- Allow @_required directive to be used in combination with configured schemas Submitted by @AndrewIngram (See #3685)
⚠️ fix bug that mutation would cause dependent operations and reexecuting operations to become the same set Submitted by @xuanduc987 (See #3665)
- Disregard write-only operation when fragment-matching with schema awareness Submitted by @JoviDeCroock (See #3621)
⚠️ Fix where we would incorrectly match all fragment concrete types because they belong to the abstract type Submitted by @JoviDeCroock (See #3603)
- Mark
@urql/core
as a peer dependency as well as a regular dependency Submitted by @kitten (See #3579)
- Only record dependencies that are changing data, this will reduce the amount of operations we re-invoke due to network-only/cache-and-network queries and mutations Submitted by @JoviDeCroock (See #3564)
- When invoking the automatic creation updater ignore the entity we are currently on in the mutation Submitted by @JoviDeCroock (See #3560)
- Add a default updater for mutation fields who are lacking an updater and where the returned entity is not present in the cache Submitted by @JoviDeCroock (See #3518)
- Remove deprecated
resolveFieldByKey
, usecache.resolve
instead Submitted by @JoviDeCroock (See #3520)
- Track abstract types being written so that we have a more reliable way of matching abstract fragments Submitted by @JoviDeCroock (See #3548)
⚠️ Fixinvalidate
not applying when using a string to invalidate an entity Submitted by @JoviDeCroock (See #3545)- Upgrade
@0no-co/graphql.web
to1.0.5
Submitted by @kitten (See #3553) - Updated dependencies (See #3520, #3553, and #3520)
- @urql/[email protected]
- Allow
@_optional
and@_required
to be placed on fragment definitions and inline fragments Submitted by @JoviDeCroock (See #3502) - Track list of entity keys for a given type name. This enables enumerating and invalidating all entities of a given type within the normalized cache Submitted by @JoviDeCroock (See #3501)
- Prevent
@defer
from being applied in child field selections. Previously, a child field (i.e. a nested field) under a@defer
-ed fragment would also become optional, which was based on a prior version of the DeferStream spec which didn't require deferred fields to be delivered as a group Submitted by @kitten (See #3517) ⚠️ Fixstore.resolve()
returning the exact link array that’s used by the cache. This can lead to subtle bugs when a user mutates the result returned bycache.resolve()
, since this directly mutates what’s in the cache at that layer Submitted by @kitten (See #3516)- Updated dependencies (See #3514, #3505, #3499, and #3515)
- @urql/[email protected]
- Set
stale: true
on cache results, even if a reexecution has been blocked by the loop protection, if the operation is already pending and in-flight Submitted by @kitten (See #3493) ⚠️ Fix@defer
state leaking into following operations Submitted by @kitten (See #3497)
- Allow the user to debug cache-misses by means of the new
logger
interface on thecacheExchange
. A field miss will dispatch adebug
log when it's not marked with@_optional
or when it's non-nullable in theschema
Submitted by @JoviDeCroock (See #3446) - Add
onCacheHydrated
as an option for theStorageAdapter
Submitted by @JoviDeCroock (See #3428) - Add optional
logger
to the options, this allows you to filter out warnings or disable them all together Submitted by @JoviDeCroock (See #3444)
⚠️ Fix a typo that caused an inverted condition, for checking owned data, to cause incorrect results when handlingnull
values and encountering them first Submitted by @kitten (See #3371)
⚠️ Fix extra variables in mutation results regressing by a change made in #3317. The original operation wasn't being preserved anymore Submitted by @kitten (See #3356)
- Reset
partial
result marker when reading from selections when a child value sees a cache miss. This only affects resolvers on child values enablinginfo.partial
while a parent may abort early instead Submitted by @kitten (See #3340) ⚠️ Fix@_optional
directive not settinginfo.partial = true
on cache miss and fix usage ofinfo.parentKey
andinfo.parentFieldKey
usage in default directives Submitted by @kitten (See #3338)- Replace implementation for
@_optional
and@_required
with built-in handling inside cache reads to allow@_optional
to work for nested selection sets Submitted by @kitten (See #3341)
- Allow scalar values on the parent to be accessed from
parent[info.fieldName]
consistently. Prior to this changeparent[fieldAlias]
would get populated, which wouldn’t always result in a field that’s consistently accessible Submitted by @kitten (See #3336) - Allow
cache.resolve
to returnundefined
when a value is not cached to make it easier to cause a cache miss in resolvers. Reminder: Returningundefined
from a resolver means a field is uncached, while returningnull
means that a field’s value isnull
without causing a cache miss Submitted by @kitten (See #3333)
- Record a dependency when
__typename
field is read. This removes a prior, outdated exception to avoid confusion when usingcache.resolve(entity, '__typename')
which doesn't cause the cache to record a dependency Submitted by @kitten (See #3335) ⚠️ Fix cases whereResolveInfo
’sparentFieldKey
was incorrectly populated with a key that isn’t a field key (allowing forcache.resolve(info.parentKey, info.parentFieldKey)
to be possible) but was instead set toinfo.parentKey
combined with the field key Submitted by @kitten (See #3336)
- Implement local directives. It’s now possible to add client-only directives to queries by adding them to the
cacheExchange
’s newdirectives
option. Directives accept an object of their arguments and return a resolver. When a field is annotated with a resolver, e.g.@_optional
or@_required
, their resolvers from thedirectives
config are executed. This means it’s now possible to use@_relayPagination
for example, by passing adding therelayPagination
helper to the config. Due to the change in #3317, any directive in queries that’s prefixed with an underscore (_
) is only visible to Graphcache and not the API. Submitted by undefined (See #3306)
- Use new
FormattedNode
/formatDocument
functionality added to@urql/core
to slightly speed up directive processing by using the client-side_directives
dictionary thatformatDocument
adds Submitted by @kitten (See #3317) - Allow
offlineExchange
to once again issue all request policies, instead of mapping them tocache-first
. When replaying operations after rehydrating it will now prioritise network policies, and before rehydrating receiving a network result will prevent a network request from being issued again Submitted by @kitten (See #3308) - Add
OperationContext.optimistic
flag as an internal indication on whether a mutation triggered an optimistic update in@urql/exchange-graphcache
'scacheExchange
Submitted by @kitten (See #3308) - Updated dependencies (See #3317 and #3308)
- @urql/[email protected]
