AsyncAPI 3.x: the action and individual a search holds by LautaroPetaccio · Pull Request #1711 · WebFuzzing/EvoMaster

LautaroPetaccio · 2026-08-24T17:10:38Z

Eighth in the AsyncAPI stack, on top of #6. Turns the parsed document into something a search can sample and mutate.

The action

AsyncApiAction is one message published on one channel, following the shape of RPCCallAction — the closest analogue in EvoMaster, being a call with no URL, made through a driver, whose mutable state is the input and whose response is read afterwards rather than searched over.

What is deliberately not a gene is the design.

The address, which the contract fixes. Sending to a channel the service does not read would only waste executions.
The correlation id, stamped fresh at each execution. Searching over it could achieve nothing — the service only echoes it back — and pairing a reply with its request needs a value unique per execution rather than one carried in the genome. So the header it is stamped into is left out of the header genes; and when it is the only header, there are no header genes at all, since an empty schema would otherwise build a free-form map gene and invite the search to invent headers the contract never declared.
The reply, which is an observation. It is read at fitness time to decide what was covered, exactly as RPC reads its response.

The individual

AsyncApiIndividual is one individual for every transport rather than one per transport. Kafka versus AMQP versus a socket appears nowhere in it: what is searched over is the operation and the payload, which are the same whatever moves the bytes, and which wire is used is decided below the driver interface. Were the transport to leak in here there would have to be a KafkaIndividual and an AmqpIndividual with nothing shared between them.

The builder

AsyncApiActionBuilder fills an action cluster, keeping the contract addActionsFromSwagger and addActionsFromSchema already use: one entry per action name, and a list of what had to be skipped rather than an exception.

Two decisions worth review:

Only operations the service consumes become actions. A send operation is what the service emits — something to subscribe to, not to drive. They are counted as skipped rather than built.
An operation carrying several messages becomes several actions. Which message to send is a choice the search makes by picking an action, not by mutating a gene. The message is named in the action only when there is more than one (recv_list_legs rather than recv_list_legs:listLegs), so a generated test reads as the operation wherever it can.

One thing not copied across

RPCIndividual.seeIndexedRPCCalls looks up children by action type, but the children are EnterpriseActionGroups, so it returns nothing. It has no callers, so it appears to be dead. I wrote the same method here, watched a test fail on it, and removed it rather than carrying the bug over — seeMainExecutableActions does the job. Worth a look on the RPC side.

Testing

15 new tests in AsyncApiActionBuilderTest, covering the action cluster built from NCS, microcks and a WebSocket document, what an action does and does not expose as genes, adding and removing messages from an individual, and copying one. 108 across the AsyncAPI package.

jgaleotti · 2026-09-04T20:58:29Z

+        }
+
+        //a field that is no longer there cannot be required either
+        (copy.get("required") as? ArrayNode)?.let { required ->


replace strings with constants

Done everywhere in this PR.

jgaleotti · 2026-09-04T21:00:04Z

+ * There is one individual for every transport rather than one per transport: Kafka versus AMQP
+ * versus a socket appears nowhere in here. What is being searched over is the operation and the
+ * payload, which are the same whatever moves the bytes; which wire is used is decided below the
+ * driver interface. Were the transport to leak in here there would have to be a


Where the transport?

jgaleotti · 2026-09-04T21:01:40Z

+            mainSize = groupsView()!!.sizeOfGroup(GroupsOfChildren.MAIN),
+            sqlSize = groupsView()!!.sizeOfGroup(GroupsOfChildren.INITIALIZATION_SQL),
+            mongoSize = groupsView()!!.sizeOfGroup(GroupsOfChildren.INITIALIZATION_MONGO),
+            redisSize = groupsView()!!.sizeOfGroup(GroupsOfChildren.INITIALIZATION_REDIS),


I think that we should keep redis outside ASYNC API until it is fully supported in REST

Dropped the redisSize parameter, the group is fixed at zero, and copyContent no longer measures it.

arcuri82 · 2026-09-08T10:56:26Z

     * a keyword either.
     */
-    private val DATA_KEYWORDS = setOf("const", "default", "enum", "example", "examples")
+    private val DATA_KEYWORDS = setOf(CONST, DEFAULT, ENUM, EXAMPLE, EXAMPLES)


recall what stated in for_developers.md. do not declare variables in the middle of the file, but at the top of the class

arcuri82 · 2026-09-08T10:56:45Z

     * document, so a field a service happens to call "const" or "default" must still be walked.
     */
-    private val SCHEMA_MAPS = setOf("properties", "patternProperties", "definitions", "\$defs")
+    private val SCHEMA_MAPS = setOf(PROPERTIES, PATTERN_PROPERTIES, DEFINITIONS, DEFS)


do not declare variables in the middle/end of the file

arcuri82 · 2026-09-08T11:00:00Z

+    mainSize: Int = allActions.size,
+    sqlSize: Int = 0,
+    mongoSize: Int = 0,
+    dnsSize: Int = 0,


we are adding several more DB to EvoMaster these days. no need to handle them here, for the moment. but can leave a TODO comment about it

