feat(cli): bug report synthesis (#131)

doc/cli.md

@@ -121,6 +121,26 @@ yes | nnsmith.model_gen model.type=torch mgen.method=symbolic-cinit \
                                          debug.viz=true
 ```
 
+## Synthesize bug reports
+
+`nnsmith.report_syn` can synthesize bug reports given a model (read from filesystem) and a backend target (user provided).
+It prints a self-contained Python script:
+
+1. For most of the cases, it should be able to directly reproduce the bug.
+2. If not, it should serve as a good starting point and developers can modify it a bit to reproduce the bug.
+
+> **Note**
+>
+> This is an experimental feature and only works for PyTorch models.
+
+> **Warning**
+>
+> `nnsmith.report_syn` is not guaranteed to reproduce the bug. For strict bug reproduction, please use `nnsmith.model_exec` instead.
+
+```shell
+nnsmith.report_syn backend.type="pt2 backend@inductor" model.type=torch model.path=nnsmith_output/model.pth
+```
+
 ## Misc
 
 TensorFlow logging can be very noisy. Use `TF_CPP_MIN_LOG_LEVEL=3` as environmental variable to depress that.

nnsmith/cli/report_syn.py

@@ -0,0 +1,61 @@
+import os
+
+import hydra
+from omegaconf import DictConfig, ListConfig
+
+from nnsmith.backends import BackendFactory
+from nnsmith.logging import RENDER_LOG
+from nnsmith.materialize import Model, Render
+
+
+@hydra.main(version_base=None, config_path="../config", config_name="main")
+def main(cfg: DictConfig):
+    RENDER_LOG.warning(
+        "The duty of `nnsmith.report_syn` is to produce a BASIC but executable Python script. It may not reproduce the original bug as the report may not use the original seed, input data, and output oracles. If you want to more strictly reproduce the bug, please use `nnsmith.model_exec`."
+    )
+
+    cmp_cfg = cfg["cmp"]
+    model_cfg = cfg["model"]
+    ModelType = Model.init(model_cfg["type"], cfg["backend"]["target"])
+
+    if isinstance(model_cfg["path"], ListConfig):
+        model_paths = model_cfg["path"]
+    else:
+        model_paths = [model_cfg["path"]]
+
+    for model_path in model_paths:
+        model = ModelType.load(model_path)
+
+        oracle_path = None
+        # Check if we can directly use oracle from `oracle.pkl`
+        if "auto" == cmp_cfg["oracle"]:
+            model_basename = os.path.basename(os.path.normpath(model_path))
+            oracle_path = model_path.replace(model_basename, "oracle.pkl")
+            if not os.path.exists(oracle_path):
+                oracle_path = None
+        elif cmp_cfg["oracle"] is not None:
+            oracle_path = cmp_cfg["oracle"]
+
+        if not os.path.exists(oracle_path):
+            oracle_path = None
+
+        this_fac = BackendFactory.init(
+            cfg["backend"]["type"],
+            target=cfg["backend"]["target"],
+            optmax=cfg["backend"]["optmax"],
+            parse_name=True,
+        )
+
+        render = Render()
+        render.emit_model(model)
+        render.emit_input(model, oracle_path)
+        render.emit_backend(this_fac)
+
+        print("#", "-" * 20)
+        print(f"# {model_path}")
+        print(render.render())
+        print("#", "-" * 20)
+
+
+if __name__ == "__main__":
+    main()

nnsmith/logging.py

@@ -6,6 +6,7 @@
 SMT_LOG = logging.getLogger("smt")
 EXEC_LOG = logging.getLogger("exec")
 DTEST_LOG = logging.getLogger("dtest")
+RENDER_LOG = logging.getLogger("render")
 CORE_LOG = logging.getLogger("core")
 
 TF_LOG = logging.getLogger("gen|tf")

nnsmith/materialize/torch/__init__.py

@@ -190,7 +190,7 @@ def import_libs(self) -> List[str]:
 
     def emit_input(self, inp_name: str, path: Optional[PathLike] = None):
         if path is not None:  # Assume NumPy tensors as inputs
-            return f"{inp_name} = [v for _, v in pickle.load(open('{path}', 'rb'))['input']]"
+            return f"{inp_name} = [v for _, v in pickle.load(open('{path}', 'rb'))['input'].items()]"
 
         # Path is None. Generate inputs from scratch.
         tensor_text = []

setup.cfg

@@ -42,3 +42,4 @@ console_scripts =
     nnsmith.model_exec = nnsmith.cli.model_exec:main
     nnsmith.dtype_test = nnsmith.cli.dtype_test:main
     nnsmith.fuzz = nnsmith.cli.fuzz:main
+    nnsmith.report_syn = nnsmith.cli.report_syn:main