Custom Functions - Part 2

[lvca]

Part 1: #281

ArcadeDB already supports custom functions written in any language supported by the JVM. The most common is Javascript and it's available out of the box.

In order to define a new function "math" in Javascript you can execute this (This syntax has been inherited from OrientDB):

define function math.sum \"return a + b\" parameters [a,b] language js

Since functions are not persistent, every time a server starts, the function must be registered. Also, there is no easy way to write and run functions from Studio or an IDE.

I believe custom functions will play an important role in the future of ArcadeDB, mainly because:

• it's hard to write a general-purpose query optimizer on graphs
• complex queries are hard for beginners because they require a deep knowledge of ArcadeDB (or Cypher)
• complex queries are hard to debug

What if ArcadeDB provides the support of IDEs and a minimal code editor in Studio? Also, if possible, the ability to debug the code in the server would be extremely helpful for the developer.

IDE Support

In order to use any IDE with ArcadeDB functions, the easiest way would be to have a new folder "functions" under the database directory containing the code of the functions in files with the suffix .js for Javascript. In this way, you can work with any IDE. A file database.js could be created by default containing all the functions (with no implementation) of the supported functions callable from ArcadeDB. Example:

database.js

function database_lookupDocumentByRID( databaseInstance, rid ) {
}

The IDE could load this file and provide a syntax completion out of the box. Of course, this file must be updated with the real server-side functions available on the Server.

customer.js

function getById( customerId ) {
  let result = database_lookupDocumentByKey( db, "Customer", ["id"], [ customerId ]  );
  return result.length > 0 ? result[0] : null;
}

The provided JS functions could mimic the current Java API with the class name as a prefix separated by an underscore. This would avoid using OO in Javascript.

At startup, the ArcadeDB server will load all the files with extensions .js from the functions directory and registers the functions. The functions above would be translated as:

define function customer.getById
"let result = database_lookupDocumentByKey( db, \"Customer\", [\"id\"], [ customerId ]  ); return result.length > 0 ? result[0] : null;" 
parameters [customerId] language js

Note that the file name becomes the library (customer) where the function is registered.

And the user can call the function from SQL (console, studio, postgres driver, etc), such as:

select `customer.getById( 12345 )` as customer

Or SQL from Java with a prepared statement:

final ResultSet result = database.command("sql", "select `customer.getById(?)` as customer", 12345);

Or directly via Java API:

Vertex customer = database.getSchema().getFunction("customer", "getById").execute(12345);

Thoughts? Ideas? WDYT?

Contributions are welcome!

[mdre]

It a great Idea. We have a lot of functions defined in OrientDB that must be ported.
Consider to do the same but with a Java. We have a FullPath function that returns all the path found between two vertex. Now I hack the code and add the functions direct inside ArcadeDB but it is not a reccomended way.

[lvca]

@mdre how did you write such functions with OrientDB? Java classes you put under the lib directory of OrientDB?

[mdre]

We put it in the plugins directory.
Here is the repo:

https://github.com/mdre/FullPath

[lvca]

Thanks @mdre, out of this scope, I'm considering adding your function to ArcadeDB's graph library. What's the graph algorithm you have used to find all the paths?

[tolgaulas]

@lvca , should you take the path of "functions" under database folder for script files, would you pls consider making it either a variable that we can set anywhere we like or functionable with symbolic link so we can point from wherever we like ( we keep things under the project repos. With orientdb we have setup_db script which init a db along with the existing embedded functions; then we can store our functions in similar fashion.)

[lvca]

a symbolic link would work, but having in a separate settings it's something we can also easily do ;-)

[ansarizafar]

I like this idea. Is there any update on this feature?

