Vedanuj Goswami

40 papers A* 9A 2C 1Journal 19Unranked 9
YearRankTypeTitle / Venue / Authors
2025 J jnl
CoRR
Rylan Schaeffer, Punit Singh Koura, Binh Tang, Ranjan Subramanian, Aaditya K. Singh, Todor Mihaylov, Prajjwal Bhargava, Lovish Madaan, Niladri S. Chatterji, Vedanuj Goswami, Sergey Edunov, Dieuwke Hupkes, Sanmi Koyejo, Sharan Narang
2025 A* conf
ISCA
Weiwei Chu, Xinfeng Xie, Jiecao Yu, Jie Wang, Amar Phanishayee, Chunqiang Tang, Yuchen Hao, Jianyu Huang, Mustafa Ozdal, Jun Wang, Vedanuj Goswami, Naman Goyal, Abhishek Kadian, Andrew Gu, Chris Cai, Feng Tian, Xiaodong Wang, Min Si, Pavan Balaji, Ching-Hsiang Chu, Jongsoo Park
2023 conf
ACL (1)
Uri Shaham, Maha Elbayad, Vedanuj Goswami, Omer Levy, Shruti Bhosale
2023 conf
EACL (Findings)
Haoran Xu, Jean Maillard, Vedanuj Goswami
2023 J jnl
CoRR
Haoran Xu, Jean Maillard, Vedanuj Goswami
2023 J jnl
CoRR
Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton-Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Hartshorn, Saghar Hosseini, Rui Hou, Hakan Inan, Marcin Kardas, Viktor Kerkez, Madian Khabsa, Isabel Kloumann, Artem Korenev, Punit Singh Koura, Marie-Anne Lachaux, Thibaut Lavril, Jenya Lee, Diana Liskovich, Yinghai Lu, Yuning Mao, Xavier Martinet, Todor Mihaylov, Pushkar Mishra, Igor Molybog, Yixin Nie, Andrew Poulton, Jeremy Reizenstein, Rashi Rungta, Kalyan Saladi, Alan Schelten, Ruan Silva, Eric Michael Smith, Ranjan Subramanian, Xiaoqing Ellen Tan, Binh Tang, Ross Taylor, Adina Williams, Jian Xiang Kuan, Puxin Xu, Zheng Yan, Iliyan Zarov, Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan Narang, Aurélien Rodriguez, Robert Stojnic, Sergey Edunov, Thomas Scialom
2023 A conf
INTERSPEECH
Mohamed Anwar, Bowen Shi, Vedanuj Goswami, Wei-Ning Hsu, Juan Pino, Changhan Wang
2023 J jnl
CoRR
Mohamed Anwar, Bowen Shi, Vedanuj Goswami, Wei-Ning Hsu, Juan Pino, Changhan Wang
2023 J jnl
CoRR
Hongyu Gong, Ning Dong, Sravya Popuri, Vedanuj Goswami, Ann Lee, Juan Pino
2023 A* conf
EMNLP
Mikel Artetxe, Vedanuj Goswami, Shruti Bhosale, Angela Fan, Luke Zettlemoyer
2023 J jnl
CoRR
Mikel Artetxe, Vedanuj Goswami, Shruti Bhosale, Angela Fan, Luke Zettlemoyer
2023 conf
ACL (1)
Jean Maillard, Cynthia Gao, Elahe Kalbassi, Kaushik Ram Sadagopan, Vedanuj Goswami, Philipp Koehn, Angela Fan, Francisco Guzmán
2023 conf
ACL (1)
Paul-Ambroise Duquenne, Hongyu Gong, Ning Dong, Jingfei Du, Ann Lee, Vedanuj Goswami, Changhan Wang, Juan Pino, Benoît Sagot, Holger Schwenk
2023 conf
EMNLP (Findings)
Haoran Xu, Maha Elbayad, Kenton Murray, Jean Maillard, Vedanuj Goswami
2023 J jnl
CoRR
Haoran Xu, Maha Elbayad, Kenton Murray, Jean Maillard, Vedanuj Goswami
2022 J jnl
CoRR
Uri Shaham, Maha Elbayad, Vedanuj Goswami, Omer Levy, Shruti Bhosale
2022 A* conf
CVPR
Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, Douwe Kiela
2022 J jnl
CoRR
Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Y. Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loïc Barrault, Gabriel Mejia Gonzalez, Prangthip Hansanti, John Hoffman, Semarley Jarrett, Kaushik Ram Sadagopan, Dirk Rowe, Shannon Spruit, Chau Tran, Pierre Andrews, Necip Fazil Ayan, Shruti Bhosale, Sergey Edunov, Angela Fan, Cynthia Gao, Vedanuj Goswami, Francisco Guzmán, Philipp Koehn, Alexandre Mourachko, Christophe Ropers, Safiyyah Saleem, Holger Schwenk, Jeff Wang
