Try to use pyghidra to get function addresses

main.py

@@ -5,32 +5,39 @@ import logging
 import os
 import tempfile
 import subprocess
+import pyghidra
+
 
 logging.basicConfig(level=logging.INFO)
 logging.info(f"CUDA available: {torch.cuda.is_available()}, CUDA version: {torch.version.cuda}")
 
+print("Starting pyghidra")
+pyghidra.start()
+
 GHIDRA_PROJECT_DIR = f"{os.getenv('HOME')}/ghidra_project"
 
 os.makedirs(GHIDRA_PROJECT_DIR, exist_ok=True)
 
 def get_functions(file):
 
-    with tempfile.TemporaryDirectory() as TEMP_DIR:
+    with pyghidra.open_program(file) as flat_api:
+        program = flat_api.getCurrentProgram()
+        function_addrs = [(f.getName(), f.getAddress().toOffset()) for f in program.getFunctionManager().getFunctions(True)]
 
-        o = subprocess.run(["/code/GhidraFunctionCPPExporter/export.bash", file, "output_dir", TEMP_DIR], shell=False, capture_output=True, encoding="utf8")
+        # o = subprocess.run(["/code/GhidraFunctionCPPExporter/export.bash", file, "output_dir", TEMP_DIR], shell=False, capture_output=True, encoding="utf8")
 
-        print(o.stdout)
-        print(o.stderr)
+        # print(o.stdout)
+        # print(o.stderr)
 
-        if (o.returncode != 0):
-            raise Exception(f"Ghidra export failed with return code {o.returncode}: {o.stderr}")
+        # if (o.returncode != 0):
+        #     raise Exception(f"Ghidra export failed with return code {o.returncode}: {o.stderr}")
 
-        # Show files in TEMP_DIR
-        print("Files in TEMP_DIR:")
-        for f in os.listdir(TEMP_DIR):
-            print(f)
+        # # Show files in TEMP_DIR
+        # print("Files in TEMP_DIR:")
+        # for f in os.listdir(TEMP_DIR):
+        #     print(f)
 
-    return o.stdout
+    return function_addrs
 
 
 # def decomp_create_prompt(input_data: str) -> str:
@@ -112,8 +119,6 @@ This is a space to experiment with our quantized 22B neural model for decompilat
     @file_widget.change(inputs=file_widget, outputs=[intro, state, col, fun_dropdown])
     def file_change_fn(file):
 
-        print("file change fn called with file:", file)
-
         if file is None:
             return {col: gr.update(visible=False), state: {"file": None}}
         else:
@@ -125,16 +130,15 @@ This is a space to experiment with our quantized 22B neural model for decompilat
                     desc=f"Analyzing binary  {os.path.basename(file.name)} with Ghidra...",
                 )
                 fun_data = get_functions(file.name)
-                # print(fun_data)
+                print(fun_data)
+
+                addrs = [
+                    (f"{name} ({hex(int(addr))}", int(addr))
+                    for name, addr in fun_data
+                ]
 
-                #addrs = [
-                #    (f"{name} ({hex(int(addr))}; {numvars} vars)", int(addr))
-                #    for addr, (name, cf, numvars) in fun_data.items()
-                #]
+                print(addrs)
 
-                #cfs = {name: cf for (name, cf, _numvars) in fun_data.values()}
-                addrs = []
-                cfs = {}
             except Exception as e:
                 print("error...", e)
                 raise gr.Error(f"Unable to analyze binary with Ghidra: {e}")
@@ -143,7 +147,7 @@ This is a space to experiment with our quantized 22B neural model for decompilat
                 col: gr.update(visible=True),
                 fun_dropdown: gr.Dropdown(choices=addrs, value=addrs[0][1] if addrs else None),
                 state: {"file": file,
-                        "cfs": cfs},
+                        "addrs": addrs},
             }
         
     #file_widget.change(