rewrite VOM LRU cache #7138

warner · 2023-03-09T00:27:58Z

rewrite virtual object manager to use the new cache

This overhauls the VOM, to drastically simplify the creation of
virtual-object Representatives, and remove the syscall sensitivity to
GC.

The old code created state (with accessors for each state property)
and context (which holds both state and self or the facets
cohort) each time it needed to create a Representative, which is every
time a virtual/durable vref is deserialized but a Representative was
not present in the slotToVal table (thus sensitive to GC). This was
partially a holdover from the time when Representatives were behavior
methods closing over state.

Since then, we moved to a model where Representatives are empty
objects, with a special prototype that holds the behavior methods. We
use defendPrototype from exo-tools.js to build this prototype,
which wraps each behavior method with code that looks the
Representative up in a "context map" to find the context
object. This reduces memory usage: we share the prototype object among
all instances of the Kind, and the Representatives have no own
properties.

It also enables the key improvement of this commit: we can defer
creating context until userspace actually invokes a method on the
Representative. We replace the context map with a contextProvider: a
function that takes the Representative and looks up (or creates) a
context. Since userspace invocation of methods is unaffected by GC,
calls to the contextProvider are too, making it a safe place to
perform syscalls (such as the vatstoreGet needed to learn the state
properties and create the state accessors).

The cache ensures that we create the context at most once per
delivery.

We use two instances of the cache: one for context (including
state), and a second for the actual state data.

We remove virtualObjectCacheSize from vat options, since the new
cache does not have a configurable size limit.

warner · 2023-03-09T00:28:09Z

cc @FUDCo

datadog-full-agoric · 2023-03-09T00:35:33Z

Datadog Report

Branch report: 6693-vo-lru-cache
Commit report: 1c96aa8

❌ agoric-sdk: 384 Failed (0 Known Flaky), 0 New Flaky, 982 Passed, 31 Skipped, 22m 48.56s Wall Time

❌ Failed Tests (384)

This report shows up to 5 failed tests.

make › make and exec - agoric.cosmic-swingset - Details

Expand for error

 ---
     name: Error
     message: Exited because no new tests completed within the last 1200000ms of inactivity
     at: >-
       Error: Exited because no new tests completed within the last 1200000ms of
       inactivity
           at TapReporter.writeTimeout (file:///home/runner/work/agoric-sdk/agoric-sdk/node_modules/ava/lib/reporters/tap.js:148:17)
           at TapReporter.consumeStateChange (file:///home/runner/work/agoric-sdk/agoric-sdk/node_modules/ava/lib/reporters/tap.js:218:10)
           at file:///home/runner/work/agoric-sdk/agoric-sdk/node_modules/ava/lib/reporters/tap.js:72:45
           at file:///home/runner/work/agoric-sdk/agoric-sdk/node_modules/emittery/index.js:365:13
 ...

attestation › attestation › before hook - agoric.inter-protocol - Details

Expand for error

 ---
     name: AssertionError
     message: Rejected promise returned by test
     values:
       'Rejected promise returned by test. Reason:': |-
         Error {
           message: 'illegal cross-facet access',
         }
     at: >-
       Object.get (packages/swingset-liveslots/src/virtualObjectManager.js:174:34)
 ...

attestation › attestation › before hook - agoric.inter-protocol - Details

Expand for error

 ---
     name: AssertionError
     message: Rejected promise returned by test
     values:
       'Rejected promise returned by test. Reason:': |-
         Error {
           message: 'illegal cross-facet access',
         }
     at: >-
       Object.get (packages/swingset-liveslots/src/virtualObjectManager.js:174:34)
 ...

attestation › userFlow › before hook - agoric.inter-protocol - Details

Expand for error

 ---
     name: AssertionError
     message: Rejected promise returned by test
     values:
       'Rejected promise returned by test. Reason:': |-
         Error {
           message: 'illegal cross-facet access',
         }
     at: >-
       Object.get (packages/swingset-liveslots/src/virtualObjectManager.js:174:34)
 ...