[mdre]

I think it's a great idea. Sorry but I was full time on other projects. I'am nearly starting again with Arcade. To answer Luca's question, I implemented the OServerPluginAbstract and put it in the lib. If you can mimic the same interface it could be great to have full speed in functions. See you soon. El lun, 20 nov 2023 a las 11:33, Zafar Ansari ***@***.***>) escribió:

…

Part 1: #281 <#281> ArcadeDB already supports custom functions written in any language supported by the JVM. The most common is Javascript and it's available out of the box. In order to define a new function "math" in Javascript you can execute this (This syntax has been inherited from OrientDB): define function math.sum \"return a + b\" parameters [a,b] language js Since functions are not persistent, every time a server starts, the function must be registered. Also, there is no easy way to write and run functions from Studio or an IDE. I believe custom functions will play an important role in the future of ArcadeDB, mainly because: - it's hard to write a general-purpose query optimizer on graphs - complex queries are hard for beginners because they require a deep knowledge of ArcadeDB (or Cypher) - complex queries are hard to debug What if ArcadeDB provides the support of IDEs and a minimal code editor in Studio? Also, if possible, the ability to debug the code in the server would be extremely helpful for the developer. IDE Support In order to use any IDE with ArcadeDB functions, the easiest way would be to have a new folder "functions" under the database directory containing the code of the functions in files with the suffix .js for Javascript. In this way, you can work with any IDE. A file database.js could be created by default containing all the functions (with no implementation) of the supported functions callable from ArcadeDB. Example: database.js function database_lookupDocumentByRID( databaseInstance, rid ) {} The IDE could load this file and provide a syntax completion out of the box. Of course, this file must be updated with the real server-side functions available on the Server. customer.js function getById( customerId ) { let result = database_lookupDocumentByKey( db, "Customer", ["id"], [ customerId ] ); return result.length > 0 ? result[0] : null;} The provided JS functions could mimic the current Java API with the class name as a prefix separated by an underscore. This would avoid using OO in Javascript. At startup, the ArcadeDB server will load all the files with extensions .js from the functions directory and registers the functions. The functions above would be translated as: define function customer.getById"let result = database_lookupDocumentByKey( db, \"Customer\", [\"id\"], [ customerId ] ); return result.length > 0 ? result[0] : null;" parameters [customerId] language js Note that the file name becomes the library (customer) where the function is registered. And the user can call the function from SQL (console, studio, postgres driver, etc), such as: select `customer.getById( 12345 )` as customer Or SQL from Java with a prepared statement: final ResultSet result = database.command("sql", "select `customer.getById(?)` as customer", 12345); Or directly via Java API: Vertex customer = database.getSchema().getFunction("customer", "getById").execute(12345); Thoughts? Ideas? WDYT? Contributions are welcome! I like this idea. Is there any update on this feature? — Reply to this email directly, view it on GitHub <#1155 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABIRCDHIM7YOL5LJDRPA3KDYFNS3PAVCNFSM6AAAAAAZ5FSHF6VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TMMRQGIYTC> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[mdre]

There's no algorithm because it just a full path. I only do a deep traverse till to node or max deep and continue with backtracking from the last node. It is really important to use the filter with the include/exclude filters to limit the function. The git repo is this: https://github.com/mdre/FullPath El lun, 20 nov 2023 a las 19:11, Luca Garulli ***@***.***>) escribió:

…

Thanks @mdre, out of this scope, I'm considering adding your function to ArcadeDB's graph library. What's the graph algorithm you have used to find all the paths? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[lvca]

Do you take care of circular references?

[mdre]

Yes. I prevent finding if a node exists in the current path before it is added. If it exists it is ignored because it is processing in the backtracking. When a path is found, it is added to the resulting collection of paths. El mar, 21 nov 2023 a las 0:47, Luca Garulli ***@***.***>) escribió:

…

Do you take care of circular references? — Reply to this email directly, view it on GitHub <#1155 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABIRCDGKFZHN5E6PZB4DG5LYFQP3JAVCNFSM6AAAAAAZ5FSHF6VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TMMRVHEYTK> . You are receiving this because you were mentioned.Message ID: ***@***.***>