Automated API traversal - program synthesis

If computers could code interactions with systems for us, we wouldn't have to write code, they could interoperate automatically

YAML 問題

We have a REST API or standard library functions in our chosen programming language. Each rest call or method call takes arguments and returns data and causes side effects. Take a HTTP client for instance, it takes data in and returns data.

We typically have to write interactions with APIs manually.

What if the programming against APIs could be automatically generated and the source code or AST generated to interact with the API.

We need a way to map a simple definition of the problem into a more detailed representation of the problem which uses APIs

In other words, computers program themselves. Rather than general intelligence we rely on mappings and equivalencies and graph traversal and scheduling.

Say I want to backup files on all my computers, compress them, encrypt them, sign them, deduplicate them, upload.

That's a lot of APIs I need to talk to (encryption, operating system file APIs) and the APIs for them require multiple lines of code to perform. There's also a data flow through a pipeline of steps.

The call graph of the code to fulfil the pipeline looks a certain way.

Each step, "encrypt" takes in an input and has an output.

These need to be defined.

I want to be able to define a list of steps to perform and have the computer synthesise all the intermediate and data transformation steps in between.

I give the computer this list: { "steps": ["deduplicate", "compress", "encrypt", "upload"] }

The computer generates code to do the following, performing all intermediate conversions such as files to bytes.

Upload(encrypt(compress(deduplicate(input)))

The actual code may look more like

A(b(c(d(e(f(g(h(I(j(k(input)))))))))

But this is a simplified example, upload might be multiple steps or multiple steps in the AST.

To implement this I need a database of records representing APIs and their inputs, outputs and types.

We also need a database of code sequences (or code snippets) which are named sequences of code that perform an overarching task - such as signing or encrypting some bytes A sequence references API method calls in order and had variables used between method calls.

We also have sequences to convert between types. And collections of types

I tell the computer what I want to be done in a simple specification of the problem and the computer works out what APIs need to be used to perform the actions. It's synthesis of a set of steps or operators to perform a function.


沒有子分類。

投票 (可選) (別通知) (可選)
請,登錄

我應該指出,OpenAPI 是可用於遍歷 API 的其餘 API 的規範格式,唯一的問題是它不是很好。

如果程序宣傳其功能、類型、輸入和輸出,我們就可以讓系統自行編程以與其他系統對話。

I should point out that OpenAPI is a specification format for rest APIs that could be used to traverse APIs the only problem is that it isn't very good.

If programs advertised their functionality, types, inputs and outputs we could have systems that program themselves to talk to other systems.



    : Inyuki
    :  -- 
    :  -- 
    

chronological,

我理解你想要什麼,並且理解這樣的程序合成必須是在沒有程序員在循環中的情況下是可能的。我什至嘗試編寫軟件,使來自軟件庫和網站的 API 更具可重用性,我稱之爲“metadrive”。我認爲我們需要驅動程序,因爲並非每個 API 都被記錄在案,而且標準也不相同,還記得 UseMe Files 文件的想法嗎?一旦我們解決了自動重用性,函數圖 是絕對可以遍歷的,並且獲得最終結果的執行路徑就像地理地圖上的路線一樣。這是一個多層次的問題,但絕對可以解決。

I understand what you want, and understand that such program synthesis must be possible without a programmer in the loop. I had even attempted to write software for making APIs from software libraries and websites more reusable, which I call "metadrive". I think we need drivers, because not every API is documented and not in the same standard, remember UseMe Files files idea? Once we solve automatic reusability, the graph of functions can definitely be traversed, and paths of execution to get final result composed like routes on a geographical map. Heck of a multilevel problem, but definitely solvable.


UseMe 文件絕對是一個與這個想法相關的應用程序。我真的很喜歡 UseMe 文件。

這個想法中未解決的問題是將一個圖擴展到另一個圖。

沒有定義“加密”是什麼意思 - 沒有將以下代碼標記爲“加密” -

生成密鑰 獲取算法 Algorithm.encrypt(輸入)

您不能將“加密”擴展到這些步驟。您必須做大量工作才能說明加密的含義。

也許我們可以使用類型來解決這個問題。只要引用正確的類型,系統就可以推斷出正確的代碼操作。

UseMe files are definitely an application/related to this idea. I really like UseMe files.

The unsolved problem in this idea is the expanding of one graph to another.

Without defining what "encrypt" means - without marking the following code as "encrypt" -

Generate key Get algorithm Algorithm.encrypt(input)

You cannot expand "encrypt" to those steps. You've got to do a lot of work to indicate what encrypt means.

Maybe we can use types to solve this problem. As long as you refer to the correct types, the system can infer the correct code operations.



    :  -- 
    : Mindey, Inyuki
    :  -- 
    

chronological,

語言