dotscope 0.6.0

//! Header fixup pass for PE file generation.
//!
//! This module provides functions to update PE headers after all content has been written.
//! Fixups are applied as the final step before checksum calculation, patching headers
//! with values that were unknown at initial write time.
//!
//! # Overview
//!
//! The fixup pass runs after all PE sections, metadata, and method bodies have been written
//! to the output file. It performs final header adjustments that require knowledge of the
//! complete file content:
//!
//! - **Optional Header**: Updates `SizeOfCode`, `SizeOfImage`, and `SizeOfHeaders`
//! - **Section Table**: Patches section headers with final sizes, RVAs, and offsets
//! - **COR20 Header**: Updates metadata RVA and size
//! - **Data Directories**: Patches directory entries for IAT, CLR header, imports/exports, etc.
//! - **Stripped Regions**: Zeros debug and certificate data for better compression
//! - **PE Checksum**: Calculated from all file bytes after all other modifications
//!
//! # Architecture
//!
//! All fixup functions operate on [`WriteContext`] which carries the positions and values
//! collected during the write phase. The functions use `write_u32_at()` to patch specific
//! offsets without changing the write position.
//!
//! ```text
//! ┌─────────────────┐
//! │  Write Phase    │ ── Collects positions, sizes, RVAs
//! └────────┬────────┘
//!          │
//!          ▼
//! ┌─────────────────┐
//! │  Fixup Phase    │ ── Patches headers with final values
//! │  (this module)  │
//! └────────┬────────┘
//!          │
//!          ▼
//! ┌─────────────────┐
//! │  Checksum       │ ── Final step: calculate PE checksum
//! └─────────────────┘
//! ```
//!
//! # PE Checksum Algorithm
//!
//! The PE checksum is a 32-bit sum computed over all 16-bit words in the file, with the
//! checksum field itself excluded. The algorithm handles carry-out by adding it back
//! (similar to IP checksum). After summing all words, the file size is added.
//!
//! The checksum is stored in the Optional Header at offset 64 from the start of the
//! Optional Header, regardless of PE32 or PE32+ format.
//!
//! # References
//!
//! - [PE Format: Optional Header](https://docs.microsoft.com/en-us/windows/win32/debug/pe-format#optional-header-image-only)
//! - [PE Format: Data Directories](https://docs.microsoft.com/en-us/windows/win32/debug/pe-format#optional-header-data-directories-image-only)
//! - ECMA-335 §II.25 - File format extensions to PE

use crate::{
    cilassembly::writer::{context::WriteContext, output::Output},
    file::pe::{constants::COR20_HEADER_SIZE, SectionTable},
    utils::align_to,
    Result,
};

/// Standard metadata stream names in the order they are written.
const METADATA_STREAM_NAMES: [&str; 5] = ["#~", "#Strings", "#US", "#GUID", "#Blob"];

/// Applies all header fixups to the output file.
///
/// This is the main entry point for the fixup phase. It calls all individual fixup
/// functions in the correct order, with checksum calculation always last.
///
/// # Arguments
///
/// * `ctx` - The write context containing all positions and values from the write phase
///
/// # Returns
///
/// Returns `Ok(())` on success, or an error if any fixup operation fails.
///
/// # Order of Operations
///
/// 1. DOS header e_lfanew fixup
/// 2. Optional header fixups (sizes)
/// 3. Section table fixups (RVAs, sizes, offsets)
/// 4. COR20 header fixup (metadata location)
/// 5. Data directory fixups
/// 6. Zero stripped data regions (debug, certificate)
/// 7. PE checksum calculation (must be last)
pub fn apply_all_fixups(ctx: &mut WriteContext) -> Result<()> {
    // Fix e_lfanew in DOS header (offset 0x3C points to PE signature)
    let pe_sig_offset = u32::try_from(ctx.pe_signature_offset).map_err(|_| {
        crate::Error::LayoutFailed("PE signature offset exceeds u32 range".to_string())
    })?;
    ctx.write_u32_at(ctx.dos_header_offset + 0x3C, pe_sig_offset)?;