2022 J jnl
CoRR
Paul-Ambroise Duquenne, Hongyu Gong, Ning Dong, Jingfei Du, Ann Lee, Vedanuj Goswami, Changhan Wang, Juan Miguel Pino, Benoît Sagot, Holger Schwenk
2022 conf
NAACL-HLT
Dheeru Dua, Shruti Bhosale, Vedanuj Goswami, James Cross, Mike Lewis, Angela Fan
2021 A* conf
ICLR
Songwei Ge, Vedanuj Goswami, Larry Zitnick, Devi Parikh
2021 J jnl
CoRR
Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, Douwe Kiela
2021 A* conf
NeurIPS
Sasha Sheng, Amanpreet Singh, Vedanuj Goswami, Jose Alberto Lopez Magana, Tristan Thrush, Wojciech Galuba, Devi Parikh, Douwe Kiela
2021 J jnl
CoRR
Sasha Sheng, Amanpreet Singh, Vedanuj Goswami, Jose Alberto Lopez Magana, Wojciech Galuba, Devi Parikh, Douwe Kiela
2021 A* conf
ICLR
Duy-Kien Nguyen, Vedanuj Goswami, Xinlei Chen
2021 A conf
WACV
Laura Sevilla-Lara, Shengxin Zha, Zhicheng Yan, Vedanuj Goswami, Matt Feiszli, Lorenzo Torresani
2021 J jnl
CoRR
Dheeru Dua, Shruti Bhosale, Vedanuj Goswami, James Cross, Mike Lewis, Angela Fan
2020 A* conf
CVPR
Jiasen Lu, Vedanuj Goswami, Marcus Rohrbach, Devi Parikh, Stefan Lee
2020 J jnl
CoRR
Amanpreet Singh, Vedanuj Goswami, Devi Parikh
2020 A* conf
KDD
Dheevatsa Mudigere, Maxim Naumov, Joe Spisak, Geeta Chauhan, Narine Kokhlikyan, Amanpreet Singh, Vedanuj Goswami
2020 J jnl
CoRR
Songwei Ge, Vedanuj Goswami, C. Lawrence Zitnick, Devi Parikh
2020 J jnl
CoRR
Duy-Kien Nguyen, Vedanuj Goswami, Xinlei Chen
2020 conf
NeurIPS (Competition and Demos)
Douwe Kiela, Hamed Firooz, Aravind Mohan, Vedanuj Goswami, Amanpreet Singh, Casey A. Fitzpatrick, Peter Bull, Greg Lipstein, Tony Nelli, Ron Zhu, Niklas Muennighoff, Riza Velioglu, Jewgeni Rose, Phillip Lippe, Nithin Holla, Shantanu Chandra, Santhosh Rajamanickam, Georgios Antoniou, Ekaterina Shutova, Helen Yannakoudakis, Vlad Sandulescu, Umut Ozertem, Patrick Pantel, Lucia Specia, Devi Parikh
2020 A* conf
NeurIPS
Douwe Kiela, Hamed Firooz, Aravind Mohan, Vedanuj Goswami, Amanpreet Singh, Pratik Ringshia, Davide Testuggine
2020 J jnl
CoRR
Douwe Kiela, Hamed Firooz, Aravind Mohan, Vedanuj Goswami, Amanpreet Singh, Pratik Ringshia, Davide Testuggine
2019 J jnl
CoRR
Jiasen Lu, Vedanuj Goswami, Marcus Rohrbach, Devi Parikh, Stefan Lee
2019 J jnl
CoRR
Laura Sevilla-Lara, Shengxin Zha, Zhicheng Yan, Vedanuj Goswami, Matt Feiszli, Lorenzo Torresani
2019 conf
ICCV Workshops
Albert Pumarola, Vedanuj Goswami, Francisco Vicente, Fernando De la Torre, Francesc Moreno-Noguer
2016 C conf
ICCBR
Spencer Rugaber, Shruti Bhati, Vedanuj Goswami, Evangelia Spiliopoulou, Sasha Azad, Sridevi Koushik, Rishikesh Kulkarni, Mithun Kumble, Sriya Sarathy, Ashok K. Goel
2015 conf
ACALCI
Vedanuj Goswami, Samir Borgohain
redb/extractors/decompiler/DecompileBinja.py
← Index redb/extractors/decompiler/DecompileBinja.py python
import hashlib
import inspect
import json
import logging
import os
import signal
import time
from datetime import datetime, timezone
from typing import Dict, Any, Optional
from pathlib import Path
import subprocess
import sys

