refactor: split min latency flags (#1351)

RE: https://sourcegraph.slack.com/archives/C05AGQYD528/p1696845420506809

refactor: replace cody autocomplete minimum latency flag with user,
provider, and language based latency flags

Replaced the `CodyAutocompleteMinimumLatency` feature flag with two new
flags:

- `CodyAutocompleteUserLatency `: Enables minimum latency for Cody
autocomplete. Works the same as the old `CodyAutocompleteMinimumLatency`
flag but rename it to make sure the old flag will not be used in older
versions.
- `CodyAutocompleteLanguageLatency`: Enables language-specific minimum
latency for Cody autocomplete for low-performance languages ONLY
- `CodyAutocompleteProviderLatency`: Enables provider-specific minimum
latency for non-anthropic provider

Expected behavior:
- latency for comments will be applied if any of the feature flags is
enabled
- latency for user-based actions will only be applied if
`CodyAutocompleteUserLatency` is enabled
- latency for language-based actions will only be applied if
`CodyAutocompleteLanguageLatency` is enabled
- latency for provider-based actions will only be applied if
`CodyAutocompleteProviderLatency` is enabled

The getLatency function now accepts feature flags and calculates latency
accordingly.

This allows more granular control over autocomplete latency based on
user, provider, and language.

## Test plan

<!-- Required. See
https://docs.sourcegraph.com/dev/background-information/testing_principles.
-->

Unit tests are updated.

---------

Co-authored-by: Philipp Spiess <hello@philippspiess.com>
Co-authored-by: Dominic Cooney <dominic.cooney@sourcegraph.com>

lib/shared/src/experimentation/FeatureFlagProvider.ts

@@ -15,9 +15,11 @@ export enum FeatureFlag {
     CodyAutocompleteLlamaCode7B = 'cody-autocomplete-default-llama-code-7b',
     CodyAutocompleteLlamaCode13B = 'cody-autocomplete-default-llama-code-13b',
     CodyAutocompleteGraphContext = 'cody-autocomplete-graph-context',
-    CodyAutocompleteMinimumLatency = 'cody-autocomplete-minimum-latency',
     CodyAutocompleteSyntacticTriggers = 'cody-autocomplete-syntactic-triggers',
     CodyAutocompleteStarCoderExtendedTokenWindow = 'cody-autocomplete-starcoder-extended-token-window',
+    CodyAutocompleteLanguageLatency = 'cody-autocomplete-language-latency',
+    CodyAutocompleteUserLatency = 'cody-autocomplete-user-latency',
+    CodyAutocompleteProviderLatency = 'cody-autocomplete-provider-latency',
 }
 
 const ONE_HOUR = 60 * 60 * 1000

vscode/src/completions/inline-completion-item-provider.ts

@@ -20,7 +20,7 @@ import {
     LastInlineCompletionCandidate,
     TriggerKind,
 } from './get-inline-completions'
-import { getLatency, resetLatency } from './latency'
+import { getLatency, LatencyFeatureFlags, resetLatency } from './latency'
 import * as CompletionLogger from './logger'
 import { CompletionEvent, SuggestionID } from './logger'
 import { ProviderConfig } from './providers/provider'
@@ -135,11 +135,15 @@ export class InlineCompletionItemProvider implements vscode.InlineCompletionItem
 
         // We start feature flag requests early so that we have a high chance of getting a response
         // before we need it.
-        const [isIncreasedDebounceTimeEnabledPromise, minimumLatencyFlagsPromise, syntacticTriggersPromise] = [
+        const [isIncreasedDebounceTimeEnabledPromise, syntacticTriggersPromise] = [
             featureFlagProvider.evaluateFeatureFlag(FeatureFlag.CodyAutocompleteIncreasedDebounceTimeEnabled),
-            featureFlagProvider.evaluateFeatureFlag(FeatureFlag.CodyAutocompleteMinimumLatency),
             featureFlagProvider.evaluateFeatureFlag(FeatureFlag.CodyAutocompleteSyntacticTriggers),
         ]
+        const minLatencyFlagsPromises = {
+            user: featureFlagProvider.evaluateFeatureFlag(FeatureFlag.CodyAutocompleteUserLatency),
+            language: featureFlagProvider.evaluateFeatureFlag(FeatureFlag.CodyAutocompleteLanguageLatency),
+            provider: featureFlagProvider.evaluateFeatureFlag(FeatureFlag.CodyAutocompleteProviderLatency),
+        }
 
         const tracer = this.config.tracer ? createTracerForInvocation(this.config.tracer) : undefined
         const graphContextFetcher = this.config.graphContextFetcher ?? undefined
@@ -254,9 +258,17 @@ export class InlineCompletionItemProvider implements vscode.InlineCompletionItem
             // latency so that we don't show a result before the user has paused typing for a brief
             // moment.
             if (result.source !== InlineCompletionsResultSource.LastCandidate) {
-                const minimumLatencyFlag = await minimumLatencyFlagsPromise
-                if (triggerKind === TriggerKind.Automatic && minimumLatencyFlag) {
+                const latencyFeatureFlags: LatencyFeatureFlags = {
+                    user: await minLatencyFlagsPromises.user,
+                    language: await minLatencyFlagsPromises.language,
+                    provider: await minLatencyFlagsPromises.provider,
+                }
+
+                const isMinLatencyEnabled =
+                    latencyFeatureFlags.user || latencyFeatureFlags.language || latencyFeatureFlags.provider
+                if (triggerKind === TriggerKind.Automatic && isMinLatencyEnabled) {
                     const minimumLatency = getLatency(
+                        latencyFeatureFlags,
                         this.config.providerConfig.identifier,
                         document.uri.fsPath,
                         document.languageId,