    // Fix header fields
    fixup_optional_header(ctx)?;
    fixup_section_table(ctx)?;
    fixup_cor20_header(ctx)?;
    fixup_data_directories(ctx)?;

    // Zero stripped data regions for better compression
    zero_stripped_data_regions(ctx)?;

    // Fix COFF characteristics if relocations were stripped
    fixup_coff_characteristics(ctx)?;

    // Calculate and write PE checksum (must be last)
    fixup_checksum(ctx)?;

    Ok(())
}

/// Fixes up the Optional Header with final size values.
///
/// Updates the following fields:
/// - `SizeOfCode` (offset +4): Size of .text section aligned to file alignment
/// - `SizeOfImage` (offset +56): Total image size in memory, aligned to section alignment
/// - `SizeOfHeaders` (offset +60): Combined size of all headers up to first section
///
/// # Arguments
///
/// * `ctx` - The write context with section information
///
/// # PE Format Reference
///
/// The offsets are the same for both PE32 and PE32+ formats for these fields:
/// - Standard fields magic is 2 bytes, so SizeOfCode is at offset 4
/// - SizeOfImage and SizeOfHeaders are in the Windows-specific fields
pub fn fixup_optional_header(ctx: &mut WriteContext) -> Result<()> {
    let text_file_size = u32::try_from(align_to(
        ctx.text_section_size,
        u64::from(ctx.file_alignment),
    ))
    .map_err(|_| crate::Error::LayoutFailed("Text file size exceeds u32 range".to_string()))?;

    // Calculate total image size from all active sections
    let mut end_rva: u32 = 0;
    for section in &ctx.sections {
        if section.removed {
            continue;
        }
        if let (Some(rva), Some(size)) = (section.rva, section.data_size) {
            let section_end = rva.saturating_add(size);
            if section_end > end_rva {
                end_rva = section_end;
            }
        }
    }

    let image_size = u32::try_from(align_to(
        u64::from(end_rva),
        u64::from(ctx.section_alignment),
    ))
    .map_err(|_| crate::Error::LayoutFailed("Image size exceeds u32 range".to_string()))?;

    // SizeOfCode at offset 4 (after magic field)
    ctx.write_u32_at(ctx.optional_header_offset + 4, text_file_size)?;

    // AddressOfEntryPoint at offset 16
    // This is the RVA of the native entry point stub that jumps to _CorExeMain/_CorDllMain
    if let Some(entry_rva) = ctx.native_entry_rva {
        ctx.write_u32_at(ctx.optional_header_offset + 16, entry_rva)?;
    }

    // SizeOfImage at offset 56
    ctx.write_u32_at(ctx.optional_header_offset + 56, image_size)?;

    // SizeOfHeaders at offset 60
    let headers_size = u32::try_from(ctx.text_section_offset)
        .map_err(|_| crate::Error::LayoutFailed("Headers size exceeds u32 range".to_string()))?;
    ctx.write_u32_at(ctx.optional_header_offset + 60, headers_size)?;

    Ok(())
}

/// Fixes up the Section Table with final values.
///
/// Rebuilds the section table by:
/// 1. Collecting all non-removed sections with updated values
/// 2. Writing them contiguously (no gaps from removed sections)
/// 3. Updating section count in COFF header
///
/// Each section header is updated with:
/// - `VirtualSize` (offset +8): Actual size of section content
/// - `VirtualAddress` (offset +12): RVA where section is loaded
/// - `SizeOfRawData` (offset +16): Size on disk, aligned to file alignment
/// - `PointerToRawData` (offset +20): File offset of section data
pub fn fixup_section_table(ctx: &mut WriteContext) -> Result<()> {
    // Build the new section table, excluding removed sections
    let mut section_headers: Vec<[u8; SectionTable::SIZE]> = Vec::new();

    for section in &ctx.sections {
        if section.removed {
            continue;
        }
        // Get section data or use original values for sections without write info
        let (Some(data_offset), Some(rva), Some(data_size)) =
            (section.data_offset, section.rva, section.data_size)
        else {
            continue; // Skip sections without data
        };

