Light
Dark
System

1.0 RC 2

This changelog summarizes new features and breaking changes in EdgeDB 1.0 Release Candidate 2 “Lacaille”.

We continue fixing bugs that affect schema definitions and migrations. The most notable changes involve static cardinality inference, so that computed links and properties can be defined to mimic their regular counterparts:

  • Fix creation of required computed links (#2985).

    This allows having required links in the schema even if they are computed:

    Copy
    type User {
        required link profile -> Profile;
    }
    
    type Profile {
        required link user := assert_exists(.<profile[is User]);
    }
  • Fix issues when migrating between regular and computed links and properties (#2411).

  • Fix issues when changing cardinality of computed links and properties (#3003).

  • Fix how alter suggestions are rejected when processing migrations, making sure that they don’t repeat (#3111).

  • Make static analysis correctly handle the cardinality of things wrapped in enumerate() (#3014).

    This means that if you wrap a single property into enumerate() it still is correctly inferred to produce no more than one result:

    Copy
    db> 
    select User {comp := enumerate(.name)};
    {default::User {comp: (0, 'alice')}}
  • Fix some issues with enumerate() applied to the results of another function (#3025).

  • Fix assert_distinct() on tuples (#2990).

  • Fix a case where for over an empty set erroneously produced a non-empty result (#3012).

  • Fix exclusive constraint issue when conflicting entries are inserted at the same time (#3022).

  • Fix issues with how tuple elements are correlated (#3001).

  • Fix how functions or operators with optional parameters such as ?? interact with shape subqueries (#3008).

    This makes it possible to have a shape as one element of a tuple to be correlated with the expression in the other tuple element:

    Copy
    db> 
    ... 
    ... 
    ... 
    select (
      User {name, friends: {name}},
      User.friends.name ?? 'n/a'
    );
    {
      (
        default::User {
          name: 'Alice',
          friends: {default::User {name: 'Billie'}},
        },
        'Billie',
      ),
      (
        default::User {
          name: 'Alice',
          friends: {default::User {name: 'Cameron'}},
        },
        'Cameron',
      ),
      (
        default::User {
          name: 'Alice',
          friends: {default::Bot {name: 'Dana'}},
        },
        'Dana',
      ),
      (default::User {name: 'Billie', friends: {}}, 'n/a'),
      (default::User {name: 'Cameron', friends: {}}, 'n/a'),
      (
        default::Bot {
          name: 'Dana',
          friends: {default::User {name: 'Billie'}},
        },
        'Billie',
      ),
    }
  • Fix issues with arrays of objects of related types (#2256).

    Trying to build an array where elements are not of the same object type, but of related types no longer produces an error:

    Copy
    db> 
    select [(select User filter .name = 'alice'), (select SystemUser)];
    {
      [
        default::User {id: 8f69777e-3129-11ec-ba91-0f55d65fd8d7},
        default::SystemUser {id: 89c0e596-3129-11ec-ba91-2f631728aea3},
      ],
    }

    Conceptually, there’s no difference between first creating a set via union and using array_agg() on it or using an array constructor directly and supply the individual elements as subqueries.

  • Fix many bugs with objects inside arrays and tuples (#2992).

    Packing and unpacking objects into arrays and tuples now works more reliably. You can make use of the different way of selecting the data, aggregate it using array_agg() or by some other means and still be able to access the elements to get their nested contents in queries:

    Copy
    db> 
    select [(User,)][0];
    {
      (default::User {id: 8f69777e-3129-11ec-ba91-0f55d65fd8d7}),
    }
    Copy
    db> 
    select ([User],).0;
    {
      [default::User {id: 8f69777e-3129-11ec-ba91-0f55d65fd8d7}],
    }
  • Fix some issues with ad-hoc computed links or properties inside arrays (#2979).

  • Fix and to consistently apply to properties and produce an {} even for the case of False and {} (#3121).

  • Restrict | operator to object types only (#3116).

  • Reserve never as a keyword for future use (#3102).

  • Make and and or in GraphQL use “short-circuiting” logic.

    In EdgeQL and and or produce {} if any of the operands are {}. There are different ways of handling the {} by using ?? or if..else operator. However, within the restricted context of operations reflected to GraphQL filter it makes sense to use “short-circuiting” versions of the operators and enforce that if any of the operands to or are true, so is the result and similarly if any of the operands to and are false so is the result.