attestation › userFlow › before hook - agoric.inter-protocol - Details

Expand for error

 ---
     name: AssertionError
     message: Rejected promise returned by test
     values:
       'Rejected promise returned by test. Reason:': |-
         Error {
           message: 'illegal cross-facet access',
         }
     at: >-
       Object.get (packages/swingset-liveslots/src/virtualObjectManager.js:174:34)
 ...

warner · 2023-03-09T00:55:02Z

Yeah, I think I reintroduced the main GC sensitivity (the one on reanimating Representatives). I want to fix this by makine state into a Proxy instead of enumerated getters/setters, so we can defer looking up the list of property names until someone looks for them (since lookup/get/set are userspace behaviors, we're allowed to react to them, but reanimate is GC-sensitive). But I think the last time we considered this, the limitations of a harden()ed Proxy came up.

warner · 2023-03-09T08:02:02Z

Actually, I believe that sensitivity was pre-existing, see #7142. If so, this branch is just as problematic, but at least it's a lot simpler.

codecov · 2023-03-24T07:45:52Z

Codecov Report

Merging #7138 (c645b4a) into master (2e05748) will decrease coverage by 0.28%.
The diff coverage is 94.52%.

❗ Current head c645b4a differs from pull request most recent head 54797ea. Consider uploading reports for the commit 54797ea to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #7138      +/-   ##
==========================================
- Coverage   71.00%   70.72%   -0.28%     
==========================================
  Files         450      449       -1     
  Lines       86390    85575     -815     
  Branches        3        3              
==========================================
- Hits        61338    60526     -812     
+ Misses      24984    24983       -1     
+ Partials       68       66       -2

Impacted Files	Coverage Δ
...ckages/SwingSet/src/controller/initializeKernel.js	`99.10% <ø> (-0.01%)`	⬇️
.../SwingSet/src/kernel/vat-loader/manager-factory.js	`73.10% <ø> (-0.23%)`	⬇️
...kages/SwingSet/src/kernel/vat-loader/vat-loader.js	`62.19% <ø> (-0.10%)`	⬇️
packages/SwingSet/src/types-external.js	`100.00% <ø> (ø)`
...ges/swingset-liveslots/src/virtualObjectManager.js	`92.50% <94.08%> (+3.41%)`	⬆️
packages/store/src/patterns/exo-makers.js	`97.10% <100.00%> (+0.06%)`	⬆️
packages/store/src/patterns/exo-tools.js	`82.01% <100.00%> (+0.26%)`	⬆️
packages/swingset-liveslots/src/liveslots.js	`92.69% <100.00%> (+0.62%)`	⬆️
...ingset-liveslots/tools/fakeVirtualObjectManager.js	`100.00% <100.00%> (ø)`
...ges/swingset-liveslots/tools/fakeVirtualSupport.js	`96.17% <100.00%> (-0.54%)`	⬇️

... and 24 files with indirect coverage changes

warner · 2023-03-24T07:48:49Z

Note: this changes the VOM, but not the collection manager (which still has two sources of GC sensitivity). I'll address those in a separate PR.

Also note that for some reason, this change increases the sensitivity of the real-GC tests to Node.js and AVA being weird (#5575), specifically in test/virtual-objects/test-virtualObjectGC.js. I increased the number of engineGC() calls to make them pass consistently again, at the cost of adding one or two seconds to the swingset-liveslots tests. I still have no idea what's happening, and I'm hoping that the extra calls will be enough to avoid new (or more frequent) intermittent failures.