from redb.extractors.enum import Tag
from redb.extractors.extractor import Extractor
import magic
import pefile
from elftools.elf.elffile import ELFFile

# Import our BinjaDecompiler (conditional)
from redb.extractors.decompiler.bninja.decompiler import BinaryNinjaDecompiler

class DecompileBinja(Extractor):
    def __init__(
        self,
        filepath,
        log,
        exporters=None,
        index_prefix=None,
        elastic_index=None,
        known_benign=False,
        known_malicious=False,
        filetype=None,
        decompile_modules=None,
    ):
        super().__init__(
            filepath,
            log,
            exporters,
            index_prefix,
            elastic_index,
            known_benign,
            known_malicious,
        )
        self.log.debug(inspect.currentframe().f_code.co_name)

        # Check if Binary Ninja is available
       # if not BINARYNINJA_AVAILABLE:
        #     self.log.error("Binary Ninja is not available in this container")
        #    raise ImportError(
        #        "Binary Ninja module not found - not available in feature extraction container"
        #    )
        self.analysis_results = None
        self.binja_decompiler = None
        self.filetype = filetype
        self.decompile_modules = decompile_modules or {"all"}
        self.goresym_data = None
        self.goresym_output_path = None

        # Convert TIMEOUT to integer with a default of 1200 seconds (20 minutes)
        try:
            self.BINJA_TIMEOUT = int(os.getenv("BINJA_TIMEOUT", "1200"))
        except ValueError:
            self.log.warning(
                "Invalid BINJA_TIMEOUT value, using default of 1200 seconds"
            )
            self.BINJA_TIMEOUT = 1200

        # Convert TIMEOUT to integer with a default of 1200 seconds (20 minutes)
        try:
            self.DECOMPILE_EXTRACTOR_TIMEOUT = int(
                os.getenv("DECOMPILE_EXTRACTOR_TIMEOUT", "2580")
            )
        except ValueError:
            self.log.warning(
                "Invalid DECOMPILE_EXTRACTOR_TIMEOUT value, using default of 2580 seconds"
            )
            self.DECOMPILE_EXTRACTOR_TIMEOUT = 2580

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_val, exc_tb):
        self.cleanup_run()

    def calculate_md5(self, input_str):
        """Calculate MD5 hash of a string."""
        return hashlib.md5(input_str.encode("utf-8")).hexdigest()

    def is_dotnet(self):
        """Check if the binary is a .NET assembly.

        Returns:
            bool: True if the file is a .NET assembly, False otherwise
        """
        try:
            if self.filetype == "pebin":
                file_type = magic.from_buffer(self.binary)
                if ".Net" in file_type:
                    return True
                pe = pefile.PE(self.filepath)
                for entry in pe.OPTIONAL_HEADER.DATA_DIRECTORY:
                    # IMAGE_DIRECTORY_ENTRY_COM_DESCRIPTOR is typically 14
                    if (
                        entry.name == "IMAGE_DIRECTORY_ENTRY_COM_DESCRIPTOR"
                        and entry.Size > 0
                    ):
                        return True
                return False
            return False
        except AttributeError as e:
            self.log.error(
                f"AttributeError error dotnet file {self.hash.sha256} Full error : {e}"
            )
            return False

    def is_golang(self):
        """Check if the binary is a Go-compiled binary (heuristic).

        Supports PE and ELF binaries.
        """
        try:
            if self.filetype == "pebin":
                pe = pefile.PE(self.filepath)
                signatures = [b"Go build ID:", b"runtime.main", b"main.main"]

                for section in pe.sections:
                    data = section.get_data()
                    if any(sig in data for sig in signatures):
                        return True

                return False

            elif self.filetype == "elf":
                with open(self.filepath, "rb") as f:
                    elf = ELFFile(f)

                    # 1. Section-based checks
                    section_names = [sec.name for sec in elf.iter_sections()]
                    if any(
                        s in section_names
                        for s in (".note.go.buildid", ".gopclntab")
                    ):
                        return True

                    # 2. String scan in loadable sections
                    signatures = [
                        b"Go build ID:",
                        b"runtime.main",
                        b"runtime.goexit",
                        b"runtime.morestack",
                        b"main.main",
                    ]

                    for sec in elf.iter_sections():
                        if sec["sh_flags"] & 0x2:  # SHF_ALLOC
                            data = sec.data()
                            if any(sig in data for sig in signatures):
                                return True

                return False

            return False

        except Exception as e:
            self.log.error(
                f"Golang detection error {self.hash.sha256}: {e}"
            )
            return False

    def cleanup_run(self):
        """Clean up after analysis."""
        try:
            # BinaryNinjaDecompiler uses context manager pattern (__enter__/__exit__)
            # Cleanup happens automatically when exiting the 'with' block
            self.binja_decompiler = None

            # Clean up goresym temp file if it exists (keep in debug mode)
            if self.goresym_output_path and os.path.exists(self.goresym_output_path):
                if self.log.isEnabledFor(logging.DEBUG):
                    self.log.debug(f"Debug mode: keeping goresym output at {self.goresym_output_path}")
                else:
                    os.remove(self.goresym_output_path)
                    self.goresym_output_path = None

