search

yonaskolb / SwagGen

OpenAPI/Swagger 3.0 Parser and Swift code generator
MIT License
625 stars 147 forks source link

Illegal Instruction: 4 on valid OAS 3.0 #136

Open dskvr opened 5 years ago

dskvr commented 5 years ago

Unfortunately, I have nothing else to report because using --verbose does not result in any additional information. My gut feeling tells me swaggen doesn't account for any types (ie: {}), which are valid OAS 3.0 types. 

$ swagger-cli validate swagger/chain.yaml
swagger/chain.yaml is valid
$ swaggen generate swagger/chain.yaml
Illegal Instruction: 4
$ swaggen version
Version: 4.0.0

Update: chain.yaml includes local references, so I tried a bundled/dereferenced version of the same specification with the exact same result.

yonaskolb commented 5 years ago

Hi @dskvr, thanks for the report. Unfortunately without a an example spec I can’t debug this. If you could put together an example that fails that would be great.

dskvr commented 5 years ago

Sure thing @yonaskolb, here's a dereferenced version of the specification. bundled-deref.yaml.zip

yonaskolb commented 5 years ago

Thanks @dskvr. Looks like the issue is the empty security:

security:
  - {}

If that is removed it's all good. That obviously shouldn't crash the parser, and will be fixed, but as a workaround you can just remove that

dskvr commented 5 years ago

Thanks!

Is there a good way to debug SwagGen on my own?

yonaskolb commented 5 years ago

To debug in Xcode you can checkout the repo and then

swift package generate-xcodeproj

Then in the swaggen scheme you can add a command line option

generate path/to/spec.yml

And then just run

dskvr commented 5 years ago

Thanks @yonaskolb!

Sidenote: I got Swaggen to generate, but the serialization logic appears to stop if it hits an anyOf member. In my case, it creates a public var trx with type Trx but Trx does not exist.

The spec I've provided is a pretty extreme case, it trips up serialization of almost every code generator I've come across aside from throwing the json-schemas (models) that make up this OAS spec into quicktype.io. It's entirely possible something is amidst in the spec despite validation (we all know how OAS 3.0 is right now...)

When I've isolated the cause in a way that can be succinctly expressed, I'll open a new issue.

Thanks for your help!

yonaskolb commented 5 years ago

Reopening this as the crash is still an issue that needs to be fixed