⚠️ Fix untranspiled class property initializer syntax being leftover in build output. (Regression in #3053) Submitted by @kitten (See #3275)
⚠️ Fixinfo.parentKey
not being correctly set for updaters or optimistic updaters Submitted by @kitten (See #3267)
- Make "Invalid undefined" warning heuristic smarter and allow for partial optimistic results. Previously, when a partial optimistic result would be passed, a warning would be issued, and in production, fields would be deleted from the cache. Instead, we now only issue a warning if these fields aren't cached already Submitted by @kitten (See #3264)
- Optimistic mutation results should never result in dependent operations being blocked Submitted by @kitten (See #3265)
⚠️ Fix torn down queries not being removed fromofflineExchange
’s failed queue on rehydration Submitted by @kitten (See #3236)
- Update build process to generate correct source maps Submitted by @kitten (See #3201)
- Prevent
offlineExchange
from issuing duplicate operations Submitted by @kitten (See #3200) ⚠️ Fix reference equality not being preserved. This is a fix on top of #3165, and was previously not addressed to avoid having to test for corner cases that are hard to cover. If you experience issues with this fix, please let us know Submitted by @kitten (See #3228)- Retry operations against offline cache and stabilize timing of flushing failed operations queue after rehydrating the storage data Submitted by @kitten (See #3196)
⚠️ Fix missing cache updates, when a query that was previously torn down restarts and retrieves results from the cache. In this case a regression caused cache updates to not be correctly applied to the queried results, since the operation wouldn’t be recognised properly Submitted by @kitten (See #3193)
- Prevent reusal of incoming API data in Graphcache’s produced (“owned”) data. This prevents us from copying the
__typename
and other superfluous fields Submitted by @kitten (See #3165) ⚠️ Fix regression which caused@defer
directives from becoming “sticky” and causing every subsequent cache read to be treated as if the field was deferred Submitted by @kitten (See #3167)- Apply
hasNext: true
and fallthrough logic to cached queries that contain deferred, uncached fields. Deferred query results will now be fetched against the API correctly, even if prior requests have been incomplete Submitted by @kitten (See #3163) ⚠️ FixofflineExchange
duplicating offline mutations in failed queue Submitted by @kitten (See #3158)
- Remove dependence on
graphql
package and replace it with@0no-co/graphql.web
, which reduces the default bundlesize impact ofurql
packages to a minimum. All types should remain compatible, even if you usegraphql
elsewhere in your app, and if other dependencies are usinggraphql
you may alias it tographql-web-lite
Submitted by @kitten (See #3097) - Update
OperationResult.hasNext
andOperationResult.stale
to be required fields. If you have a custom exchange creating results, you'll have to add these fields or use themakeResult
,mergeResultPatch
, ormakeErrorResult
helpers Submitted by @kitten (See #3061)
- Update exchanges to drop redundant
share
calls, since@urql/core
’scomposeExchanges
utility now automatically does so for us Submitted by @kitten (See #3082)
⚠️ Fix source maps included with recently published packages, which lost theirsourcesContent
, including additional source files, and had incorrect paths in some of them Submitted by @kitten (See #3053)- Upgrade to
wonka@^6.3.0
Submitted by @kitten (See #3104) - Restore variables correctly on mutations Submitted by @JoviDeCroock (See #3046)
- Use
stringifyDocument
inofflineExchange
rather thanprint
and serializeoperation.extensions
as needed Submitted by @kitten (See #3094) - Add missing
hasNext
andstale
passthroughs on caching exchanges Submitted by @kitten (See #3059) - Add TSDocs for all exchanges, documenting API internals Submitted by @kitten (See #3072)
- Updated dependencies (See #3101, #3033, #3054, #3053, #3060, #3081, #3039, #3104, #3082, #3097, #3061, #3055, #3085, #3079, #3087, #3059, #3055, #3057, #3050, #3062, #3051, #3043, #3063, #3054, #3102, #3097, #3106, #3058, and #3062)
- @urql/[email protected]
- Add
isOfflineError
option to theofflineExchange
to allow it to be customized to different conditions to determine whether an operation has failed because of a network error Submitted by @robertherber (See #3020) - Allow
updates
config to react to arbitrary type updates other than justMutation
andSubscription
fields. You’ll now be able to write updaters that react to any entity field being written to the cache, which allows for more granular invalidations. Note: If you’ve previously usedupdates.Mutation
andupdated.Subscription
with a custom schema with custom root names, you‘ll get a warning since you’ll have to update yourupdates
config to reflect this. This was a prior implementation mistake! Submitted by @kitten (See #2979)
⚠️ Fix regression which caused partial results, whose refetches were blocked by the looping protection, to not have astale: true
flag added to them. This is a regression from #2831 and only applies tocacheExchange
s that had theschema
option set Submitted by @kitten (See #2999)- Add
invariant
to data layer that prevents cache writes during cache query operations. This preventscache.writeFragment
,cache.updateQuery
, andcache.link
from being called inresolvers
for instance Submitted by @kitten (See #2978) - Updated dependencies (See #3007, #2962, #3007, #3015, and #3022)
- @urql/[email protected]
⚠️ Fix potential data loss inofflineExchange
that's caused whenonOnline
triggers and flushes mutation queue before the mutation queue is used, by @trcoffman (See #2945)- Patch message for
(16) Heuristic Fragment Matching
, by @inokawa (See #2923) - Patch message for (19) Can't generate a key for invalidate(...) error, by @inokawa (See #2918)
⚠️ Fix operation being blocked for looping due to it not cancelling the looping protection when ateardown
is received. This bug could be triggered when a shared query operation triggers again and causes a cache miss (e.g. due to an error). The re-execution of the operation would then be blocked as Graphcache considered it a "reexecution loop" rather than a legitimate execution triggered by the UI. (See #2737 for more information), by @kitten (See #2876)
⚠️ Fix type-generation, with a change in TS/Rollup the type generation took the paths as src and resolved them into the types dir, by @JoviDeCroock (See #2870)- Updated dependencies (See #2872, #2870, and #2871)
- @urql/[email protected]
- Solve issue where partial data could cause loops between related queries, by @JoviDeCroock (See #2831)
- Add skipping of garbage collection runs when the cache is waiting for optimistic, deferred or other results in layers. This means that we only take an opportunity to run garbage collection after results have settled and are hence decreasing the chance of hogging the event loop when a run isn't needed, by @kitten (See #2862)