        // SizeOfRawData must be a multiple of FileAlignment per PE spec.
        // This is required for all sections including the last one.
        let file_size = u32::try_from(align_to(
            u64::from(data_size),
            u64::from(ctx.file_alignment),
        ))
        .map_err(|_| {
            crate::Error::LayoutFailed(format!(
                "Section {} file size exceeds u32 range",
                section.name
            ))
        })?;

        let offset_u32 = u32::try_from(data_offset).map_err(|_| {
            crate::Error::LayoutFailed(format!("Section {} offset exceeds u32 range", section.name))
        })?;

        // Build section header
        let mut header = [0u8; SectionTable::SIZE];

        // Name (8 bytes, null-padded)
        let name_bytes = section.name.as_bytes();
        let copy_len = std::cmp::min(name_bytes.len(), 8);
        header[..copy_len].copy_from_slice(&name_bytes[..copy_len]);

        // VirtualSize
        header[8..12].copy_from_slice(&data_size.to_le_bytes());
        // VirtualAddress
        header[12..16].copy_from_slice(&rva.to_le_bytes());
        // SizeOfRawData
        header[16..20].copy_from_slice(&file_size.to_le_bytes());
        // PointerToRawData
        header[20..24].copy_from_slice(&offset_u32.to_le_bytes());
        // COFF relocation and line number fields are always 0 for PE executables.
        // These are legacy fields from COFF object files used during linking.
        // PointerToRelocations
        header[24..28].copy_from_slice(&0u32.to_le_bytes());
        // PointerToLinenumbers (deprecated)
        header[28..32].copy_from_slice(&0u32.to_le_bytes());
        // NumberOfRelocations
        header[32..34].copy_from_slice(&0u16.to_le_bytes());
        // NumberOfLinenumbers (deprecated)
        header[34..36].copy_from_slice(&0u16.to_le_bytes());
        // Characteristics
        header[36..40].copy_from_slice(&section.characteristics.to_le_bytes());

        section_headers.push(header);
    }

    // Write the rebuilt section table
    let mut offset = ctx.section_table_offset;
    for header in &section_headers {
        ctx.write_at(offset, header)?;
        offset += SectionTable::SIZE as u64;
    }

    // Zero any remaining space (from removed sections)
    let original_table_size = ctx.sections.len() * SectionTable::SIZE;
    let new_table_size = section_headers.len() * SectionTable::SIZE;
    if new_table_size < original_table_size {
        let zeros = vec![0u8; original_table_size - new_table_size];
        ctx.write_at(offset, &zeros)?;
    }

    // Update section count in COFF header
    let new_count = u16::try_from(section_headers.len()).unwrap_or(0);
    ctx.write_u16_at(ctx.coff_header_offset + 2, new_count)?;

    Ok(())
}

/// Fixes up the COR20 (CLR) header with metadata location.
///
/// Updates:
/// - `MetaData.VirtualAddress` (offset +8): RVA of metadata root
/// - `MetaData.Size` (offset +12): Size of entire metadata section
///
/// # Arguments
///
/// * `ctx` - The write context with metadata position information
///
/// # CLR Header Layout
///
/// The COR20 header (IMAGE_COR20_HEADER) has a fixed structure defined in ECMA-335.
/// The MetaData field is a data directory (RVA + Size) at offset 8.
pub fn fixup_cor20_header(ctx: &mut WriteContext) -> Result<()> {
    // MetaData directory (offset 8-15)
    let metadata_rva = ctx.offset_to_rva(ctx.metadata_offset);
    let metadata_size = u32::try_from(ctx.metadata_size)
        .map_err(|_| crate::Error::LayoutFailed("Metadata size exceeds u32 range".to_string()))?;

    ctx.write_u32_at(ctx.cor20_header_offset + 8, metadata_rva)?; // MetaData RVA
    ctx.write_u32_at(ctx.cor20_header_offset + 12, metadata_size)?; // MetaData Size

    // Resources directory (offset 24-31)
    // COR20 header layout:
    //   0-3: cb (header size)
    //   4-5: MajorRuntimeVersion
    //   6-7: MinorRuntimeVersion
    //   8-11: MetaData RVA
    //   12-15: MetaData Size
    //   16-19: Flags
    //   20-23: EntryPointToken
    //   24-27: Resources RVA
    //   28-31: Resources Size