            # Force garbage collection
            import gc

            gc.collect()

        except Exception as e:
            self.log.error(f"Error in cleanup: {e}")

    def run_goresym(self, binary_path, output_json_path):
        """
            Run goresym on a Go binary and export its JSON output to a file.
            binary_path: path to the Go binary to analyze
            output_json_path: path where the JSON output will be saved
            """
        binary_path = str(Path(binary_path).resolve())
        output_json_path = str(Path(output_json_path).resolve())
        goresym_path = os.getenv("GORESYM_PATH", "GoReSym")

        try:
            # Example: goresym -t json /path/to/binary
            self.log.info("DEBUG: starting GoReSym")
            result = subprocess.run(
                [goresym_path, binary_path],
                stdout=subprocess.PIPE,
                stderr=subprocess.PIPE,
                text=True,
            )
        except FileNotFoundError:
            self.log.error("Error: 'goresym' not found in PATH. Make sure it is installed.")
            raise
        except subprocess.CalledProcessError as e:
            self.log.error("Error during goresym execution:")
            self.log.error(e.stderr)


        # Assuming goresym emits valid JSON to stdout.
        try:
            parsed = json.loads(result.stdout)
            # Store parsed JSON for later export to ClickHouse
            self.goresym_data = parsed
        except json.JSONDecodeError:
            # If it's not valid JSON, save the raw output instead.
            self.log.error("Warning: goresym output is not valid JSON; saving raw.")
            with open(output_json_path, "w", encoding="utf-8") as f:
                f.write(result.stdout)
            return

        # Save pretty-printed JSON for readability and debugging (used by BinaryNinja)
        with open(output_json_path, "w", encoding="utf-8") as f:
            json.dump(parsed, f, ensure_ascii=False, indent=2)
        self.goresym_output_path = output_json_path

        self.log.debug(f"goresym output saved to: {output_json_path}")

    def analyze_binary(self) -> Optional[Dict[str, Any]]:
        """Run Binary Ninja analysis and return results."""
        self.log.debug("Starting binary analysis")
        # if the binary is dotnet (only PE)
        if self.is_dotnet():
            self.log.debug("Skipping .NET binary - decompilation not supported")
            return None

        output_json_path = None
        ## if golang: run goresym
        try:
            if self.is_golang():
                output_json_path = "./goResym.json"
                self.run_goresym(self.filepath, output_json_path)
        except Exception as e:
            self.log.error(f"Error on GoReSym extraction: {e}")

        try:
            # Use BinaryNinjaDecompiler as a context manager to ensure proper setup/cleanup
            with BinaryNinjaDecompiler(
                filepath=self.filepath,
                timeout=self.BINJA_TIMEOUT,
                log=self.log,
                exporters=self.exporters,
                index_prefix=self.index_prefix,
                filetype=self.filetype,
                goresym=output_json_path,
                decompile_modules=self.decompile_modules,
            ) as decompiler:
                self.binja_decompiler = decompiler

                if decompiler.extract():
                    # Store results before context manager exits
                    results = decompiler.analysis_results
                    return results
                else:
                    self.log.error("BinaryNinjaDecompiler extraction failed")
                    return None

        except Exception as e:
            self.log.error(f"Error in Binary Ninja analysis: {e}")
            import traceback
            self.log.error(f"Traceback: {traceback.format_exc()}")
            return None

        finally:
            self.cleanup_run()

    def extract(self):
        """Extract and process all analysis results."""
        self.log.debug(inspect.currentframe().f_code.co_name)

        # Create a flag to track if extraction completed
        extraction_completed = False
        extraction_result = False
        extraction_error = None

        # Define the extraction process as a separate function
        def do_extraction():
            nonlocal extraction_completed, extraction_result, extraction_error
            try:
                results = self.analyze_binary()
                if not results:
                    extraction_result = False
                else:
                    self.analysis_results = results
                    # Add file hashes from parent Extractor class to analysis results
                    self.analysis_results["sha256"] = self.sha256
                    self.analysis_results["sha1"] = self.sha1
                    self.analysis_results["md5"] = self.md5
                    extraction_result = True
            except Exception as e:
                extraction_error = e
                extraction_result = False
            finally:
                extraction_completed = True

        # Run extraction directly with signal-based timeout (no thread overhead).
        # SIGALRM is delivered by the OS, so there's no GIL contention or polling.
        old_handler = signal.getsignal(signal.SIGALRM)
        def _timeout_handler(signum, frame):
            raise TimeoutError("Extraction timed out")

        signal.signal(signal.SIGALRM, _timeout_handler)
        signal.alarm(self.DECOMPILE_EXTRACTOR_TIMEOUT)
        try:
            do_extraction()
        except TimeoutError:
            self.log.error(
                f"Extraction timed out after {self.DECOMPILE_EXTRACTOR_TIMEOUT} seconds"
            )
            self.cleanup_run()
            return None
        finally:
            signal.alarm(0)
            signal.signal(signal.SIGALRM, old_handler)

        if extraction_error:
            self.log.error(f"Error in extraction: {extraction_error}")
            return None