⚠️ Fix a deadlock condition in Graphcache's layers, which is caused by subscriptions (or other deferred layers) starting before one-off mutation layers. This causes the mutation to not be completed, which keeps its data preferred above the deferred layer. That in turn means that layers stop squashing, which causes new results to be missing indefinitely, when they overlap, by @kitten (See #2861)- Updated dependencies (See #2843, #2847, #2850, and #2846)
- @urql/[email protected]
- Set operations when updating the cache with a result, by @JoviDeCroock (See #2782)
- Ensure we aren't eagerly removing layers that are caused by subscriptions, by @JoviDeCroock (See #2771)
⚠️ Fix case where a mutation would also be counted in the loop-protection, this prevented partial queries from initiating refetches, by @JoviDeCroock (See #2761)- Updated dependencies (See #2758 and #2762)
- @urql/[email protected]
- Preserve the original
DocumentNode
AST when updating the cache, to prevent results after a network request from differing and breaking referential equality due to added__typename
fields, by @kitten (See #2736) ⚠️ Fix optimistic mutations containing partial results (undefined
fields), which previously actually caused a hidden cache miss, which may then affect a subsequent non-optimistic mutation result, by @kitten (See #2740)- Prevent cache misses from causing infinite network requests from being issued, when two operations manipulate each other while experiencing cache misses or are partially uncacheable, by @kitten (See #2737)
⚠️ Fix operation identities preventing users from deeply cloning operation contexts. Instead, we now use a client-wide counter (rolling over as needed). While this changes an internal data structure in@urql/core
only, this change also affects theofflineExchange
in@urql/exchange-graphcache
due to it relying on the identity being previously an object rather than an integer, by @kitten (See #2732)⚠️ Fix referential equality preservation in Graphcache failing after API results, due to a typo writing the API result rather than the updated cache result, by @kitten (See #2741)- Updated dependencies (See #2691, #2692, and #2732)
- @urql/[email protected]
- Adjust timing of when an introspected schema will be processed into field maps, interface maps, and union type maps. By making this lazy we can avoid excessive work when these maps aren't actually ever used, by @kitten (See #2640)
- Goodbye IE11! 👋 This major release removes support for IE11. All code that is shipped will be transpiled much less and will not be ES5-compatible anymore, by @kitten (See #2504)
- Prevent cache-hydration from buffering operations, by @JoviDeCroock (See #2612)
- Implement stricter variables types, which require variables to always be passed and match TypeScript types when the generic is set or inferred. This is a breaking change for TypeScript users potentially, unless all types are adhered to, by @kitten (See #2607)
- Upgrade to Wonka v6 (
wonka@^6.0.0
), which has no breaking changes but is built to target ES2015 and comes with other minor improvements. The library has fully been migrated to TypeScript which will hopefully help with making contributions easier!, by @kitten (See #2504)
- Remove the
babel-plugin-modular-graphql
helper, this because the graphql package hasn't converted to ESM yet which gives issues in node environments, by @JoviDeCroock (See #2551) - Allow passing in
fragmentName
forwrite
andread
operations, by @JoviDeCroock (See #2609)
- Graphcache's
optimistic
option now accepts optimistic mutation resolvers that return fields by name rather than alias. Previously, depending on which mutation was run, the optimistic resolvers would read your optimistic data by field alias (i.e. "alias" foralias: id
rather than "id"). Instead, optimistic updates now correctly use field names and allow you to also pass resolvers as values on your optimistic config, by @JoviDeCroock (See #2616) - Updated dependencies (See #2551, #2504, #2619, #2607, and #2504)
- @urql/[email protected]
- Correctly reorder optimistic layers when we see repeated keys coming in, by @JoviDeCroock (See #2489)
- Keep track of mutations in the offline exchange so we can accurately recreate the original variables, there could be more variables for use in updater functions which we strip away in graphCache before sending to the API, by @JoviDeCroock (See #2472)
- Switch
isFragmentHeuristicallyMatching()
to always returntrue
for writes, so that we give every fragment a chance to be applied and to write to the cache, by @kitten (See #2455) ⚠️ Fix default storage persisting data afterclear()
was called on it, by @JoviDeCroock (See #2458)- Updated dependencies (See #2446, #2456, and #2457)
- @urql/[email protected]
- Fix issues with continuously updating operations (i.e. subscriptions and
hasNext: true
queries) by shifting their layers in Graphcache behind those that are still awaiting a result. This causes continuous updates to not overwrite one-off query results while still keeping continuously updating operations at the highest possible layer, by @kitten (See #2419)
⚠️ Fix ignore empty relay edges when using therelayPagination
resolver, by @tgriesser (See #2431)- Prevent creating unnecessary layers, which should improve performance slightly, by @kitten (See #2419)
- Extend peer dependency range of
graphql
to include^16.0.0
. As always when upgrading across many packages ofurql
, especially including@urql/core
we recommend you to deduplicate dependencies after upgrading, usingnpm dedupe
ornpx yarn-deduplicate
, by @kitten (See #2133) - Updated dependencies (See #2133)
- @urql/[email protected]
⚠️ Fix regression from #1869 that caused nullable lists to always cause a cache miss, if schema awareness is enabled, by @kitten (See #1983)- Updated dependencies (See #1985)
- @urql/[email protected]
- Improve perf by using String.indexOf in getField, by @JoviDeCroock (See #1957)
- Updated dependencies (See #1944)
- @urql/[email protected]
- Remove
hasNext: true
flag from stale responses. This was erroneously added in debugging, but leads to stale responses being marked withhasNext
, which means thededupExchange
will keep waiting for further network responses, by @kitten (See #1911)
- Cleanup the previous
onOnline
event-listener when called again, by @JoviDeCroock (See #1896)
⚠️ Fix previous results'null
values spilling into the next result that Graphcache issues, which may prevent updates from being issued until the query is reexecuted. This was affecting anynull
links on data, and any queries that were issued before non-optimistic mutations, by @kitten (See #1885)- Updated dependencies (See #1870 and #1880)
- @urql/[email protected]
- Improve referential equality of deeply queried objects from the normalised cache for queries. Each query operation will now reuse the last known result and only incrementally change references as necessary, scanning over the previous result to identify whether anything has changed.
