[WIP][REST] Extend the file format conversion APIs to support DSL items

[lolodomo]

Related to #4585

[lolodomo]

This PR is not yet ready for review because not yet finished. I consider it finished at around 75%.
I submitted it only because @mherwege wanted to know how it looks like.

[spacemanspiff2007]

It seems you decided to use the following naming schema:

 /file-format/{file-format-name}/parse
 /file-format/{file-format-name}/create

I think this is very good and I like it very much.

However for file-format-name you chose things and items.
These are the names for the core objects and not the file format and thus confusing.
I think it would be better if we use items-file and things-file.
What do you think?

[lolodomo]

/file-format/items is IMHO obvious that we are talking about file format for items. /file-format/items-file is redundant IMHO.. But if an official reviewer thinks your suggestion is better than mine, I will then change.

[spacemanspiff2007]

But GET /file-format/items will return the existing items in a certain file format.
There the name is used to indicate that one is accessing the object item and not the file format items.
That's very confusing:

GET  /file-format/items       <-- items mean openHAB items
POST /file-format/items/parse <-- items mean *.items file

[mherwege]

I don't want to steer this discussion too much and I would be fine with any conclusion on this, but here is another alternative that could be considered:

POST /file-format/parse
POST /file-format/create

As the 'what' is already in the MIME type (e.g. text/vnd.openhab.dsl.item), it is not strictly required here. And it also makes it possible in the future to just use something like text/vnd.openhab.dsl for a to be defined format that wraps multiple concepts in one file, without creating another end point.

[spacemanspiff2007]

For parse the input is always a string and the output schema could be set according to the MIME type (thank you @mherwege for showing that that's possible).
For create the output is always a string and the input schema could be set according to the MIME type.

That would also be fine with me.

[lolodomo]

For output, as type is Response, there to s no problem to return any DTO object inside the Response.
But for input (provided as body), I still don't know how you define the type of the input parameter for the method as it can be ItemXXXDTO or ThingXXXDTO depending on content type header

[lolodomo]

@mherwege : the day we have a YAML file format, media type will be application/yaml. This is not sufficient for the parse API to know if it should return a ItemXXXDTO or ThingXXXDTO.. Not sufficient to specify what kind of information you are requesting.

[lolodomo]

Or you imagine to rather define text/vnd.openhab.yaml.item and text/vnd.openhab.yaml.thing ?

[spacemanspiff2007]

I think it's important to differentiate between the existing ui yaml files and a possible new yaml format.
I suggest we somehow indicate that it's a ui yaml format.
So it would be text/yaml-ui.itemor text/yaml.ui.item or something like that.

[lolodomo]

I think it's important to differentiate between the existing ui yaml files and a possible new yaml format.

There is no plan to create 2 different YAML formats.

[spacemanspiff2007]

My understanding is that you want to parse the already existing yaml from the ui code tab.

That would be the currently existing UI yaml with tow sub formats for item and thing.

Then there is also discussion about a new file format which contains more information and thus is not compatible to the ui yaml.
That would be the new config yaml (which still has to be properly defined and created).
This format should then be parsed, too.

The naming should make it clear which of the yaml formats you are referring to.

[lolodomo]

My understanding is that you want to parse the already existing yaml from the ui code tab.

No, this never was my intention. Please re-read the pointed issue you apparently missed the general intention. My intention is to show DSL and new YAML file formats in Main UI code tab. Management of YAML would be moved from Main UI to core. Main UI would just call API to create/parse YAML.

[lolodomo]

Then there is also discussion about a new file format which contains more information and thus is not compatible to the ui yaml.

The UI YAML will disappear in my solution and be replaced by any file format.

[spacemanspiff2007]

Then that was a misunderstanding. I thought additionaly to the new yaml file the plan is to provide a parse function for the existing ui yamls so the forum examples will still work.