    // Entry point token (offset 20-23) - may need remapping if methods were deleted
    if ctx.entry_point_token != 0 && !ctx.token_remapping.is_empty() {
        if let Some(&new_token) = ctx.token_remapping.get(&ctx.entry_point_token) {
            ctx.write_u32_at(ctx.cor20_header_offset + 20, new_token)?;
        }
    }

    if ctx.resource_data_size > 0 {
        let resource_rva = ctx.offset_to_rva(ctx.resource_data_offset);
        let resource_size = u32::try_from(ctx.resource_data_size).map_err(|_| {
            crate::Error::LayoutFailed("Resource size exceeds u32 range".to_string())
        })?;

        ctx.write_u32_at(ctx.cor20_header_offset + 24, resource_rva)?; // Resources RVA
        ctx.write_u32_at(ctx.cor20_header_offset + 28, resource_size)?; // Resources Size
    }

    Ok(())
}

/// Fixes up the Data Directories in the Optional Header.
///
/// Data directories are an array of (RVA, Size) pairs that point to various PE
/// structures. This function updates directories for:
///
/// - **Index 0**: Export Table (if native exports present)
/// - **Index 1**: Import Table (if native imports present)
/// - **Index 2**: Resource Table (if .rsrc section present)
/// - **Index 5**: Base Relocation Table (if .reloc section present)
/// - **Index 12**: IAT (Import Address Table) - always at start of .text
/// - **Index 14**: CLR Runtime Header - always 8 bytes after .text start
///
/// # Arguments
///
/// * `ctx` - The write context with data directory information
///
/// # Directory Offsets
///
/// Data directories start at different offsets in the Optional Header:
/// - PE32: offset 96 (28 standard + 68 windows-specific fields)
/// - PE32+: offset 112 (24 standard + 88 windows-specific fields)
///
/// Each directory entry is 8 bytes (4-byte RVA + 4-byte Size).
///
/// # Note
///
/// Debug (index 6) and Certificate (index 4) directories are zeroed during the
/// write phase because they become invalid after assembly modification.
pub fn fixup_data_directories(ctx: &mut WriteContext) -> Result<()> {
    let dd_offset = if ctx.is_pe32_plus { 112 } else { 96 };
    let dd_base = ctx.optional_header_offset + dd_offset;

    // IAT (index 12) and CLR Runtime Header (index 14)
    // When the assembly has native imports (IAT was written), the layout is:
    //   .text start → IAT → COR20 header → ...
    // When no native imports exist (.NET Core PE32+ without mscoree.dll):
    //   .text start → COR20 header → ...
    let has_iat = ctx.iat_size > 0;
    if has_iat {
        let iat_rva = ctx.text_section_rva;
        let iat_size = u32::try_from(ctx.iat_size).unwrap_or(8);
        ctx.write_u32_at(dd_base + 12 * 8, iat_rva)?;
        ctx.write_u32_at(dd_base + 12 * 8 + 4, iat_size)?;

        // CLR header sits immediately after IAT
        let clr_rva = ctx.text_section_rva + iat_size;
        ctx.write_u32_at(dd_base + 14 * 8, clr_rva)?;
        ctx.write_u32_at(dd_base + 14 * 8 + 4, COR20_HEADER_SIZE)?;
    } else {
        // No IAT - zero the IAT data directory
        ctx.write_u32_at(dd_base + 12 * 8, 0)?;
        ctx.write_u32_at(dd_base + 12 * 8 + 4, 0)?;

        // CLR header sits at the very start of .text section
        let clr_rva = ctx.text_section_rva;
        ctx.write_u32_at(dd_base + 14 * 8, clr_rva)?;
        ctx.write_u32_at(dd_base + 14 * 8 + 4, COR20_HEADER_SIZE)?;
    }

    // Import Table (index 1)
    if let (Some(rva), Some(size)) = (ctx.import_data_rva, ctx.import_data_size) {
        ctx.write_u32_at(dd_base + 8, rva)?;
        ctx.write_u32_at(dd_base + 8 + 4, size)?;
    } else {
        // No import table - zero the directory entry
        ctx.write_u32_at(dd_base + 8, 0)?;
        ctx.write_u32_at(dd_base + 8 + 4, 0)?;
    }