@erights Please note that I'm changing exo-tools.js from a contextMap to a contextProvider function. I considered sticking with a Map-like interface ("an object with a .get method"), but changing the types to allow a non-Map was just as much surgery as switching to a proper function. We'll have to figure out how to upstream this into Endo eventually.. for now I think it's ok for us to keep pulling defendPrototype/etc from packages/store.

datadog-full-agoric · 2023-03-24T07:49:50Z

Datadog Report

Branch report: 6693-vo-lru-cache
Commit report: db193b6

❌ agoric-sdk: 6 Failed (0 Known Flaky), 0 New Flaky, 3458 Passed, 55 Skipped, 21m 25.16s Wall Time

❌ Failed Tests (6)

This report shows up to 5 failed tests.

virtualObjects › representatives › exercise cache - agoric.SwingSet - Details

Expand for error

 ---
     name: AssertionError
     assertion: deepEqual
     values:
       'Difference (- actual, + expected):': |2-
           [
         -   [
         -     'get',
         -     'v1.vs.vom.rc.o+v10/2',
         -     undefined,
 ...

virtualObjects › representatives › strongly held cohort retains representative - agoric.SwingSet - Details

Expand for error

 ---
     name: AssertionError
     assertion: deepEqual
     values:
       'Difference (- actual, + expected):': |2-
           [
         -   'compare old === new : false',
         +   'compare old === new : true',
           ]
     at: >-
 ...

virtualObjects › representatives › strongly held state retains representative - agoric.SwingSet - Details

Expand for error

 ---
     name: AssertionError
     assertion: deepEqual
     values:
       'Difference (- actual, + expected):': |2-
           [
         -   'compare old === new : false',
         +   'compare old === new : true',
           ]
     at: >-
 ...

virtualObjects › representatives › weakly held cohort retains representative - agoric.SwingSet - Details

Expand for error

 ---
     name: AssertionError
     assertion: deepEqual
     values:
       'Difference (- actual, + expected):': |2-
           [
         -   'compare old === new : false',
         +   'compare old === new : true',
           ]
     at: >-
 ...

virtualObjects › representatives › weakly held state retains representative - agoric.SwingSet - Details

Expand for error

 ---
     name: AssertionError
     assertion: deepEqual
     values:
       'Difference (- actual, + expected):': |2-
           [
         -   'compare old === new : false',
         +   'compare old === new : true',
           ]
     at: >-
 ...

gibson042

Partial review, but the three files still to go are the ones with all the substance:

packages/swingset-liveslots/test/test-liveslots-mock-gc.js
packages/swingset-liveslots/test/test-liveslots-real-gc.js
packages/swingset-liveslots/src/virtualObjectManager.js

packages/swingset-liveslots/test/virtual-objects/test-cross-facet.js

gibson042 · 2023-03-27T18:32:24Z

packages/store/src/patterns/exo-tools.js

+/**
+ * @typedef { (representative: any) => any } ContextProvider
+ */


How about narrowing this?

Suggested change

/**

* @typedef { (representative: any) => any } ContextProvider

*/

/**

* @typedef {{self: any, state: any}} ClassContext

* @typedef {{self: any, facets: Record<string, any>}} ClassKitContext

* @typedef {(representative: any) => ClassContext} ContextProvider

* @typedef {Record<facetName: string, (facet: any) => ClassKitContext>} ContextProviderKit

*/

oh, that changes some work that @erights just did to synchronize the agoric-sdk copy of exo-tools with the Endo one

@erights what do you think about this? If you prefer this, let's do another round of (fix in agoric-sdk, fix in Endo), and I'll rebase this branch to include those changes.

oh, also note: the change this comment is referring to has been landed already, as part of #7304 , so further improvements to the type shouldn't affect this PR (unless swingset is doing something wrong)

gibson042 · 2023-03-27T19:10:42Z

packages/swingset-liveslots/src/cache.js

+ * This cache is part of the virtual object manager and is not intended to be
+ * used independently; it is exported only for the benefit of test code.


Do we really not already have a suitable preexisting implementation of this? I wonder if it merits a ticket/TODO for e.g. merging with makeReadCachingStorage.

