wirl-lang 0.1.0

WIRL workflow DSL grammar and parser

wirl-lang — Workflow DSL (BNF + Parser)

This package contains the grammar and parser for the WIRL workflow DSL. The DSL compiles to an explicit directed graph with optional guarded loops. This subfolder is self-contained: a BNF grammar and a Python parser that produces a structured AST you can feed into an engine.

Scope of this folder

• Grammar: grammar/wirl.bnf
• Parser and AST dataclasses: grammar/wirl_parser.py
• Public API: parse_wirl_to_objects(path: str) -> Workflow

Quick start

from wirl_lang import parse_wirl_to_objects

workflow = parse_wirl_to_objects(
    "workflow_definitions/paper_rename_workflow/paper_rename_workflow.wirl"
)

print(workflow.name)
for node in workflow.nodes:
    print(type(node).__name__, getattr(node, "name", None))

DSL at a glance

• Workflow file defines exactly one runnable graph: workflow <Name> { ... }.
• Nodes are external function calls with declared inputs/outputs and immutable const { ... }.
• Data dependencies are implicit: a node becomes eligible once its inputs are available.
• No shared mutable state. All values are immutable except outputs with reducer (append) which accumulate into a list.
• Optional inputs/outputs use a trailing ? in their declaration. Optional inputs do not create execution dependencies.
• Guarded loops via cycle with a guard { ... } clause and max_iterations: <int>.
• Conditional execution via when { <boolean-expr> } on nodes and guards.
• Human-in-the-loop hitl { ... } and retry { ... } are present in the grammar; see status below.

Example (excerpt)

This is a shortened excerpt of the working workflow in workflow_definitions/paper_rename_workflow/paper_rename_workflow.wirl:

workflow PaperRenameWorkflow {

  metadata {
    description: "Paper Rename workflow"
    owner: "madmag77"
    version: "1.0"
    files_extension: "pdf"
  }

  inputs {
    String drafts_folder_path
    String processed_folder_path
  }

  outputs {
    List<String> processed_files = ReturnProcessedFiles.processed_files
  }

  node GetFiles {
    call get_files
    inputs {
      String drafts_folder_path = drafts_folder_path
    }
    outputs {
      List<String> file_paths
    }
  }

  cycle RenameLoop {
    inputs {
        List<String> initial_file_paths_to_process = GetFiles.file_paths
        List<String> remaining_file_paths_to_process = ReadPdfFile.remaining_file_paths
        String processed_folder_path = processed_folder_path
    }
    outputs {
        List<String> processed_files = CheckAllFilesProcessed.processed_files
    }

    node ReadPdfFile {
        call read_pdf_file
        inputs {
           List<T<Image>> file_paths = RenameLoop.remaining_file_paths_to_process?
           List<String> initial_file_paths_to_process = RenameLoop.initial_file_paths_to_process
        }
        const {
            pages_to_read: 2
        }
        outputs {
           List<T<Image>> pages
           String file_path
           List<String> remaining_file_paths
           Bool no_files_to_process
        }
    }

    ...

    guard {
      inputs {
        Bool is_done = CheckAllFilesProcessed.is_done
      }
      when {
        CheckAllFilesProcessed.is_done
      }
    }
    max_iterations: 10
  }

Grammar highlights

• Top-level:
  • workflow <Name> { metadata? inputs? outputs? node* cycle* }
• Nodes:
  • node <Name> { call <module_or_func> inputs { ... } outputs { ... } const { ... } when { ... } hitl { ... } retry { ... } }
• Parameters:
  • Declarations: TYPE name (= value)? ?
  • Values: literal or reference OtherNode.outputName
  • Optional: trailing ? (e.g., String clarifications?)
  • Reducers on outputs: (append) TYPE name or default (last)
• Cycles:
  • cycle <Name> { inputs { ... } outputs { ... } node* guard { inputs { ... } when { ... } } max_iterations: INT }
• Conditionals:
  • when { <boolean-expr> } on nodes and inside guards

Parser API

• parse_wirl_to_objects(path: str) -> Workflow
  • Returns a Workflow dataclass with:
    • name: str
    • metadata: Optional[Metadata] (entries: Dict[str, str])
    • inputs: List[Input], outputs: List[Output]
    • nodes: List[NodeClass | CycleClass]
  • NodeClass: name, call, inputs, outputs, when, hitl, retry, constants
  • CycleClass: name, inputs, outputs, nodes, guard, max_iterations
  • Output.reducer: "last" (default) or "append"

Type system

• Supported tokens in the grammar: Bool, Int, Float, String, File, Object<...>, List<T> (free-form TYPE token covers these and generics).
• Current parser does not enforce types; it records the declared type name string.

Feature status

• Implemented in grammar and parser:
  • Workflows, nodes, inputs/outputs, constants, reducers (append)
  • when { ... } conditions
  • cycle with guard { ... } and max_iterations
• Present in grammar but not fully wired in the AST yet:
  • retry { attempts, backoff, policy } (parsed token exists; transformer wiring pending)
  • hitl { correlation, timeout } (accepted; current transformer sets a placeholder/default)
• Planned validations (not yet implemented):
  • Missing required inputs, unused outputs, type mismatches
  • Single entry/exit checks and reachability

Design notes (execution semantics)

• One file ⇒ one graph with a single entry and a single exit.
• Dataflow drive: node executes when all non-optional inputs become available and when block is evaluated to True (if any).
• All values are immutable; (append) outputs accumulate across iterations/paths.
• Guarded loops model Pregel-style supersteps; parallelism can be expressed with multiple nodes eligible in the same step. A parallel block may be added in the future.

When Block Evaluation Rules

The when blocks use special truthiness evaluation that differs from standard Python boolean logic:

• False conditions: Only None or explicit False evaluate to false
• True conditions: All other values evaluate to true, including:
  • Empty containers: [], {}, ""
  • Zero values: 0, 0.0
  • Objects with custom types not defined in the evaluation context

This design ensures that:

1. Only explicit absence (None) or negation (False) prevents node execution
2. Empty results from previous nodes don't block subsequent processing
3. Type definitions don't need to be available during condition evaluation

Examples:

node ProcessIfResults {
  when {
    DataLoader.items  // True even if items is []
  }
}

node SkipIfDisabled {
  when {
    Config.enabled  // False only if None or explicit False
  }
}

node HandleCustomObjects {
  when {
    Parser.objects  // True even if objects contains undefined types
  }
}

Local development

• Edit the grammar in grammar/wirl.bnf and the transformer in grammar/wirl_parser.py.
• After grammar changes, re-run your parser-driven test or the quick-start snippet above.

FAQ

• Can I reference optional outputs? Yes. Declare the consuming input as optional with a trailing ? and assign from Producer.optionalOutput.
• Do optional inputs block execution? No. Absence of an optional input does not prevent node scheduling.
• How do I append across iterations? Mark the output with (append) in its declaration, e.g., (append) List<String> processed_files.

License

This package is released under the repository’s root license.