    // Export Table (index 0)
    if let (Some(rva), Some(size)) = (ctx.export_data_rva, ctx.export_data_size) {
        ctx.write_u32_at(dd_base, rva)?;
        ctx.write_u32_at(dd_base + 4, size)?;
    }

    // Resource Table (index 2) - find .rsrc section or embedded PE resources
    let rsrc_section = ctx
        .sections
        .iter()
        .find(|s| s.name.starts_with(".rsrc") && !s.removed);
    if let Some(section) = rsrc_section {
        if let (Some(rva), Some(size)) = (section.rva, section.data_size) {
            ctx.write_u32_at(dd_base + 2 * 8, rva)?;
            ctx.write_u32_at(dd_base + 2 * 8 + 4, size)?;
        }
    } else if ctx.pe_resource_size > 0 {
        // Resources were embedded in .text and carried over
        let rva = ctx.offset_to_rva(ctx.pe_resource_offset);
        ctx.write_u32_at(dd_base + 2 * 8, rva)?;
        ctx.write_u32_at(dd_base + 2 * 8 + 4, ctx.pe_resource_size)?;
    } else {
        // No resources at all - zero the directory entry
        ctx.write_u32_at(dd_base + 2 * 8, 0)?;
        ctx.write_u32_at(dd_base + 2 * 8 + 4, 0)?;
    }

    // Base Relocation Table (index 5) - find .reloc section in sections vector
    let reloc_section = ctx
        .sections
        .iter()
        .find(|s| s.name.starts_with(".reloc") && !s.removed);
    if let Some(section) = reloc_section {
        if let (Some(rva), Some(size)) = (section.rva, section.data_size) {
            ctx.write_u32_at(dd_base + 5 * 8, rva)?;
            ctx.write_u32_at(dd_base + 5 * 8 + 4, size)?;
        } else {
            // Section exists but no data written - zero the directory
            ctx.write_u32_at(dd_base + 5 * 8, 0)?;
            ctx.write_u32_at(dd_base + 5 * 8 + 4, 0)?;
        }
    } else {
        // No reloc section or it was removed - zero out the data directory entry
        ctx.write_u32_at(dd_base + 5 * 8, 0)?;
        ctx.write_u32_at(dd_base + 5 * 8 + 4, 0)?;
    }

    Ok(())
}

/// Fixes up the metadata root stream headers with actual offsets and sizes.
///
/// The metadata root header contains an array of stream headers, each with:
/// - `offset` (u32): Offset from metadata root to stream data
/// - `size` (u32): Size of stream data (aligned to 4 bytes)
/// - `name`: Null-terminated string, padded to 4-byte boundary
///
/// During initial write, these headers contain placeholder values. This function
/// patches them with the actual offsets and sizes after all streams are written.
///
/// # Arguments
///
/// * `ctx` - The write context with stream positions
/// * `metadata_root_offset` - File offset of the metadata root header
/// * `stream_headers_offset` - File offset where stream headers begin
///
/// # Returns
///
/// Returns `Ok(())` on successful fixup of all stream headers.
///
/// # Errors
///
/// Returns an error if writing to the output file fails.
///
/// # Stream Order
///
/// Streams are written and fixed up in this order:
/// 1. `#~` - Tables stream
/// 2. `#Strings` - String heap
/// 3. `#US` - User string heap
/// 4. `#GUID` - GUID heap
/// 5. `#Blob` - Blob heap
pub fn fixup_metadata_stream_headers(
    ctx: &mut WriteContext,
    metadata_root_offset: u64,
    stream_headers_offset: u64,
) -> Result<()> {
    let mut offset = stream_headers_offset;