I'm really hesitant to share an implementation with other customers whose needs may diverge, or which might provoke the package that hosts the cache to add new dependencies (I'd want to put the cache in its own package, with no additional dependencies). Also I think the "hold until flushed" behavior might not be useful in as many other contexts as the more traditional LRU.

That said, makeReadCachingStorage (in cosmic-swingset) looks like an 80% match. The overhead of a new package feels like it would outweigh the savings made by sharing this with the other cache, but maybe it'd make sense.

packages/swingset-liveslots/src/cache.js

packages/swingset-liveslots/test/test-cache.js

packages/swingset-liveslots/test/virtual-objects/test-empty-data.js

packages/swingset-liveslots/test/virtual-objects/test-virtualObjectManager.js

packages/swingset-liveslots/test/virtual-objects/test-vo-real-gc.js

warner · 2023-03-30T23:57:39Z

two known tests still failing:

(Swingset): yarn test test/virtualObjects/test-representatives.js -m "weakly held cohort retains representative" (and others in the same file)
(zoe): yarn test test/swingsetTests/upgradeCoveredCall/test-coveredCall-service-upgrade.js
- somehow it's not reattaching a Kind during upgrade

erights · 2023-03-31T01:59:49Z

Be aware of #7286 , which I think is a complementary optimization. But I'm not sure.

gibson042

The overall structure of virtualObjectManager.js makes sense and seems good, but I would like some more complete commentary. Everything else is just suggested tweaks to code organization and/or messaging.

I haven't yet reviewed test-liveslots-{mock,real}-gc.js, but am pausing this for now to work on liveslots export/representative visibility and #7083.

packages/swingset-liveslots/src/virtualObjectManager.js

FUDCo

Looks good. I got nuthin' but nit picks.

packages/swingset-liveslots/src/liveslots.js

FUDCo · 2023-04-04T19:08:45Z

packages/swingset-liveslots/test/test-cache.js

@@ -26,12 +26,9 @@ test('cache', t => {

  // new reads pass through immediately to backing store
  t.is(c.get('key0'), 'value0');
+  t.deepEqual(log.splice(0), [['read', 'key0']]);


splice is a confusing operation that I have to look up the docs for each time I use it. I'm wondering if it would be clearer to have a helper function that consumes the head of the log and returns it (which of course would be written with splice, but only in one place). Seems like we have something like that in other tests.

Heh, see #7138 (comment)

I like splice for supporting the ability to make assertions about a batch of entries rather than one at a time, and really this should be an external test helper where the implementation is explained inline but never really considered at call sites because it Just Works as expected (above suggestion updated with that in mind).

/** * Clear the contents of a log array and assert that they match expectations. * * @param {import('ava').ExecutionContext} t * @param {unknown[]} log * @param {unknown[]} expected * @returns {void} */ const clearAndAssertLog = (t, log, expected) => { // `log.splice(0)` removes elements starting at index 0 (i.e., every element) // and returns them in an array. const entries = log.splice(0); t.deepEqual(entries, expected); }

ok, two votes for a helper, I'll start using one in the future

This cache accumulates all data until an explicit flush, at which point it writes back all dirty state, and forgets all state. This avoids the somewhat unpredictable behavior of an LRU cache, which caused us problems with GC sensitivity. It suffers from the potential for larger RAM usage between flushes, but is quite simple, and yields easy-to-debug syscall behavior. We have other caches in our platform, but liveslots needs something with these specific properties, and I don't know what the needs in the rest of the platform are, so I'm fine with not sharing this implementation with other packages, at least for now. refs #6693