The new yaml format can contain both items and things at the same time.
Why would you need to differentiate between thing-yaml and items-yaml then?
That's why I thought they are the UI yamls because these are the only yamls where this differentiation exists.

[lolodomo]

Why would you need to differentiate between thing-yaml and items-yaml then?

Because UI need is to work either on item or thing, depending on the configuration WEB page.
But the YAML format will of course allow multiple kind of data. @J-N-K already provided an architecture allowing that easily.

One time again, we start discuss YAML in a PR which is not about YAML. Why not having this discussion rather in the issue itself?

[spacemanspiff2007]

One time again, we start discuss YAML in a PR which is not about YAML. Why not having this discussion rather in the issue itself?

Because the way the format looks has implications and obviously we're not on the same page how the format looks.
For me the differentiation between a items and a things yaml makes no sense because the format makes it clear what is what and depending on the input it's clear what should be generated.

So please let's have this discussion (not here) otherwise it will be an endless source of confusion

[mherwege]

Or you imagine to rather define text/vnd.openhab.yaml.item and text/vnd.openhab.yaml.thing ?

That was my thinking, but I am fine either way. You could then also have text/vnd.openhab.yaml which is a wrapper format that has it defined on a higher level with lists for things, items... So depending on the MIME type consumed, the outcome would be different. It has the disadvantage that these custom MIME types do not automatically get documented though. So I am on the fence.

For multiple types of objects content, if you want some analogies in the UI, we currently support working with this on multiple levels in the UI in a few places:

• Pages: you can look at and edit the code of a page, which is a YAML structure that includes the full definition of the page with all its components, but you can also look at and modify the YAML code on an individual component level.
• Rules: depends a bit on the rule type, but mostly the rule code page is a YAML with all triggers, actions (UI defined) and conditions, and if you have an inline script as an action, you will see the code under script

I would not get stuck on the exact format in the UI. That is a later concern. For now, I fully agree with @lolodomo that the first target should be making DSL import work without the parser code in the UI, just by calling a REST endpoint.

For output, as type is Response, there to s no problem to return any DTO object inside the Response.
But for input (provided as body), I still don't know how you define the type of the input parameter for the method as it can be ItemXXXDTO or ThingXXXDTO depending on content type header

I believe you can define multiple possible input MIME types as well in the @Consumes annotation and use @RequestBody to document it. I called in ChatGPT for some help:

@POST
@Path("/upload")
@Consumes({MediaType.APPLICATION_JSON, MediaType.APPLICATION_XML})
@Produces(MediaType.APPLICATION_JSON)
@Operation(summary = "Upload data", description = "Accepts JSON or XML input")
@ApiResponses(value = {
    @ApiResponse(responseCode = "200", description = "Success"),
    @ApiResponse(responseCode = "400", description = "Invalid input")
})
public Response uploadData(
    @RequestBody(description = "Input data", required = true, content = {
        @Content(mediaType = "application/json", schema = @Schema(implementation = MyData.class)),
        @Content(mediaType = "application/xml", schema = @Schema(implementation = MyData.class))
    }) MyData data) {
    return Response.ok().entity("Data uploaded").build();
}

[lolodomo]

In you example, the type is the same in both cases, MyData.
But ok, we can create a new MyData class that contain a collection of items and a collection of things. That was I believe more or less what @spacemanspiff2007 also suggested.
In that case, I agree we don't need to mention what is as input or what to provide at output when the file format supports both items and things (like YAML in the future(. As DSL format cannot contain items and things in the same fle, we need two different MIME types to distinguish items and things So we should keep text/vnd.openhab.dsl.item and text/vnd.openhab.dsl.thing and in the future add only text/vnd.openhab.yaml.
So yes I am going to switch to these unique APIs:
POST /file-format/parse
POST /file-format/create

[mherwege]

You can achieve that with something like this:

