request moves them into a shared folder and also checks them when
they are used as model or command arguments.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Add a new `arguments` resolve step. It doesn't fit inside the usual
`commands` or `models` step as we need the data sources for both
validated, plus access to all the resolved `relationships` outputs.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: f713659962e3f20b2c85f287b6c362fb52ffa1ed
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Much like https://github.com/hasura/v3-engine/pull/774, we split the big
files into separate modules. Functional no-op.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Copy pasta, `just fix-local`, mostly.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: ba7e40057583f98df54573c72663a4a2d2c4a4ab
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
Tiny no-op PR to split the `command` stage of `metadata-resolve` into
smaller modules the same way `models` and others work.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
Copy and paste, run `just fix-local` to remove unrequired imports.
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: ab2846be83ba0e948ea222b42eb4fd7ffa5b3523
### What
Output all traces to stdout.
---------
Co-authored-by: Daniel Harvey <danieljamesharvey@gmail.com>
V3_GIT_ORIGIN_REV_ID: 06330076ca305a331996530ddcd4d4c13d46bd95
### What
This adds a flag, `--partial-supergraph`, which instructs the metadata
resolver to prune relationships to unknown subgraphs rather than failing
to resolve.
### How
The flag gets passed through as
`metadata_resolve::configuration::Configuration`, and known subgraphs
are now tracked in `MetadataAccessor`. If the flag is set and a
relationship target refers to an unknown subgraph, we return an empty
list of relationships instead of failing.
Some test infrastructure has been added to set configuration flags per
test.
V3_GIT_ORIGIN_REV_ID: 6f0de2442a3bfc7c7a4c48e3dc7296dc1538cd67
### What
This PR adds the ability to include internal errors in API responses via
a command line argument `--expose-internal-errors`.
The default behavior remains not to show the contents of internal error
messages.
V3_GIT_ORIGIN_REV_ID: 11c47286d3fbceeda71df3a224853633aeea8902
<!-- The PR description should answer 2 (maybe 3) important questions:
-->
### What
A scalar `BooleanExpressionType` lets us give operators fancy names.
This PR makes them actually work.
Functional no-op as this feature is behind a feature flag.
<!-- What is this PR trying to accomplish (and why, if it's not
obvious)? -->
<!-- Consider: do we need to add a changelog entry? -->
### How
<!-- How is it trying to accomplish it (what are the implementation
steps)? -->
V3_GIT_ORIGIN_REV_ID: b49ef95d3d6672a1e27371fa5f4df63acd0849fc
### What
There was a report that using a different argument name other than
`headers` for `DataConnectorLink.argumentPresets` doesn't work. This PR
updates the test to use a different name, and this seems to work.
### How
Update the headers argument name in custom connector schema, and the
argument name in `DataConnectorLink.argumentPresets` metadata.json.
V3_GIT_ORIGIN_REV_ID: 91346c01573f666e5707c0f41f4635d689bf5b98
### What
This removes an unused field. The remaining struct had just one field
and is private to the crate, so I converted it to a type alias.
### How
In order to make `rustc` highlight the issue, I reduced the number of
types we export as `pub`, changing some to `pub(crate)`.
Then I just deleted the field once the warning showed up.
V3_GIT_ORIGIN_REV_ID: 7a26e99f062ed0f2c7449e1f57bc76068f059afb
This PR fixes the version of the NDC Spec that Open DD's JSON schema
references when it uses NDC's schema and capabilities types in the
`DataConnectorLink`. OpenDD references 0.1.4 of NDC, but uses 0.1.3 in
the JSON schema. This corrects that.
V3_GIT_ORIGIN_REV_ID: bdbb417b3227861dae7835f6d3bda0d1bf935ea7
### What
Using `String` everywhere to represent subgraph identifiers is
definitely going to cause problems at some point. We can avoid this by
using `open_dds::identifier::SubgraphIdentifier` instead.
`Identifier` and `SubgraphIdentifier` were wrappers around `String`. I
have also changed them to wrap `SmolStr` instead, for better performance
as they're cloned a lot.
### How
First of all, I hid the innards of `Identifier` and
`SubgraphIdentifier`, instead exposing certain behaviors as methods.
Mostly, this means exposing `as_str()` and `to_string()`.
Then I replaced the internals with `SmolStr`. I had to change one place
where `RefCast` was used, but that was pretty much it.
Finally, I switched out `String` for `SubgraphIdentifier` in
`QualifiedObject`. This necessitated a couple of constants for the two
"magic" subgraphs, `__globals` and `__unknown_namespace`, but was
otherwise fine.