Turns the parsed document into something the search can sample and mutate. AsyncApiAction is one message published on one channel, following the shape of RPCCallAction -- the closest analogue in EvoMaster, being a call with no URL, made through a driver, whose mutable state is the input and whose response is read afterwards rather than searched over. What is deliberately not a gene is the point of the design: - the address, which the contract fixes. Sending to a channel the service does not read would only waste executions. - the correlation id, stamped fresh at each execution. Searching over it could achieve nothing, since the service only echoes it, and pairing a reply with its request needs a value unique per execution rather than one carried in the genome. The header it is stamped into is therefore left out of the header genes, and when it is the only header there are no header genes at all -- an empty schema would otherwise build a free-form map and invite the search to invent headers the contract never declared. - the reply, which is an observation, read at fitness time. AsyncApiIndividual is one individual for every transport rather than one per transport: Kafka versus AMQP versus a socket appears nowhere in it. What is searched over is the operation and the payload, which are the same whatever moves the bytes. AsyncApiActionBuilder fills an action cluster, keeping the contract the REST and GraphQL builders use: one entry per action name, and a list of what had to be skipped rather than an exception. Only operations the service consumes become actions, since a 'send' is something to subscribe to rather than to drive. An operation carrying several messages becomes several actions -- which message to send is a choice made by picking an action, not by mutating a gene -- and the message is named in the action only when there is more than one, so a generated test reads as the operation where it can. That multi-message case is not hypothetical, but it does not arise the way one would expect. No document in the corpus narrows to more than one message with a `messages:` array; what happens is the opposite -- an operation declares no array at all, and so inherits everything its channel carries. scalar.yaml is the real case, vendored unmodified under MIT with its source noted: its userEvents channel carries five genuinely different payloads, and subscribeToUserEvents does not say which of them it means. Note RPCIndividual.seeIndexedRPCCalls is not copied over. It looks up children by action type, but the children are enterprise action groups, so it returns nothing; it has no callers. seeMainExecutableActions does the job.

It is vendored unmodified from github.com/scalar/scalar under MIT and names its source at the top of the file, but docs/for_developers.md also asks for third-party material to be listed in docs/reused_code.md.

Review feedback: the keywords read and written while turning a payload into genes -- properties, required, const, enum, type, minimum, maximum and the type names -- were literals at fifteen sites, two of them also inside the DATA_KEYWORDS and SCHEMA_MAPS sets. They are object-level constants now, beside INLINE_PREFIX, and the two sets are built from them. These are JSON Schema's vocabulary rather than AsyncAPI's, so they belong here and not with the parser's keywords.

@experimental

Review feedback: keep Redis out of AsyncAPI until it is fully supported for REST. Every Redis option in EMConfig is still @experimental, so a new problem type should not inherit that seam on day one. The redisSize parameter goes, the group size is fixed at zero, and copyContent no longer measures it. Mongo and DNS stay: neither is experimental. RPCIndividual and RestIndividual both still carry a Redis group; this is deliberately stricter than the classes it is modelled on.

Review feedback asked where the transport is. The KDoc said "below the driver interface" without naming it; it now says SutController.executeAsyncApiAction, the one driver method that publishes and awaits, so the reference can be found by search. The core hands it an address, a payload and a correlation id, and never learns what broker is behind them.

Review feedback: DATA_KEYWORDS and SCHEMA_MAPS were declared after every function, and AsyncApiAction.singleMessage after its first override. docs/for_developers.md puts fields first. The two sets now follow the keyword constants they are built from, and singleMessage opens the class body.

Review feedback: EvoMaster is adding several more databases; they need no handling here yet, but a TODO beside the three that are handled says where they will go.

Review against docs/for_developers.md: a suite is named after the class it exercises, and the four tests of AsyncApiIndividual were sitting in AsyncApiActionBuilderTest. Moved, unchanged, to AsyncApiIndividualTest.

addAction and removeAction took any position and left the group bookkeeping to object, which it does only when the index falls off the children altogether. With a setup action in front of the messages, removing "message -1" removed the setup action instead. Both now check the position against the main group, as RestIndividual does, and say so.