@POST
@Path("/upload")
@Consumes({MediaType.APPLICATION_JSON, MediaType.APPLICATION_XML})
public Response uploadData(@Context HttpHeaders headers, String rawData) {
    String contentType = headers.getHeaderString(HttpHeaders.CONTENT_TYPE);

    if (MediaType.APPLICATION_JSON.equals(contentType)) {
        // Process JSON input
    } else if (MediaType.APPLICATION_XML.equals(contentType)) {
        // Process XML input
    } else {
        return Response.status(Response.Status.UNSUPPORTED_MEDIA_TYPE).entity("Unsupported format").build();
    }
    
    return Response.ok("Processed successfully").build();
}

[lolodomo]

For create API, content type will be only JSON and output will accept text/vnd.openhab.dsl.item and text/vnd.openhab.dsl.thing.
For parse API, content type will accept text/vnd.openhab.dsl.item and text/vnd.openhab.dsl.thing and output will be only JSON.

[lolodomo]

For the errors, yes you have line and column. See in the current code:
https://github.com/openhab/openhab-core/blob/main/bundles/org.openhab.core.model.core/src/main/java/org/openhab/core/model/core/internal/ModelRepositoryImpl.java#L289
For warnings, I am not sure:

[spacemanspiff2007]

The argument pro including it is that the REST endpoint for items also includes it I believe.

Yes - that's why I'd put it there, too.
Metadata in the file formats is always item bound so there's no use case for providing metadata without and item, in fact its most likely an error.
Imho metadata is only separate for historical reasons and today metadata and items are so intertwined, that this separation doesn't make any sense any more. So putting it in the item is the right way to do it.
Additionally this allows copy pasting rest responses without transformation.

For warnings, I am not sure:

That would require to return a json:

{
  "file": "FileText",
  "warnings": ["SomeWarningDto"]
}

I'm not really sure it it's a valid use case.
The format generation should always return a file that loads without warning and if the input is ambiguous it should raise an error.
Format parsing should always parse the file, warnings can be silently dropped there.
@mherwege do you have a specific case in mind where the warnings are of interest?

[lolodomo]

The format generation should always return a file that loads without warning

We are talking about the other way.
There is no warning for file format generation.
There are warnings for file format parsing.

[kaikreuzer]

@lolodomo Is this still WIP as the title suggests?

[lolodomo]

@lolodomo Is this still WIP as the title suggests?

Yes, it is still in progress.

[kaikreuzer]

Then maybe also convert it to a "draft" PR, that makes it more explicit.

[lolodomo]

Finally I plan to add a unique new API: POST /file-format/convert
It takes a format at input and convert it into the output format.
Input and output formats are defined by media types in request headers.

[jimtng]

Is this now superseded by #4779?

[lolodomo]

Yes, partially.. This one contains also stuff for DSL items that I will reuse later.
I will rebase it when the other is merged.

[lolodomo]

I rebased this PR, it now depends on #4779 and only add DSL items as DSL things are already covered by #4779 .

[jimtng]

I tried this and noticed that item's yaml generator is not implemented yet. However, item DSL generator works!

[lolodomo]

I tried this and noticed that item's yaml generator is not implemented yet.

Yes, because PR for YAML items is not yet merged.

[jimtng]

I tried this and noticed that item's yaml generator is not implemented yet.

Yes, because PR for YAML items is not yet merged.

is this what it needs: #4776 ?

[lolodomo]

I tried this and noticed that item's yaml generator is not implemented yet.

Yes, because PR for YAML items is not yet merged.

is this what it needs: #4776 ?

This PR does not require #4776 as it deals only with DSL items.
But to consider YAML items in the future, of course, we need first #4776 and few additional lines of code.
I could already create a PR but that would require to combine two not yet merged PRs, a little complicated to do, I prefer to wait for at least one of these PRs to be merged.
I am at the point I am now slowing down until some of my code is first reviewed and merged.

[lolodomo]

I am closing this PR, a new PR #4793 covers its scope.