This overhauls the VOM, to drastically simplify the creation of virtual-object Representatives, and remove the syscall sensitivity to GC. The old code created `state` (with accessors for each state property) and `context` (which holds both `state` and `self` or the `facets` cohort) each time it needed to create a Representative, which is every time a virtual/durable vref is deserialized but a Representative was not present in the slotToVal table (thus sensitive to GC). This was partially a holdover from the time when Representatives were behavior methods closing over `state`. Since then, we moved to a model where Representatives are empty objects, with a special prototype that holds the behavior methods. We use `defendPrototype` from `exo-tools.js` to build this prototype, which wraps each behavior method with code that looks the Representative up in a "context map" to find the `context` object. This reduces memory usage: we share the prototype object among all instances of the Kind, and the Representatives have no own properties. It also enables the key improvement of this commit: we can defer creating `context` until userspace actually invokes a method on the Representative. We replace the context map with a `contextProvider`: a function that takes the Representative and looks up (or creates) a context. Since userspace invocation of methods is unaffected by GC, calls to the contextProvider are too, making it a safe place to perform syscalls (such as the vatstoreGet needed to learn the state properties and create the `state` accessors). The cache ensures that we create the context at most once per delivery. We use two instances of the cache: one for `context` (including `state`), and a second for the actual state data. For each baseref, the state data cache holds the full `capdatas` record (separate capdata for each property name). It also holds a sparsely/lazily-populated `valueMap`, a Map of unserialized values. These are made available to `state` getters. When modifying a property on `state`, we immediately serialize it and hold it in the cache. We also perform refcount delta management at that time. Later, during flush(), we reassemble the capdatas (some modified, some pristine) into the value of a `syscall.vatstoreSet`. We defer unserialization until a getter is called, which reduces the constraints on userspace behavior, and making virtual objects behave more like virtual collections. We remove `virtualObjectCacheSize` from vat options, since the new cache size is unlimited, and thus not configurable. This changes the `slotToVal` mapping to hold the cohort *record*, where previously it held an Array. This is for multi-facet Kinds, whose list of named facets is learned during `defineKindMulti`. Each Kind has a fixed list of facet names, so each facet has both a name and an index. Userspace receives these as a record, with one property per facet. Internally, we need the index to encode into the vref, e.g. `o+11/5:1` means the second (0-indexed) facet of the 5th (1-indexed) instance of the Kind with KindID 11. We also use the index when looking up the export status, which is recorded in one vatstore key per baseref, with one character per facet, so 'rrn' means the first two facets are reachable, and the third is unreachable. Previously, during construction or reanimation, we also created an *Array* of facets, and recorded them in the slotToVal Map (under the baseref as a key). When doing `convertSlotToVal()`, we'd extract the baseref from the vref, use it to fetch the array, then use the integer facet index to extract the specific facet. This complicated the construction process, because we needed to create two separate containers for the set of facets, which could have separate lifetimes. We needed a third object (named `keepAlive`) to help maintain the cyclic GC-reference subgraph containing all facets and the user-visible facet record. With this commit, we discard this extra Array, and store the facet record itself in slotToVal. `convertSlotToVal()` asks the VRM to convert the facet index (extracted from the vref) into a facet name, and then looks it up in the slotToVal record. When registering a cohort, `registerValue` iterates through the facet names to make sure we register each separate facet in the reverse-pointing valToSlot table. This gets rid of `toRegister`, `toExpose`, and `keepAlive`.

* remove virtualObjectCacheSize from options * change expectations about when vatstoreSet calls happen * add some tests * fix test-cross-facet to expect a different error message

for some reason, these recent changes have caused 20% of 'yarn test' runs to fail, in test-virtualObjectGC.js, in varying test functions. This happens even when I run just test-virtualObjectGC.js, with --serial, --no-worker-threads, --concurrency 1. The symptom is consistent with GC not successfully releasing the target object. We never figured out why one engineGC() call seemed to help, but apparently three calls work even better.

* remove cacheSize / virtualObjectCacheSize from options * update convertSlotToVal to expect facets object, not array

This affects both static-vat creation options in `config.vats.NAME.creationOptions.virtualObjectCacheSize`, the dynamic-vat options to `E(vatAdminSvc).createVat()`, and the vatAdmin `changeOptions()` facility.