This should help improve the performance of processing updates in UI frameworks (e.g. in React with
useMemo
orReact.memo
). (See #1859) - Add experimental support for
@defer
and@stream
responses for GraphQL. This implements the "GraphQL Defer and Stream Directives" and "Incremental Delivery over HTTP" specifications. If a GraphQL API supportsmultipart/mixed
responses for deferred and streamed delivery of GraphQL results,@urql/core
(and all its derived fetch implementations) will attempt to stream results. This is only supported on browsers supporting streamed fetch responses, which excludes IE11. The implementation of streamed multipart responses is derived frommeros
by@maraisr
, and is subject to change if the RFCs end up changing, by @kitten (See #1854)
⚠️ Fix missing values cascading into lists causing anull
item without the query being marked as stale and fetched from the API. This would happen in schema awareness when a required field, which isn't cached, cascades into a nullable list, by @kitten (See #1869)- Updated dependencies (See #1854)
- @urql/[email protected]
⚠️ Fix issue where operations that get dispatched synchronously after the cache restoration completes get forgotten, by @JoviDeCroock (See #1789)
- Type the
relayPagination
andsimplePagination
helpers return value asResolver<any, any, any>
as there's no way to match them consistently to either generated or non-generated resolver types anymore, by @kitten (See #1778) - Updated dependencies (See #1776 and #1755)
- @urql/[email protected]
- Apply
bivarianceHack
in thegraphcache
types to better support code-generated configs, by @JoviDeCroock (See #1687) - Updated dependencies (See #1709)
- @urql/[email protected]
⚠️ Fix: add theENTRIES_STORE_NAME
to the clear transaction, by @JoviDeCroock (See #1685)- Updated dependencies (See #1695)
- @urql/[email protected]
- Loosen type constraint on
ScalarObject
to account for custom scalar deserialization likeDate
forDateTime
s, by @kitten (See #1648) - Loosen the typing constraint on the cacheExchange generic, by @JoviDeCroock (See #1675)
⚠️ Fix an edge-case for which an introspection query during runtime could fail when schema-awareness was enabled in Graphcache, since built-in types weren't recognised as existent, by @kitten (See #1631)
- Add
cache.link(...)
method to Graphcache. This method may be used in updaters to update links in the cache. It is hence the writing-equivalent ofcache.resolve()
, which previously didn't have any equivalent as such, which meant that onlycache.updateQuery
orcache.writeFragment
could be used, even to update simple relations, by @JoviDeCroock (See #1551) - Add on a generic to
cacheExchange
andofflineExchange
for future, experimental type-generation support, by @JoviDeCroock (See #1562)
⚠️ Fix up internal types in Graphcache to improve their accuracy for catching more edge cases in its implementation. This only affects you if you previously imported any type related toScalarObject
from Graphcache which now is a more opaque type. We've also adjusted theNullArray
types to be potentially nested, since lists in GraphQL can be nested arbitarily, which we were covering but didn't reflect in our types, by @kitten (See #1591)- Remove closure-compiler from the build step (See #1570)
⚠️ Fix list items being returned asnull
even for non-nullable lists, when the entities are missing in the cache. This could happen when a resolver was added returning entities or their keys. This behaviour is now (correctly) only applied to partial results with schema awareness, by @JoviDeCroock (See #1566)- Allow for the schema subscription and mutationType to be null, by @JoviDeCroock (See #1530)
- Updated dependencies (See #1570, #1509, #1600, and #1515)
- @urql/[email protected]
- Add improved error awareness to Graphcache. When Graphcache now receives a
GraphQLError
(via aCombinedError
) it checks whether theGraphQLError
'spath
matches up withnull
values in thedata
. Anynull
values that the write operation now sees in the data will be replaced with a "cache miss" value (i.e.undefined
) when it has an associated error. This means that errored fields from your GraphQL API will be marked as uncached and won't be cached. Instead the client will now attempt a refetch of the data so that errors aren't preventing future refetches or with schema awareness it will attempt a refetch automatically. Additionally, theupdates
functions will now be able to check whether the current field has any errors associated with it withinfo.error
, by @kitten (See #1356)
- Allow
schema
option to be passed with a partial introspection result that only containsqueryType
,mutationType
, andsubscriptionType
with their respective names. This allows you to pass{ __schema: { queryType: { name: 'Query' } } }
and the likes to Graphcache'scacheExchange
to alter the default root names without enabling full schema awareness, by @kitten (See #1379)
- Updated dependencies (See #1374, #1357, and #1375)
- @urql/[email protected]
- Warn when using an interface or union field in the graphCache resolvers config, by @JoviDeCroock (See #1304)
⚠️ Fix edge-case where query results would pick up invalidated fields from mutation results as they're written to the cache. This would cause invalid cache misses although the result was expected to just be passed through from the API result, by @kitten (See #1300)⚠️ Fix a Relay Pagination edge case where overlapping ends of pages queried using thelast
argument would be in reverse order, by @JoviDeCroock (See #1311)
⚠️ Fix, add null as a possible type for the variables argument incache.invalidate
, by @JoviDeCroock (See #1269)
- Update
cache.resolve(parent, ...)