        # Return the actual analysis results, not just a boolean
        return self.analysis_results if extraction_result else None

    def prepare_export_data(self, exporter_type: str) -> Any:
        """Prepare data for database export."""
        self.log.debug(inspect.currentframe().f_code.co_name)
        if not self.analysis_results:
            return None

        # # Delegate to the BinjaDecompiler for consistent export formatting
        # if self.binja_decompiler:
        #     return self.binja_decompiler.prepare_export_data(exporter_type)
        # else:
        #     self.log.error("BinjaDecompiler not available for export preparation")
        #     return None

        if exporter_type == "ClickHouseExporter":
            now = datetime.now(timezone.utc)

            def prepare_array_field(value, array_type):
                """Helper to prepare array fields with proper null handling"""
                if value is None:
                    return []
                return value

            # Add a helper function to handle empty strings
            def ensure_not_empty(value, default="UNKNOWN"):
                """Ensure a string value is not empty"""
                if value is None or value == "":
                    return default
                return value

            def prepare_register_usage_map(register_dict):
                """Convert register usage dict to Map format with tuples
                Input: {"rbx": {"reads": 3, "writes": 1}, ...}
                Output: {"rbx": (3, 1), ...}
                """
                if not register_dict:
                    return {}
                return {
                    reg: (info.get("reads", 0), info.get("writes", 0))
                    for reg, info in register_dict.items()
                }


            decompile_modules = getattr(self, "decompile_modules", {"all"})
            run_all = "all" in decompile_modules
            run_decompilation = run_all or "decompilation" in decompile_modules
            run_disassembly = run_all or "disassembly" in decompile_modules
            run_llil = run_all or "llil" in decompile_modules
            run_cfg = run_all or "cfg" in decompile_modules
            run_strings = run_all or "strings" in decompile_modules

            export = {"multi_table": True}

            # Decompilation tables
            if run_decompilation:
                export["decompiled_content"] = {
                    "table": "code_binja_decompiled_functions_content",
                    "data": [
                        [
                            f["decompiled_function_hash"],
                            f["decompiled_function"],
                            f["function_type"],
                            f.get("flattened_score"),
                            f.get("mba_score"),
                            now,
                        ]
                        for f in self.analysis_results["decompiled"]
                    ],
                    "column_names": [
                        "decompiled_function_hash",
                        "decompiled_function",
                        "function_type",
                        "flattened_score",
                        "mba_score",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "String",
                        "Enum8('USER'=1, 'LIBRARY'=2, 'THUNK'=3, 'EXTERNAL'=4, 'UNKNOWN'=5)",
                        "Nullable(Float64)",
                        "Nullable(Float64)",
                        "DateTime64(3, 'UTC')",
                    ],
                }
                export["decompiled_refs"] = {
                    "table": "code_binja_decompiled_functions_references",
                    "data": [
                        [
                            self.analysis_results["sha256"],
                            self.analysis_results["sha1"],
                            self.analysis_results["md5"],
                            f["decompiled_function_hash"],
                            f.get("disassembled_function_hash"),
                            f["decompiled_function_name"],
                            f["decompiled_function_prototype"],
                            f["decompiled_function_address"],
                            prepare_array_field(f.get("functions_caller"), "Array(String)"),
                            prepare_array_field(f.get("functions_call"), "Array(String)"),
                            now,
                        ]
                        for f in self.analysis_results["decompiled"]
                    ],
                    "column_names": [
                        "sha256",
                        "sha1",
                        "md5",
                        "decompiled_function_hash",
                        "disassembled_function_hash",
                        "decompiled_function_name",
                        "decompiled_function_prototype",
                        "decompiled_function_address",
                        "functions_caller",
                        "functions_call",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "FixedString(40)",
                        "FixedString(32)",
                        "FixedString(64)",
                        "Nullable(FixedString(64))",
                        "LowCardinality(String)",
                        "LowCardinality(String)",
                        "UInt64",
                        "Array(String)",
                        "Array(String)",
                        "DateTime64(3, 'UTC')",
                    ],
                }