I saw a test failure here, "historical inaccuracy in replay", which might have been due to GC happening one way in the original, and a different way in the replay (when running under Node.js). This feels like an aspect of #5575, and this test isn't trying to exercise anything about GC, so I'm just going to set defaultReapInterval to 'never' to inhibit BOYD, which should remove the problem.

warner mentioned this pull request Mar 9, 2023

GC syscall sensitivity through reanimate and reanimateDurableKindID #7142

Closed

warner force-pushed the 6693-vo-lru-cache branch 2 times, most recently from 0a29ba3 to 9264d6b Compare March 13, 2023 16:21

warner force-pushed the 6693-vo-lru-cache branch 2 times, most recently from 7d562a0 to ef539c1 Compare March 24, 2023 07:42

warner force-pushed the 6693-vo-lru-cache branch from ef539c1 to 2ed0d9e Compare March 24, 2023 08:03

This was referenced Mar 24, 2023

enforce that buildRootObject() returns an ephemeral, not durable #7236

Closed

vats must not use virtual/durable as root object #7240

Closed

warner force-pushed the 6693-vo-lru-cache branch from 2ed0d9e to 0a1c34e Compare March 25, 2023 04:30

warner changed the base branch from master to 7240-forbid-durable-root March 25, 2023 04:30

warner force-pushed the 7240-forbid-durable-root branch from 03c779e to b01ac26 Compare March 25, 2023 20:34

warner force-pushed the 6693-vo-lru-cache branch from b24d868 to 4a83063 Compare March 25, 2023 20:36

gibson042 reviewed Mar 27, 2023

View reviewed changes

warner force-pushed the 7240-forbid-durable-root branch from ae15780 to 9c3b8f5 Compare March 28, 2023 01:24

warner force-pushed the 6693-vo-lru-cache branch from 4a83063 to c645b4a Compare March 28, 2023 01:25

Base automatically changed from 7240-forbid-durable-root to master March 28, 2023 02:16

warner force-pushed the 6693-vo-lru-cache branch from c645b4a to 54797ea Compare March 30, 2023 23:30

erights mentioned this pull request Mar 31, 2023

refactor(swingset-liveslots): stateShape causes accessor sharing #7286

Closed

gibson042 reviewed Mar 31, 2023

View reviewed changes

This was referenced Apr 1, 2023

fix: sync with shadows in agoric-sdk endojs/endo#1532

Merged

fix: only the exo api change #7304

Merged

FUDCo approved these changes Apr 4, 2023

View reviewed changes

warner force-pushed the improve-swingset-tests branch 2 times, most recently from 4513d7e to 7ca461c Compare April 4, 2023 23:40

warner force-pushed the 6693-vo-lru-cache branch 2 times, most recently from 608857b to 14af542 Compare April 5, 2023 00:56

Base automatically changed from improve-swingset-tests to master April 5, 2023 01:14

warner force-pushed the 6693-vo-lru-cache branch from 14af542 to 9beef4a Compare April 5, 2023 01:19

warner self-assigned this Apr 5, 2023

warner added SwingSet package: SwingSet liveslots requires vat-upgrade to deploy changes labels Apr 5, 2023

warner added 11 commits April 4, 2023 21:55

split out and simplify assessFacetiousness

5ed636d

update liveslots tests to match

ff969ab

* remove virtualObjectCacheSize from options * change expectations about when vatstoreSet calls happen * add some tests * fix test-cross-facet to expect a different error message

one test-gc-sensitivity test is no longer failing

5a60ea6

update fake-virtual support

50ef3ac

* remove cacheSize / virtualObjectCacheSize from options * update convertSlotToVal to expect facets object, not array

remove virtualObjectCacheSize from SwingSet options

e397fc3

This affects both static-vat creation options in `config.vats.NAME.creationOptions.virtualObjectCacheSize`, the dynamic-vat options to `E(vatAdminSvc).createVat()`, and the vatAdmin `changeOptions()` facility.