  • Hide id as well as computed links and properties from mutation (#3109).

A lot of the configuration settings used to be strings that simply mapped to Postgres settings. We now use more appropriate types for expressing configuration values: bool, int64, duration, in addition to str.

We also introduce a new scalar cfg::memory to represent size of various memory storage in a clear and unambiguous way.

We now have the client abstraction superceding the connection and pool abstractions. There’s no conceptual difference between a pool of size 1 and a single connection, so we decided that the API should reflect that. So we introduce a single concept of client to send queries to the database and let this client encapsulate the handling of concurrent connections.

We’re making our binding more robust by adding “retry options” to our interface. What it means is that read-only queries and all transactions can be automatically retried in case of certain kinds of errors (such as network errors) without the need for any extra code on the part of the developers. This also means that we no longer have “raw” and “retrying” transactions in the APIs, but all transactions are retrying. To get the same behavior as for “raw” transactions the maximum number of attempts can be explicitly specified as 1.

We’re also in the process of cleaning up our APIs and removing the deprecated functions.

  • Support connection pooling by default and retry features for edgedb-python and release v0.18.0a2 driver.

  • Support connection pooling by default and retry features for edgedb-js and release v0.15.3 driver.

  • Support connection pooling by default and retry features for edgedb-go and release v0.8.3 driver.

For clarity and simplicity, several environment variables have been renamed and the set of supported values has been modified. The old variables have been deprecated and will be removed in a future release.

Old name

New name

EDGEDB_TLS_VERIFY_HOSTNAME

EDGEDB_CLIENT_TLS_SECURITY

EDGEDB_NO_VERSION_CHECK

EDGEDB_RUN_VERSION_CHECK

EDGEDB_SKIP_DOCKER_CHECK

EDGEDB_INSTALL_IN_DOCKER

EDGEDB_SERVER_ALLOW_INSECURE_BINARY_CLIENTS

EDGEDB_SERVER_BINARY_ENDPOINT_SECURITY

EDGEDB_SERVER_ALLOW_INSECURE_HTTP_CLIENTS

EDGEDB_SERVER_HTTP_ENDPOINT_SECURITY

EDGEDB_SERVER_INSECURE_DEV_MODE

EDGEDB_SERVER_SECURITY

EDGEDB_SERVER_SKIP_MIGRATIONS

EDGEDB_DOCKER_APPLY_MIGRATIONS

EDGEDB_SERVER_GENERATE_SELF_SIGNED_CERT

EDGEDB_SERVER_TLS_CERT_MODE

EDGEDB_HIDE_GENERATED_CERT

EDGEDB_DOCKER_SHOW_GENERATED_CERT

EDGEDB_CLIENT_TLS_SECURITY
  • insecure

  • no_host_verification

  • strict

  • default: no_host_verification if custom certificate is supplied, otherwise strict

New: EDGEDB_CLIENT_SECURITY
  • default

  • insecure_dev_mode: for now, equivalent to setting EDGEDB_CLIENT_TLS_SECURITY=insecure though this mode may encapsulate other behavior later

EDGEDB_RUN_VERSION_CHECK
  • cached == default — run occasionally

  • never — skips the check

EDGEDB_INSTALL_IN_DOCKER (CLI only)
  • forbid == default

  • allow — skips the check

EDGEDB_SERVER_BINARY_ENDPOINT_SECURITY
  • tls == default

  • optional — allow no TLS

EDGEDB_SERVER_HTTP_ENDPOINT_SECURITY
  • tls == default

  • optional — allow no TLS

EDGEDB_SERVER_SECURITY
  • strict == default

  • insecure_dev_mode — disable password-based authentication and allow unencrypted HTTP traffic

EDGEDB_DOCKER_APPLY_MIGRATIONS (Docker only)
  • always == default

  • never

EDGEDB_SERVER_TLS_CERT_MODE
  • require_file — requires a valid TLS certificate and key to be specified

  • generate_self_signed generate self-signed certificate and private key on bootstrap if certificate or key are not specified or missing

  • default (equals to require_file if EDGEDB_SERVER_SECURITY is set to strict, equals to generate_self_signed if EDGEDB_SERVER_SECURITY is set to insecure_dev_mode)

EDGEDB_DOCKER_SHOW_GENERATED_CERT
  • always == default

  • never

Light
Dark
System