            # Disassembly tables
            if run_disassembly:
                export["disassembled_content"] = {
                    "table": "code_binja_disassembled_functions_content",
                    "data": [
                        [
                            f["disassembled_function_hash"],
                            f.get("disassembled_function", ""),
                            f.get("disassembled_function_no_addresses", ""),
                            f.get("function_type", "UNKNOWN"),
                            f.get("instructions_count", 0),
                            prepare_array_field(
                                f.get("instructions_types"), "LowCardinality(String)"
                            ),
                            f.get("control_flow_count", 0),
                            prepare_array_field(
                                f.get("memory_access_pattern"), "LowCardinality(String)"
                            ),
                            prepare_array_field(
                                f.get("register_usage"), "LowCardinality(String)"
                            ),
                            f.get("data_references_count", 0),
                            f.get("max_block_size"),
                            f.get("num_calls"),
                            f.get("stack_size"),
                            now,
                        ]
                        for f in self.analysis_results["disassembled"]
                    ],
                    "column_names": [
                        "disassembled_function_hash",
                        "disassembled_function",
                        "disassembled_function_no_addresses",
                        "function_type",
                        "instructions_count",
                        "instructions_types",
                        "control_flow_count",
                        "memory_access_pattern",
                        "register_usage",
                        "data_references_count",
                        "max_block_size",
                        "num_calls",
                        "stack_size",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "String",
                        "String",
                        "Enum8('USER'=1, 'LIBRARY'=2, 'THUNK'=3, 'EXTERNAL'=4, 'UNKNOWN'=5)",
                        "UInt32",
                        "Array(LowCardinality(String))",
                        "UInt32",
                        "Array(LowCardinality(String))",
                        "Array(LowCardinality(String))",
                        "UInt32",
                        "Nullable(UInt32)",
                        "Nullable(UInt32)",
                        "Nullable(Int32)",
                        "DateTime64(3, 'UTC')",
                    ],
                }
                export["disassembled_refs"] = {
                    "table": "code_binja_disassembled_functions_references",
                    "data": [
                        [
                            self.analysis_results["sha256"],
                            self.analysis_results["sha1"],
                            self.analysis_results["md5"],
                            f["disassembled_function_hash"],
                            f.get("decompiled_function_hash"),
                            f["disassembled_function_name"],
                            f["disassembled_function_address"],
                            f.get("tlsh_disassembly"),
                            f.get("tlsh_llil"),
                            now,
                        ]
                        for f in self.analysis_results["disassembled"]
                    ],
                    "column_names": [
                        "sha256",
                        "sha1",
                        "md5",
                        "disassembled_function_hash",
                        "decompiled_function_hash",
                        "disassembled_function_name",
                        "disassembled_function_address",
                        "tlsh_disassembly",
                        "tlsh_llil",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "FixedString(40)",
                        "FixedString(32)",
                        "FixedString(64)",
                        "Nullable(FixedString(64))",
                        "LowCardinality(String)",
                        "UInt64",
                        "Nullable(FixedString(72))",
                        "Nullable(FixedString(72))",
                        "DateTime64(3, 'UTC')",
                    ],
                }
                # Function similarity metrics table (derived from disassembly data)
                # Lookup maps to join LLIL/MLIL features by disassembled_function_hash
                llil_by_hash = {
                    l.get("disassembled_function_hash"): l
                    for l in self.analysis_results.get("llil", [])
                    if l and l.get("disassembled_function_hash")
                }
                mlil_by_hash = {
                    m.get("disassembled_function_hash"): m
                    for m in self.analysis_results.get("mlil", [])
                    if m and m.get("disassembled_function_hash")
                }

                export["function_similarity_metrics"] = {
                    "table": "code_binja_function_similarity_metrics",
                    "data": [
                        [
                            f["disassembled_function_hash"],
                            f.get("cyclomatic_complexity"),
                            f.get("tlsh_disassembly"),
                            f.get("tlsh_llil"),
                            prepare_array_field(f.get("minhash"), "Array(UInt8)"),
                            (llil_by_hash.get(f["disassembled_function_hash"]) or {}).get("tlsh_llil"),
                            (llil_by_hash.get(f["disassembled_function_hash"]) or {}).get(
                                "tlsh_instruction_typed_llil"),
                            prepare_array_field(
                                (llil_by_hash.get(f["disassembled_function_hash"]) or {}).get("minhash_llil_skeleton"),
                                "Array(UInt8)",
                            ),
                            prepare_array_field(
                                (llil_by_hash.get(f["disassembled_function_hash"]) or {}).get("minhash_llil_typed"),
                                "Array(UInt8)",
                            ),
                            (mlil_by_hash.get(f["disassembled_function_hash"]) or {}).get("tlsh_mlil_skeleton"),
                            (mlil_by_hash.get(f["disassembled_function_hash"]) or {}).get("tlsh_mlil_typed"),
                            prepare_array_field(
                                (mlil_by_hash.get(f["disassembled_function_hash"]) or {}).get("minhash_mlil_skeleton"),
                                "Array(UInt8)",
                            ),
                            prepare_array_field(
                                (mlil_by_hash.get(f["disassembled_function_hash"]) or {}).get("minhash_mlil_typed"),
                                "Array(UInt8)",
                            ),
                            now,
                        ]
                        for f in self.analysis_results.get("disassembled", [])
                    ],
                    "column_names": [
                        "disassembled_function_hash",
                        "cyclomatic_complexity",
                        "tlsh_disassembly",
                        "tlsh_llil",
                        "minhash",
                        "tlsh_llil_new",
                        "tlsh_instruction_typed_llil",
                        "minhash_llil_skeleton",
                        "minhash_llil_typed",
                        "tlsh_mlil_skeleton",
                        "tlsh_mlil_typed",
                        "minhash_mlil_skeleton",
                        "minhash_mlil_typed",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "Nullable(UInt16)",
                        "Nullable(FixedString(72))",
                        "Nullable(FixedString(72))",
                        "Array(UInt8)",
                        # new
                        "Nullable(FixedString(72))",
                        "Nullable(FixedString(72))",
                        "Array(UInt8)",
                        "Array(UInt8)",
                        "Nullable(FixedString(72))",
                        "Nullable(FixedString(72))",
                        "Array(UInt8)",
                        "Array(UInt8)",
                        "DateTime64(3, 'UTC')",
                    ],
                }