    // Stream order: #~, #Strings, #US, #GUID, #Blob
    let streams = [
        (
            ctx.tables_stream_offset,
            ctx.tables_stream_size,
            METADATA_STREAM_NAMES[0],
        ),
        (
            ctx.strings_heap_offset,
            ctx.strings_heap_size,
            METADATA_STREAM_NAMES[1],
        ),
        (
            ctx.us_heap_offset,
            ctx.us_heap_size,
            METADATA_STREAM_NAMES[2],
        ),
        (
            ctx.guid_heap_offset,
            ctx.guid_heap_size,
            METADATA_STREAM_NAMES[3],
        ),
        (
            ctx.blob_heap_offset,
            ctx.blob_heap_size,
            METADATA_STREAM_NAMES[4],
        ),
    ];

    for (stream_offset, stream_size, name) in &streams {
        // Calculate offset relative to metadata root
        let relative_offset =
            u32::try_from(*stream_offset - metadata_root_offset).map_err(|_| {
                crate::Error::LayoutFailed("Stream relative offset exceeds u32 range".to_string())
            })?;
        let aligned_size = u32::try_from(align_to(*stream_size, 4)).map_err(|_| {
            crate::Error::LayoutFailed("Stream aligned size exceeds u32 range".to_string())
        })?;

        // Write offset
        ctx.write_u32_at(offset, relative_offset)?;
        // Write size
        ctx.write_u32_at(offset + 4, aligned_size)?;

        // Advance past this stream header (offset + size + name with alignment)
        let name_with_null = name.len() + 1;
        let aligned_name = align_to(name_with_null as u64, 4);
        offset += 8 + aligned_name;
    }

    Ok(())
}

/// Handles stripped data regions (debug, certificates) after assembly modification.
///
/// When modifying a .NET assembly, certain data becomes invalid:
/// - **Debug data**: IL offsets change, PDB correlation breaks
/// - **Certificate data**: Digital signatures are invalidated
///
/// # Debug Data
///
/// Debug data is NOT explicitly zeroed because:
/// 1. The `.text` section is completely rebuilt with a new layout, so original
///    debug data (which was in the old `.text`) simply isn't included
/// 2. The debug directory entry is already zeroed, so loaders won't look for it
/// 3. Using the original debug RVA would corrupt our new data since the layout changed
///
/// # Certificate Data
///
/// Certificates use a file offset (not RVA) and are typically appended after all
/// sections. They are usually excluded by file truncation to `bytes_written`.
/// If certificate data falls within written bounds, it's zeroed.
///
/// # Arguments
///
/// * `ctx` - The write context with original directory locations
pub fn zero_stripped_data_regions(ctx: &mut WriteContext) -> Result<()> {
    // Debug data handling:
    //
    // We do NOT attempt to zero the original debug data region because:
    //
    // 1. The .text section is completely rebuilt with a new layout. The original
    //    debug RVA pointed to data in the OLD .text layout, which no longer exists.
    //    Using that RVA to calculate a file offset in our new file would corrupt
    //    unrelated data (e.g., blob heap, method bodies).
    //
    // 2. The debug directory entry in the data directories is already zeroed
    //    during generation (see write_data_directories). This tells loaders
    //    there is no debug info, so they won't look for any debug data.
    //
    // 3. Since we rebuild .text from scratch (metadata, method bodies, etc.),
    //    the original debug data simply isn't included - we never copy it.
    //
    // For assemblies where debug data is in a separate section (e.g., .rdata),
    // we don't copy those sections either - we only preserve .rsrc and .reloc.
    let _ = ctx.original_debug_dir; // Stored for reference but not used

    // Certificate data handling:
    //
    // Certificates use a FILE OFFSET (not RVA) in the data directory, and are
    // typically appended after all sections. Since we truncate the output to
    // `bytes_written`, certificate data that was beyond our content is naturally
    // excluded. If somehow certificate data falls within our written bounds
    // (unusual but possible), we zero it since the signature is invalid after
    // any modification.
    if let Some((cert_offset, cert_size)) = ctx.original_certificate_dir {
        let cert_offset_u64 = u64::from(cert_offset);
        if cert_offset_u64 + u64::from(cert_size) <= ctx.bytes_written {
            let zeros = vec![0u8; cert_size as usize];
            ctx.write_at(cert_offset_u64, &zeros)?;
        }
    }