case to enable even more cases, for instance whereparent.__typename
isn't set yet. This was intended to be enabled in the previous patch but has been forgotten, by @kitten (See #1219) - Deprecate
cache.resolveFieldByKey
in favour ofcache.resolve
, which functionally was already able to do the same, by @kitten (See #1219) - Updated dependencies (See #1225)
- @urql/[email protected]
- Update
cache
methods, for instancecache.resolve
, to consistently accept theparent
argument fromresolvers
andupdates
and alias it to the parent's key (which is usually found oninfo.parentKey
). This usage ofcache.resolve(parent, ...)
was intuitive and is now supported as expected, by @kitten (See #1208)
⚠️ Fix reusing original query data from APIs accidentally, which can lead to subtle mismatches in results when the API's incomingquery
results are being updated by thecacheExchange
, to apply resolvers. Specifically this may lead to relations from being set back tonull
when the resolver returns a different list of links than the result, since somenull
relations may unintentionally exist but aren't related. If you're usingrelayPagination
then this fix is critical, by @kitten (See #1196)
- Increase the consistency of when and how the
__typename
field is added to results. Instead of adding it by default and automatically first, the__typename
field will now be added along with the usual selection set. Thewrite
operation now automatically issues a warning if__typename
isn't present where it's expected more often, which helps in debugging. Also the__typename
field may now not proactively be added to root results, e.g."Query"
, by @kitten (See #1185)
- Replace
graphql/utilities/buildClientSchema.mjs
with a custom-tailored, lighter implementation built into@urql/exchange-graphcache
. This will appear to increase its size by about0.2kB gzip
but will actually save around8.5kB gzip
to9.4kB gzip
in any production bundle by using less ofgraphql
's code, by @kitten (See #1189) - Updated dependencies (See #1187, #1186, and #1186)
- @urql/[email protected]
- Add a
mergeMode: 'before' | 'after'
option to thesimplePagination
helper to define whether pages are merged before or after preceding ones when pagination, similar torelayPagination
's option, by @hoangvvo (See #1174)
- Updated dependencies (See #1168)
- @urql/[email protected]
- Add support for
TypedDocumentNode
to infer the type of theOperationResult
andOperation
for all methods, functions, and hooks that either directly or indirectly accept aDocumentNode
. Seegraphql-typed-document-node
and the corresponding blog post for more information., by @kitten (See #1113) - Updated dependencies (See #1119, #1113, #1104, and #1123)
- @urql/[email protected]
⚠️ Fix a strayoperationName
deprecation warning in@urql/exchange-graphcache
's exchange logic, which adds themeta.cacheOutcome
field to the operation's context, by @kitten (See #1103)
⚠️ Fix the production build overwriting the development build. Specifically in the previous release we mistakenly replaced all development bundles with production bundles. This doesn't have any direct influence on how these packages work, but prevented development warnings from being logged or full errors from being thrown, by @kitten (See #1097)- Updated dependencies (See #1097)
- @urql/[email protected]
- Add missing
.mjs
extension to all imports fromgraphql
to fix Webpack 5 builds, which require extension-specific import paths for ESM bundles and packages. This change allows you to safely upgrade to Webpack 5., by @kitten (See #1094) - Deprecate the
Operation.operationName
property in favor ofOperation.kind
. This name was previously confusing asoperationName
was effectively referring to two different things. You can safely upgrade to this new version, however to mute all deprecation warnings you will have to upgrade allurql
packages you use. If you have custom exchanges that spread operations, please use the newmakeOperation
helper function instead, by @bkonkle (See #1045) - Updated dependencies (See #1094 and #1045)
- @urql/[email protected]
- Enforce atomic optimistic updates so that optimistic layers are cleared before they're reapplied. This is important for instance when an optimistic update is performed while offline and then reapplied while online, which would previously repeat the optimistic update on top of its past data changes, by @kitten (See #1080)
⚠️ Fix optimistic updates not being allowed to be cumulative and apply on top of each other. Previously in #866 we explicitly deemed this as unsafe which isn't correct anymore given that concrete, non-optimistic updates are now never applied on top of optimistic layers, by @kitten (See #1074)
- Changes some internals of how selections are iterated over and remove some private exports. This will have no effect or fixes on how Graphcache functions, but may improve some minor performance characteristics of large queries, by @kitten (See #1060)
⚠️ Fix inline fragments being skipped when they were missing a full type condition as per the GraphQL spec (e.g{ ... { field } }
), by @kitten (See #1040)
⚠️ Fix a case where theofflineExchange
would not start processing operations after hydrating persisted data when no operations arrived in time by the time the persisted data was restored. This would be more evident in Preact and Svelte due to their internal short timings, by @kitten (See #1019)
⚠️ Fix small pieces of code where polyfill-less ES5 usage was compromised. This was unlikely to have affected anyone in production asArray.prototype.find
(the only usage of an ES6 method) is commonly used and polyfilled, by @kitten (See #991)⚠️ Fix queries that have erroed with aNetworkError
(isOfflineError
) not flowing back completely through thecacheExchange
. These queries should also now be reexecuted when the client comes back online, by @kitten (See #1011)- Updated dependencies (See #1011)
- @urql/[email protected]
⚠️ Fix updaters config not working when Mutation/Subscription root names were altered. For instance, a Mutation namedmutation_root
could causestore.updates
to be misread and cause a runtime error, by @kitten (See #984)⚠️ Fix operation results being obstructed by theofflineExchange
when the network request has failed due to being offline and no cache result has been issued. Instead theofflineExchange
will now retry withcache-only
policy, by @kitten (See #985)
- Add support for
nodes
fields to therelayPagination
helper, instead of only supporting the standardedges
. (See #897)
- Updated dependencies (See #947, #962, and #957)
- @urql/[email protected]
- Add special-case for fetching an introspection result in our schema-checking, this avoids an error when urql-devtools fetches the backend graphql schema, by @JoviDeCroock (See #893)
- Mute warning when using built-in GraphQL fields, like
__type
, by @kitten (See #919) ⚠️ Fix return type for resolvers to allow data objects to be returned with__typename
as expected, by @JoviDeCroock (See #927)- Updated dependencies (See #911 and #908)
- @urql/[email protected]
- Add warning for queries that traverse an Operation Root Type (Mutation / Subscription types occuring in a query result), by @kitten (See #859)
⚠️ Fix storage implementation not preserving deleted values correctly or erroneously checking optimistically written entries for changes. This is fixed by adding a new default serializer to the@urql/exchange-graphcache/default-storage
implementation, which will be incompatible with the old one, by @kitten (See #866)- Replace unnecessary
scheduleTask
polyfill with inlinePromise.resolve().then(fn)
calls, by @kitten (See #861) - Updated dependencies (See #860 and #861)
- @urql/[email protected]
This major release comes with a couple of fixes and new experimental offline support, which we're very excited for! Please give it a try if your application is targeting Offline First!