            # LLIL tables
            if run_llil:
                export["llil_content"] = {
                    "table": "code_binja_llil_functions_content",
                    "data": [
                        [
                            f["sha256_llil"],
                            f["function_type"],
                            prepare_array_field(
                                f.get("instructions_types_llil"), "LowCardinality(String)"
                            ),
                            f.get("control_flow_count_llil", 0),
                            prepare_array_field(
                                f.get("memory_access_pattern_llil"), "LowCardinality(String)"
                            ),
                            prepare_register_usage_map(f.get("register_usage", {})),
                            f.get("total_reg_reads", 0),
                            f.get("total_reg_written", 0),
                            f.get("data_references_count", 0),
                            f.get("max_block_size"),
                            f.get("num_calls"),
                            f.get("stack_size"),
                            prepare_array_field(f.get("body_llil_vector"), "Array(Tuple(UInt32, Array(UInt16)))"),
                            now,
                        ]
                        for f in self.analysis_results.get("llil", [])
                    ],
                    "column_names": [
                        "llil_function_hash",
                        "function_type",
                        "instructions_types_llil",
                        "control_flow_count_llil",
                        "memory_access_pattern_llil",
                        "register_usage_llil",
                        "total_reg_reads",
                        "total_reg_written",
                        "data_references_count",
                        "max_block_size",
                        "num_calls",
                        "stack_size",
                        "body_llil_vector",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "Enum8('USER'=1, 'LIBRARY'=2, 'THUNK'=3, 'EXTERNAL'=4, 'UNKNOWN'=5)",
                        "Array(LowCardinality(String))",
                        "UInt32",
                        "Array(LowCardinality(String))",
                        "Map(LowCardinality(String), Tuple(UInt32, UInt32))",
                        "UInt32",
                        "UInt32",
                        "UInt32",
                        "Nullable(UInt32)",
                        "Nullable(UInt32)",
                        "Nullable(Int32)",
                        "Array(Tuple(UInt32, Array(UInt16)))",
                        "DateTime64(3, 'UTC')",
                    ],
                }
                export["llil_refs"] = {
                    "table": "code_binja_llil_functions_references",
                    "data": [
                        [
                            self.analysis_results["sha256"],
                            self.analysis_results["sha1"],
                            self.analysis_results["md5"],
                            f.get("sha256_llil"),
                            f.get("disassembled_function_hash"),
                            f.get("function_address"),
                            f.get("tlsh_disassembly"),
                            f.get("tlsh_llil"),
                            now,
                        ]
                        for f in self.analysis_results.get("llil", [])
                    ],
                    "column_names": [
                        "sha256",
                        "sha1",
                        "md5",
                        "llil_function_hash",
                        "disassembled_function_hash",
                        "function_address",
                        "tlsh_disassembly",
                        "tlsh_llil",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "FixedString(40)",
                        "FixedString(32)",
                        "Nullable(FixedString(64))",
                        "FixedString(64)",
                        "UInt64",
                        "Nullable(FixedString(72))",
                        "Nullable(FixedString(72))",
                        "DateTime64(3, 'UTC')",
                    ],
                }

            # Errors table (always include if per-function loop ran)
            if run_decompilation or run_disassembly or run_llil or run_cfg:
                export["function_analysis_errors"] = {
                    "table": "new_function_analysis_errors_binja",
                    "data": [
                        [
                            self.analysis_results["sha256"],
                            f["function_name"],
                            f["function_address"],
                            f.get("error_location", "unknown"),
                            f.get("error_message", ""),
                            f.get("error_details", ""),
                            f.get("error_type", "unknown"),
                            self.calculate_md5(
                                f"{f.get('error_message', '')}{f['function_name']}{f['function_address']}{f.get('error_location', 'unknown')}"
                            ),
                            "new",
                            now,
                        ]
                        for f in self.analysis_results.get("errors", [])
                    ],
                    "column_names": [
                        "sha256",
                        "function_name",
                        "function_address",
                        "error_location",
                        "error_message",
                        "error_details",
                        "error_type",
                        "error_hash",
                        "status",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "Nullable(String)",
                        "UInt64",
                        "LowCardinality(String)",
                        "Nullable(String)",
                        "Nullable(String)",
                        "Nullable(String)",
                        "FixedString(32)",
                        "Enum8('new'=1, 'investigating'=2, 'fixed'=3, 'wontfix'=4)",
                        "DateTime64(3, 'UTC')",
                    ],
                }