fix test-representatives to new cache vatstore syscalls

03060bc

update test-facet-retention.js for new contextCache

b3cea3a

warner force-pushed the 6693-vo-lru-cache branch from 9beef4a to 99522ea Compare April 5, 2023 05:01

warner added the automerge:rebase Automatically rebase updates, then merge label Apr 5, 2023

mergify bot merged commit 8e6a425 into master Apr 5, 2023

mergify bot deleted the 6693-vo-lru-cache branch April 5, 2023 06:15

warner mentioned this pull request Apr 5, 2023

cache the collection schema, only read from vatstore once per crank #7333

Merged

gibson042 mentioned this pull request Apr 7, 2023

chore(store): Define narrow VOM context types #7360

Merged

warner mentioned this pull request Apr 12, 2023

change virtual-object state LRU cache: write-through, write-back at end-of-crank, or remove entirely #6693

Closed

erights mentioned this pull request Apr 18, 2023

refactor: stateShape causes accessor sharing #7444

Merged

gibson042 mentioned this pull request Jul 15, 2023

chore(packages): fix default param lint warnings #8052

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

rewrite VOM LRU cache #7138

rewrite VOM LRU cache #7138

warner commented Mar 9, 2023 •

edited

Loading

warner commented Mar 9, 2023

datadog-full-agoric bot commented Mar 9, 2023 •

edited

Loading

warner commented Mar 9, 2023

warner commented Mar 9, 2023

codecov bot commented Mar 24, 2023 •

edited

Loading

warner commented Mar 24, 2023

datadog-full-agoric bot commented Mar 24, 2023 •

edited

Loading

gibson042 left a comment •

edited

Loading

gibson042 Mar 27, 2023

warner Apr 1, 2023

warner Apr 2, 2023

gibson042 Mar 27, 2023

warner Mar 30, 2023

warner commented Mar 30, 2023

erights commented Mar 31, 2023

gibson042 left a comment

FUDCo left a comment

FUDCo Apr 4, 2023

gibson042 Apr 4, 2023

warner Apr 4, 2023

-/**
- * @typedef { (representative: any) => any } ContextProvider
- */
+/**
+ * @typedef {{self: any, state: any}} ClassContext
+ * @typedef {{self: any, facets: Record<string, any>}} ClassKitContext
+ * @typedef {(representative: any) => ClassContext} ContextProvider
+ * @typedef {Record<facetName: string, (facet: any) => ClassKitContext>} ContextProviderKit
+ */

		* This cache is part of the virtual object manager and is not intended to be
		* used independently; it is exported only for the benefit of test code.

rewrite VOM LRU cache #7138

rewrite VOM LRU cache #7138

Conversation

warner commented Mar 9, 2023 • edited Loading

warner commented Mar 9, 2023

datadog-full-agoric bot commented Mar 9, 2023 • edited Loading

Datadog Report

❌ Failed Tests (384)

warner commented Mar 9, 2023

warner commented Mar 9, 2023

codecov bot commented Mar 24, 2023 • edited Loading

Codecov Report

warner commented Mar 24, 2023

datadog-full-agoric bot commented Mar 24, 2023 • edited Loading

Datadog Report

❌ Failed Tests (6)

gibson042 left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

warner commented Mar 30, 2023

erights commented Mar 31, 2023

gibson042 left a comment

Choose a reason for hiding this comment

FUDCo left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

warner commented Mar 9, 2023 •

edited

Loading

datadog-full-agoric bot commented Mar 9, 2023 •

edited

Loading

codecov bot commented Mar 24, 2023 •

edited

Loading

datadog-full-agoric bot commented Mar 24, 2023 •

edited

Loading

gibson042 left a comment •

edited

Loading