To migrate to this new major version, check the major breaking changes below. Mainly you will have
to watch out for cache.invalidateQuery
which has been removed. Instead you should now invalidate
individual entities and fields using cache.invalidate
. Learn more about this method on our
docs.
- Remove the deprecated
populateExchange
export from@urql/exchange-graphcache
. If you're using thepopulateExchange
, please install the separate@urql/exchange-populate
package and import it from there, by @kitten (See #840) - The deprecated
cache.invalidateQuery()
method has been removed. Please migrate over tocache.invalidate()
instead, which operates on individual fields instead of queries, by @kitten (See #840)
- Implement experimental Offline Support in Graphcache. Read more about how to use the Offline Support in our docs., by @kitten (See #793)
- Issue warnings when an unknown type or field has been included in Graphcache's
opts
configuration to help spot typos. Checksopts.keys
,opts.updates
,opts.resolvers
andopts.optimistic
. (See #820 and #826)
⚠️ Fix resolvers being executed for data even when data is currently written. This behaviour could lead to interference with custom updaters that update fragments or queries, e.g. an updater that was receiving paginated data due to a pagination resolver. We've determined that generally it is undesirable to have any resolvers run during the cache update (writing) process, since it may lead to resolver data being accidentally written to the cache or for resolvers to interfere with custom user updates, by @olistic (See #812)- Upgrade to a minimum version of wonka@^4.0.14 to work around issues with React Native's minification builds, which use uglify-es and could lead to broken bundles, by @kitten (See #842)
- Updated dependencies (See #838 and #842)
- @urql/[email protected]
- Add
source
debug name to alldispatchDebug
calls during build time to identify events by which exchange dispatched them, by @JoviDeCroock (See #780) ⚠️ Fix Introspection Queries (or internal types in general) triggering lots of warnings for unkeyed entities, by @kitten (See #779)- Updated dependencies (See #780)
- @urql/[email protected]
- Add a
"./package.json"
entry to thepackage.json
's"exports"
field for Node 14. This seems to be required by packages likerollup-plugin-svelte
to function properly, by @JoviDeCroock (See #771) ⚠️ Fix traversal issue, where when a prior selection set has set a nested result field tonull
, a subsequent traversal of this field attempts to accessprevData
onnull
, by @kitten (See #772)- Updated dependencies (See #771 and #771)
- @urql/[email protected]
- @urql/[email protected]
This release heavily improves on the intuitiveness of how Optimistic Updates work. It ensures that optimistic updates aren't accidentally discarded, by temporarily blocking some refetches when necessary. It also prevents optimistic mutation updates from becoming permanent, which could previously happen if an updater read optimistic data and rewrote it again. This isn't possible anymore as mutation results are applied as a batch.
- Implement refetch blocking for queries that are affected by optimistic update. When a query would normally be refetched, either because it was partial or a cache-and-network operation, we now wait if it touched optimistic data for that optimistic mutation to complete. This prevents optimistic update data from unexpectedly disappearing, by @kitten (See #750)
- Implement optimistic mutation result flushing. Mutation results for mutation that have had optimistic updates will now wait for all optimistic mutations to complete at the same time before being applied to the cache. This sometimes does delay cache updates to until after multiple mutations have completed, but it does prevent optimistic data from being accidentally committed permanently, which is more intuitive, by @kitten (See #750)
- Adjust mutation results priority to always override query results as they arrive, similarly to subscriptions. This will prevent race conditions when mutations are slow to execute at the cost of some consistency, by @kitten (See #745)
- Improve warning and error console output in development by cleaning up the GraphQL trace stack, by @JoviDeCroock (See #751)
Sorry for the many updates; Please only upgrade to >=2.3.8
and don't use the deprecated 2.3.7
and 2.3.6
release.
⚠️ Fix nested package path for @urql/core/internal and @urql/exchange-graphcache/extras, by @kitten (See #734)- Updated dependencies (See #734)
- @urql/[email protected]
- Make the extension of the main export unknown, which fixes a Webpack issue where the resolver won't pick
module
fields inpackage.json
files once it's importing from another.mjs
file, by @JoviDeCroock (See #733) - Updated dependencies (See #733)
- @urql/[email protected]
- Add debugging events to exchanges that add more detailed information on what is happening internally, which will be displayed by devtools like the urql Chrome / Firefox extension, by @andyrichardson (See #608)
⚠️ Fix persistence using special tab character in serialized keys and add sanitization to persistence key serializer, by @kitten (See #715)- Updated dependencies (See #608, #718, and #722)
- @urql/[email protected]
⚠️ Fix @urql/exchange-populate visitWithTypeInfo import by bumping babel-plugin-modular-graphql, by @kitten (See #709)- Updated dependencies (See #709)
- @urql/[email protected]
- Pick modules from graphql package, instead of importing from graphql/index.mjs, by @kitten (See #700)
- Change invalidation to check for undefined links since null is a valid value in graphql, by @JoviDeCroock (See #694)
- Updated dependencies (See #700)
- @urql/[email protected]
- @urql/[email protected]
- Add graphql@^15.0.0 to peer dependency range, by @kitten (See #688)
- Forcefully bump @urql/core package in all bindings and in @urql/exchange-graphcache.