            # Strings table
            if run_strings:
                export["strings_raw"] = {
                    "table": "code_binja_strings_raw",
                    "data": [
                        [
                            self.analysis_results["sha256"],
                            s["string"],
                            s["string_raw"],
                            s["string_encoding"],
                            s["string_offset"],
                            s["string_length"],
                            s["string_raw_length"],
                            s["string_entropy"],
                        ]
                        for s in self.analysis_results.get("strings", [])
                    ],
                    "column_names": [
                        "sha256",
                        "string",
                        "string_raw",
                        "string_encoding",
                        "string_offset",
                        "string_length",
                        "string_raw_length",
                        "string_entropy",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "String",
                        "String",
                        "LowCardinality(String)",
                        "UInt64",
                        "UInt32",
                        "UInt32",
                        "Float32",
                    ],
                }

            # CFG function-level features table
            if run_cfg:
                export["cfg_functions"] = {
                    "table": "code_binja_cfg_functions",
                    "data": [
                        [
                            cfg.get("disassembled_function_hash"),
                            cfg["cfg_topology_hash"],
                            cfg["block_count"],
                            cfg["edge_count"],
                            cfg.get("llil_total_operations", 0),
                            cfg.get("call_count", 0),
                            cfg["cyclomatic_complexity"],
                            cfg.get("loop_count", 0),
                            cfg.get("max_depth", 0),
                            cfg.get("max_fan_out", 0),
                            cfg.get("md_index_topdown", 0),
                            cfg.get("md_index_bottomup", 0),
                            cfg.get("prime_product_llil", 0),
                            cfg.get("cfg_feature_tlsh"),
                            cfg.get("wl_minhash", []),
                            cfg.get("bb_features", []),
                            cfg.get("cfg_adjacency", []),
                            now,
                        ]
                        for cfg in self.analysis_results.get("cfg", [])
                        if cfg is not None
                    ],
                    "column_names": [
                        "disassembled_function_hash",
                        "cfg_topology_hash",
                        "block_count",
                        "edge_count",
                        "llil_total_operations",
                        "call_count",
                        "cyclomatic_complexity",
                        "loop_count",
                        "max_depth",
                        "max_fan_out",
                        "md_index_topdown",
                        "md_index_bottomup",
                        "prime_product_llil",
                        "cfg_feature_tlsh",
                        "wl_minhash",
                        "bb_features",
                        "cfg_adjacency",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "FixedString(16)",
                        "UInt16",
                        "UInt16",
                        "UInt32",
                        "UInt16",
                        "UInt16",
                        "UInt8",
                        "UInt16",
                        "UInt8",
                        "UInt64",
                        "UInt64",
                        "UInt64",
                        "Nullable(FixedString(72))",
                        "Array(UInt8)",
                        "Array(Array(UInt16))",
                        "Array(UInt32)",
                        "DateTime64(3, 'UTC')",
                    ],
                }

            # GoReSym metadata table (only if goresym data exists)
            if self.goresym_data:
                export["golang_metadata"] = {
                    "table": "redb_golang_metadata",
                    "data": [
                        [
                            self.analysis_results["sha256"],
                            json.dumps(self.goresym_data),
                            now,
                        ]
                    ],
                    "column_names": [
                        "sha256",
                        "goresym",
                        "analysis_date",
                    ],
                    "column_type_names": [
                        "FixedString(64)",
                        "JSON",
                        "DateTime64(3, 'UTC')",
                    ],
                }

            return export

    def tag(self) -> str:
        """Return the tag for this extractor."""
        return Tag.DECOMPILED.value

    def get_clickhouse_table(self) -> str:
        """Not used directly as we're handling multiple tables."""
        pass


if __name__ == "__main__":
    # Setup basic logging
    import logging
    import time

    logging.basicConfig(level=logging.INFO)
    logger = logging.getLogger("DecompileBinja")

    # Parse command line arguments
    import argparse

    parser = argparse.ArgumentParser(description="Binary Ninja Decompiler Wrapper")
    parser.add_argument("filepath", help="Path to the binary file to analyze")
    parser.add_argument(
        "--output", "-o", help="Output JSON file path (default: stdout)"
    )
    parser.add_argument(
        "--timeout",
        "-t",
        type=int,
        default=1200,
        help="Analysis timeout in seconds (default: 1200)",
    )
    args = parser.parse_args()
    start = time.perf_counter()
    # Create and run the extractor
    with DecompileBinja(args.filepath, logger) as extractor:
        success = extractor.extract()
        end = time.perf_counter()
        if not success:
            logger.error("Analysis failed")
            exit(1)

        # Output results
        if args.output:
            with open(args.output, "w") as f:
                json.dump(extractor.analysis_results, f)
            logger.info(f"Results written to {args.output}")
        else:
            print((extractor.analysis_results))
            with open("diff", "w") as f:
                f.write(str(f"{end - start:.3f} seconds"))