I didn't touch `Qualified` for now.
V3_GIT_ORIGIN_REV_ID: 28664609c3173b181c3789093cb9796896642eb7
### What
The `lazy_static` macro is poorly maintained, fairly bloated, and has
been mostly superseded by
[`OnceLock`](https://doc.rust-lang.org/stable/std/sync/struct.OnceLock.html)
in the stdlib.
### How
1. I turned a couple of `static ref` values into `const`, sometimes by
creating `const fn` equivalents to other functions.
2. I inlined static behavior to construct a JSON pointer into some
tests, where we don't care too much about losing a few milliseconds.
3. For the rest, I replaced `lazy_static` with a `static OnceLock` and a
call to `OnceLock::get_or_init`.
V3_GIT_ORIGIN_REV_ID: 18e4150a5fb24fe71f6ed77fe6178b7942405aa3
NOTE: This PR is stacked on #756 and should be shipped after that is
merged.
This PR enables the existing aggregate relationships work (see #725,
#731, #756) by default by removing the experimental flag it used to be
disabled behind.
The new OpenDD schema changes that were added are also unhidden so that
they are visible in the OpenDD JSON Schema.
V3_GIT_ORIGIN_REV_ID: cfd86d8a9ea61887ccf0f1a5d08bdcc3dda59cdc
## Description
This PR implements the GraphQL schema and execution for aggregate
relationships.
In the `schema` crate, the new `model_aggregate_relationship_field`
function handles generating schema for ModelAggregateTarget
relationships. It mostly delegates the meat of its implementation to
reused logic; some refactoring has occurred to make this possible.
This involved changes in `select_many`, `select_aggregate` and
`model_arguments`. The creation of the model arguments field argument
now exists in `model_arguments` and is reused by `select_many` and
`select_aggregate`. The creation of all aggregate field arguments is now
in `select_aggregate::generate_select_aggregate_arguments`, and is then
reused when generating the aggregate relationship field. That field is
annotated with the new `RelationshipToModelAggregate` annotation.
In the `execute` crate, the logic around generating an the aggregate
selection IR was moved from `select_aggregate` into `model_selection`.
This was so it can be reused by the logic in `relationship` that now
uses it to generate an aggregate selection when encountering an
`RelationshipToModelAggregate` field.
Inside `relationship` some rearranging was done so that
`build_local_model_relationship` and `build_remote_relationship` could
work with either a normal model selection IR or the new aggregate
selection IR. The necessitated moving the creation of that IR outside
those functions into the caller, so the different callers can create
different IR (normal vs aggregate IR). This also reduced code
duplication.
New tests have been added to `engine` that cover aggregate relationships
and also remote joined aggregate relationships.
This PR also corrects two bugs in metadata resolve revealed by new
testing:
* The filter input field name in `GraphqlConfig` must be specified if
using an aggregate relationship
* The filter input type name defined on a `Model` must be specified if
that model is the target of an aggregate relationship. Conversely, the
filter input type name can be specified if the `Model` itself doesn't
define an aggregate, but is still involved in a aggregate relationship
(this previously produced an error).
This PR completes the feature, but it is still hidden behind the
experimental flag. There will be a follow up PR to remove that and
expose the functionality by default.
JIRA: [V3ENGINE-160](https://hasurahq.atlassian.net/browse/V3ENGINE-160)
[V3ENGINE-160]:
https://hasurahq.atlassian.net/browse/V3ENGINE-160?atlOrigin=eyJpIjoiNWRkNTljNzYxNjVmNDY3MDlhMDU5Y2ZhYzA5YTRkZjUiLCJwIjoiZ2l0aHViLWNvbS1KU1cifQ
V3_GIT_ORIGIN_REV_ID: d499371906f7af71a4017c7c3ae75b7693cd3fa7
### What
In order to more easily monitor and review changes to metadata
resolution, this introduces snapshot testing for both successful and
failing calls to `resolve`. I used [Insta](https://insta.rs/) for this.
### How
For tests of the failure case, we already had a text file with the
expected error, so I have turned those files into snapshot files. I
wrote a small script to move the files rather than deleting and
recreating them so I could guarantee that the contents have not changed.
(Unfortunately, Git's diff doesn't always recognise the move as a move
because Insta has added a header.)
For tests of the successful case, I added a line to snapshot the
metadata rather than discarding it.
I also rewrote the tests to use `insta::glob` so we could get rid of
`test_each`.
V3_GIT_ORIGIN_REV_ID: 41bef4cf77bddb8d20d7c101df52ae149e8b0476