…ses microcks.yaml too addActionsFromSchema clears the cluster it is given before filling it, as the REST builder does; its KDoc now says so. reused_code.md said microcks.yaml was a test resource of the parser module only, but a copy sits in core as well since the action builder tests.

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from f5a5dd5 to fee5769 Compare August 25, 2026 16:11

jgaleotti force-pushed the feature/asyncapi-action-individual branch from fee5769 to cef304b Compare September 4, 2026 17:27

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from cef304b to bc8930e Compare September 4, 2026 18:44

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from bc8930e to 7f14a88 Compare September 4, 2026 19:18

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from 7f14a88 to 2fe6c8d Compare September 4, 2026 19:27

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from 2fe6c8d to 17bd03e Compare September 4, 2026 19:44

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from 17bd03e to d712fe2 Compare September 4, 2026 20:09

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from d712fe2 to 0b02805 Compare September 4, 2026 20:13

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from 0b02805 to 25df3bf Compare September 4, 2026 20:22

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from 25df3bf to d9376a5 Compare September 4, 2026 20:32

jgaleotti requested changes Sep 4, 2026

View reviewed changes

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from d9376a5 to ee8e1bf Compare September 4, 2026 21:24

LautaroPetaccio mentioned this pull request Sep 5, 2026

AsyncAPI 3.x: the driver contract for publishing a message #1738

Open

jgaleotti approved these changes Sep 7, 2026

View reviewed changes

jgaleotti requested a review from arcuri82 September 7, 2026 15:01

arcuri82 requested changes Sep 8, 2026

View reviewed changes

arcuri82 force-pushed the feature/asyncapi-action-individual branch from 60d5918 to 3e3dd4e Compare September 8, 2026 11:35

arcuri82 force-pushed the feature/asyncapi-action-individual branch from 3e3dd4e to 9709fab Compare September 8, 2026 11:41

arcuri82 force-pushed the feature/asyncapi-action-individual branch from 9709fab to 621abde Compare September 8, 2026 11:42

arcuri82 force-pushed the feature/asyncapi-action-individual branch from 621abde to 07e585e Compare September 8, 2026 11:45

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from 07e585e to 8fc83db Compare September 8, 2026 16:24

LautaroPetaccio force-pushed the feature/asyncapi-action-individual branch from 5e4064a to eec4004 Compare September 8, 2026 17:19

Base automatically changed from feature/asyncapi-payload-genes to master September 9, 2026 10:52

LautaroPetaccio added 10 commits September 9, 2026 12:52

AsyncAPI 3.x: list scalar.yaml in reused_code.md

c4871d4

It is vendored unmodified from github.com/scalar/scalar under MIT and names its source at the top of the file, but docs/for_developers.md also asks for third-party material to be listed in docs/reused_code.md.

AsyncAPI 3.x: note that more databases are coming

6c22b66

Review feedback: EvoMaster is adding several more databases; they need no handling here yet, but a TODO beside the three that are handled says where they will go.

AsyncAPI 3.x: the individual gets its own test suite

b708723

Review against docs/for_developers.md: a suite is named after the class it exercises, and the four tests of AsyncApiIndividual were sitting in AsyncApiActionBuilderTest. Moved, unchanged, to AsyncApiIndividualTest.

arcuri82 force-pushed the feature/asyncapi-action-individual branch from eec4004 to 07e181b Compare September 9, 2026 10:52

Navigation Menu

Uh oh!

AsyncAPI 3.x: the action and individual a search holds - #1711

AsyncAPI 3.x: the action and individual a search holds#1711
LautaroPetaccio wants to merge 10 commits into
masterfrom
feature/asyncapi-action-individual

LautaroPetaccio commented Aug 24, 2026 •

edited

Loading

Uh oh!

jgaleotti Sep 4, 2026

Uh oh!

LautaroPetaccio Sep 5, 2026

Uh oh!

jgaleotti Sep 4, 2026

Uh oh!

jgaleotti Sep 4, 2026

Uh oh!

LautaroPetaccio Sep 5, 2026

Uh oh!

arcuri82 Sep 8, 2026

Uh oh!

LautaroPetaccio Sep 8, 2026

Uh oh!

arcuri82 Sep 8, 2026

Uh oh!

LautaroPetaccio Sep 8, 2026

Uh oh!

arcuri82 Sep 8, 2026

Uh oh!

LautaroPetaccio Sep 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

LautaroPetaccio commented Aug 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

The action

The individual

The builder

One thing not copied across

Testing

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

LautaroPetaccio commented Aug 24, 2026 •

edited

Loading