We're aware that in some cases users may not have upgraded to @urql/core, even though that's within
the typical patch range. Since the latest @urql/core version contains a patch that is required for
cache-and-network
to work, we're pushing another patch that now forcefully bumps everyone to the new version that includes this fix, by @kitten (See #684) - Reimplement persistence support to take commutative layers into account, by @kitten (See #674)
- Updated dependencies (See #688 and #678)
- @urql/[email protected]
- @urql/[email protected]
- Support optimistic values for mutations without a selectionset, by @JoviDeCroock (See #657)
- Refactor to replace dictionary-based (
Object.create(null)
) results with regular objects, by @kitten (See #651) ⚠️ Fix case where a mutation-rootfield would cause an empty call to the cache.updates[mutationRootField], by @JoviDeCroock (See #654)- Updated dependencies (See #658 and #650)
- @urql/[email protected]
⚠️ Fix node resolution when using Webpack, which experiences a bug where it only resolvespackage.json:main
instead ofmodule
when an.mjs
file imports a package, by @JoviDeCroock (See #642)- Updated dependencies (See #642)
- @urql/[email protected]
- @urql/[email protected]
⚠️ Fix critical ordering bug in commutative queries and mutations. Subscriptions and queries would ad-hoc be receiving an empty optimistic layer accidentally. This leads to subscription results potentially being cleared, queries from being erased on a second write, and layers from sticking around on every second write or indefinitely. This affects versions> 2.2.2
so please upgrade!, by @kitten (See #638)⚠️ Fix multipart conversion, in theextract-files
dependency (used by multipart-fetch) there is an explicit check for the constructor property of an object. This made the files unretrievable, by @JoviDeCroock (See #639)⚠️ Fix Node.js Module support for v13 (experimental-modules) and v14. If your bundler doesn't support.mjs
files and fails to resolve the new version, please double check your configuration for Webpack, or similar tools, by @JoviDeCroock (See #637)- Updated dependencies (See #637)
- @urql/[email protected]
- @urql/[email protected]
⚠️ Fix cache.inspectFields causing an undefined error for uninitialised or cleared commutative layers, by @kitten (See #626)- Improve Store constructor to accept an options object instead of separate arguments, identical to the cacheExchange options. (This is a patch, not a minor, since we consider Store part of the private API), by @kitten (See #622)
- Allow a single field to be invalidated using cache.invalidate using two additional arguments, similar to store.resolve; This is a very small addition, so it's marked as a patch, by @kitten (See #627)
- Prevent variables from being filtered and queries from being altered before they're forwarded, which prevented additional untyped variables from being used inside updater functions, by @kitten (See #629)
- Expose generated result data on writeOptimistic and passthrough data on write operations, by @kitten (See #613)
- Updated dependencies (See #621)
- @urql/[email protected]
- Refactor parts of Graphcache for a minor performance boost and bundlesize reductions, by @kitten (See #611)
⚠️ Fix Rollup bundle output being written to .es.js instead of .esm.js, by @kitten (See #609)- Updated dependencies (See #609)
- @urql/[email protected]
- Apply commutative layers to all operations, so now including mutations and subscriptions, to ensure that unordered data is written in the correct order, by @kitten (See #593)
- Updated dependencies (See #607 and #601)
- @urql/[email protected]
⚠️ Fix commutative layer edge case when lowest-priority layer comes back earlier than others, by @kitten (See #587)- Externalise @urql/exchange-populate from bundle, by @kitten (See #590)
⚠️ Fix teardown events leading to broken commutativity, by @kitten (See #588)
- Remove the shared package, this will fix the types file generation for graphcache, by @JoviDeCroock (See #579)
- Updated dependencies (See #577)
- @urql/[email protected]
- Add
cache.invalidate
to invalidate an entity directly to remove it from the cache and all subsequent cache results, e.g.cache.invalidate({ __typename: 'Todo', id: 1 })
, by @JoviDeCroock (See #566)
⚠️ Fixcache-only
operations being forwarded and triggering fetch requests, by @JoviDeCroock (See #551)- Apply Query results in-order and commutatively even when results arrive out-of-order, by @kitten (See #565)
- Updated dependencies (See #551, #542, and #544)
- @urql/[email protected]
- Update the
updater
function type ofcache.updateQuery
to have a return type ofDataFields
so that__typename
does not need to be defined, by @JoviDeCroock (See #538) ⚠️ Fix updates not being triggered when optimistic updates diverge from the actual result. (See #160)- Refactor away SchemaPredicates helper to reduce bundlesize. (See #161)
- Ensure that pagination helpers don't confuse pages that have less params with a query that has more params. (See #156)
- Updated dependencies (See #533, #519, #515, #512, and #518)
- @urql/[email protected]
This release adds support for cache persistence which is bringing us one step closer to full offline-support, which we hope to bring you soon.
It also allows wonka@^4.0.0
as a dependency to be compatible with [email protected]
. It also fixes a couple of issues in our
new populateExchange
.
- Refactor internal store code and simplify
Store
(see #134) - ✨ Implement store persistence support (see #137)
- ✨ Apply GC to store persistence (see #138)
- Remove unused case where scalars are written from an API when links are expected (see #142)
⚠️ Add support for resolvers causing cache misses (see #143)⚠️ Fix nested types (e.g.[Item!]!
) inpopulateExchange
(see #150)- Fix duplicate fragments in
populateExchange
output (see #151) - Allow
wonka@^3.2.1||^4.0.0
to be used (see #153)
Note: The minimum required version of
urql
for this release is now1.7.0
!
Christmas came early! This version improves performance again by about 25% over 1.2.2
. It also
now ships with two new features: The populateExchange
and automatic garbage collection.
Including the populateExchange
is optional. It records all fragments in any active queries, and
populates mutation selection sets when the @populate
directive is used based on typenames. If your
schema includes viewer
fields on mutations, which resolve back to your Query
type, you can use
this to automatically update your app's data when a mutation is made. (More documentation on this
is coming soon!)
The garbage collection works by utilising an automatic reference counting algorithm rather than a mark & sweep algorithm. We feel this is the best tradeoff to maintain good performance during runtime while minimising the data that is unnecessarily retained in-memory. You don't have to do anything! Graphcache will do its newly added magic in the background.
There are some breaking changes, if you're using cache.resolveConnections
or resolveValueOrLink
then you now need to use inspectFields
and resolveFieldByKey
instead. You may also now make
use of cache.keyOfField
. (More info on #128)
- ✨ Implement
populateExchange
(see #120) - Improve type safety of
invariant
andwarning
(see #121) - Reduce size of
populateExchange
(see #122) - Move more code to KVMap (see #125)
- Move deletion to setting
undefined
instead (see #126) - Fix multiple edge cases in the
relayPagination
helper, by @rafeca (see #127) - ✨
⚠️ Reimplement data structure and add garbage collection (see #128) - Use Closure Compiler (see #131)
- Switch to using
urql/core
on1.7.0
(see #132)
This patch replaces pessimism
(our former underlying data structure) with a smaller implementation
that just uses Map
s, since we weren't relying on any immutability internally. This cuts down
on bundlesize and massively on GC-pressure, which provides a large speedup on low-end devices.
- Replace Pessimism with mutable store to prevent excessive GC work (see #117)
- Fix viewer fields (which return
Query
types) not being written or read correctly (see #116)
⚠️ Fix unions not being checked supported by schema predicates, by @StevenLangbroek (see #113)- ✨ Add
simplePagination
helper for resolving simple, paginated lists (see #115)
This is a minor release since it increases the peer dependency of urql
to >= 1.6.0
, due to the addition
of the stale
flag to partial responses and cache-and-network
responses. This flag is useful to check
whether more requests are being made in the background by @urql/exchange-graphcache
.
Additionally, this release adds a small stack to every error and warning that indicates where an error has occured. It lists out the query and all subsequent fragments it has been traversing so that errors and warnings can be traced more easily.
- Add a query/fragment stack to all errors and warnings (see #107)
- Add
stale: true
to allcache-and-network
and partial responses (see #108)
- Fix
relayPagination
helper merging pages with different field arguments (see #104)
- Deduplicate connections in
Store.writeConnection
when possible (see #103) - Fix early bail-out in
relayPagination
helper (see #103)
- Trims down the size by 100 bytes (see #96)
- Include the
/extras
build in the published version (see #97) - Invariant and warnings will now have an error code associated with a more elaborate explanation (see #99)
- Invariant errors will now be included in your production bundle (see #100)
- Fixes the relayPagination helper to correctly return partial results (see #101)
- Add special case to relayPagination for first and last during inwards merge (see #102)
Note: The minimum required version of
urql
for this release is now1.5.1
!
Hooray it's v1.0.0
time! This doesn't mean that we won't be changing little things anymore, but we're so far happy with our API and trust Graphcache to work correctly. We will further iterate on this version with some planned features, like "fragment invalidation", garbage collection, and more.
This version refactors the cache resolvers and adds some new special powers to them! You can now return almost anything from cache resolvers and trust that it'll do the right thing:
- You can return entity keys, which will resolve the cached entities
- You can return keyable entities, which will also be resolved from cache
- You may also return unkeyable entities, which will be partially resolved from cache, with your resolved values taking precedence
This can also be nested, so that unkeyable entities can eventually lead back to normal, cached entities!
This has enabled us to expose the relayPagination()
helper! This is a resolver that you can just drop into the cacheExchange
's resolvers
config. It automatically does Relay-style pagination, which is now possible due to our more powerful resolvers! You can import it from @urql/exchange-graphcache/extras
.
- ✨ Add full cache resolver traversal (see #91)
- ✨ Add a new
relayPagination
helper (see #91) - Add a
Cache
interface with all methods (that are safe for userland) having documentation (see #91) - ⚠ Fix non-default root keys (that aren't just
Query
) not being respected (see #87)
- Fix
updates
not being called foroptimistic
results (see #83) - Add optional
variables
argument toreadFragment
andwriteFragment
(see #84) - ⚠ Fix field arguments not normalising optional
null
values (see #85)
- ⚠ Fix removing cache entries by upgrading to Pessimism
1.1.4
(see ae72d3)
- ⚠ Fix optimistic updates by upgrading to Pessimism
1.1.3
(see #81)
- Fix warnings being shown for Relay
Connection
andEdge
embedded types (see #79) - Implement
readFragment
method onStore
(see #73) - Implement
readQuery
method onStore
(see #73) - Improve
writeFragment
method onStore
(see #73)
- ⚠ Fix reexecuted operations due to dependencies not using
cache-first
(see 0bd58f6)
- ⚠ Fix concurrency issue where a single operation is reexecuted multiple times (see #70)
- Skip writing
undefined
to the cache and log a warning in development (see #71) - Allow
query
to be passed as a string tostore.updateQuery
(see #72)
- ⚠ Fix user-provided
keys
config not being able to returnnull
(see #68)
- ⚠ Fix development warnings throwing an error for root fields (see #65)
Note: This is release contains a bug that v1.0.0-rc.4
fixes
- Fix warning condition for missing entity keys (see 98287ae)
Note: This is release contains a bug that v1.0.0-rc.3
fixes
- Add warnings for unknown fields based on the schema and deduplicate warnings (see #63)
This is the first release that adds schema awareness. Passing a schema to Graphcache allows it to make deterministic
assumptions about the cached results it generates from its data. It can deterministically match fragments to interfaces,
instead of resorting to a heuristic, and it can provide partial results for queries. With a schema
passed to Graphcache,
as long as only nullable fields are uncached and missing, it will still provide an initial cached result.
- ✨ Add schema awareness using the
schema
option (see #58) - ✨ Allow for partial results to cascade missing values upwards (see #59)
- Fix
store.keyOfEntity
not using root names from the schema (see #62)
This is where this CHANGELOG starts.
For a log on what happened in beta
and alpha
releases, please read the commit history.