    Ok(())
}

/// Fixes up COFF characteristics when relocations are stripped.
///
/// Sets the `IMAGE_FILE_RELOCS_STRIPPED` flag (0x0001) in the COFF header
/// characteristics field when the assembly has no relocations. This is
/// required by CoreCLR for IL-only x64 EXEs without relocations.
///
/// # Arguments
///
/// * `ctx` - The write context with relocs_stripped flag
///
/// # CoreCLR Requirements
///
/// From CoreCLR's `CheckILOnlyBaseRelocations()`:
/// - If no base relocation directory AND not a DLL, RELOCS_STRIPPED must be set
/// - This applies to IL-only x64 EXEs where no relocations are needed
///
/// # COFF Header Layout
///
/// The characteristics field is at offset 18 (0x12) from the start of the
/// COFF header. It's a 16-bit field containing various PE flags.
pub fn fixup_coff_characteristics(ctx: &mut WriteContext) -> Result<()> {
    if ctx.relocs_stripped {
        const IMAGE_FILE_RELOCS_STRIPPED: u16 = 0x0001;

        // Read current characteristics (offset +18 from COFF header)
        let chars_offset = ctx.coff_header_offset + 18;

        // Read the current value from the output
        let current_bytes = ctx.output.as_slice();
        #[allow(clippy::cast_possible_truncation)]
        let chars_offset_usize = chars_offset as usize;
        if chars_offset_usize + 2 <= current_bytes.len() {
            let current = u16::from_le_bytes([
                current_bytes[chars_offset_usize],
                current_bytes[chars_offset_usize + 1],
            ]);

            // Set the RELOCS_STRIPPED flag
            ctx.write_u16_at(chars_offset, current | IMAGE_FILE_RELOCS_STRIPPED)?;
        }
    }

    Ok(())
}

/// Calculates and writes the PE checksum.
///
/// This must be the last fixup operation since the checksum covers all file content.
/// The checksum field itself is excluded from the calculation.
///
/// # Arguments
///
/// * `ctx` - The write context with output file reference
///
/// # Algorithm
///
/// 1. Sum all 16-bit words in the file using 32-bit arithmetic
/// 2. Skip the checksum field (4 bytes at optional_header_offset + 64)
/// 3. Fold carry bits back into the lower 16 bits
/// 4. Add file size to get final checksum
///
/// # Checksum Location
///
/// The checksum is at offset 64 from the start of the Optional Header,
/// which is the same for both PE32 and PE32+ formats.
///
/// # Note
///
/// Uses `ctx.bytes_written` as the actual file size, not the over-allocated
/// mmap size. This ensures the checksum matches what the final truncated
/// file will contain.
pub fn fixup_checksum(ctx: &mut WriteContext) -> Result<()> {
    let checksum_offset = ctx.optional_header_offset + 64;
    let actual_size = usize::try_from(ctx.bytes_written)
        .map_err(|_| crate::Error::LayoutFailed("File size exceeds usize range".to_string()))?;
    let checksum = calculate_pe_checksum(&ctx.output, checksum_offset, actual_size);
    ctx.write_u32_at(checksum_offset, checksum)?;

    Ok(())
}

/// Calculates the PE checksum for the file.
///
/// This implements the standard Windows PE checksum algorithm used by the
/// `MapFileAndCheckSum` API and the Windows loader.
///
/// # Algorithm Details
///
/// 1. Sum all 16-bit words in the file using 32-bit arithmetic
/// 2. Skip the checksum field itself (4 bytes = 2 words at checksum_offset)
/// 3. Fold any carry bits back into the lower 16 bits
/// 4. Add the file length to get the final checksum
///
/// # Arguments
///
/// * `output` - The memory-mapped output file to calculate checksum for
/// * `checksum_offset` - File offset of the checksum field (to exclude from sum)
/// * `actual_size` - The actual file size (may be less than the mmap size due to over-allocation)
///
/// # Returns
///
/// The calculated PE checksum value.
///
/// # Performance
///
/// Since the output is memory-mapped, we access the data directly without
/// any buffer allocation or copying. This is significantly faster than
/// chunked I/O for large files.
fn calculate_pe_checksum(output: &Output, checksum_offset: u64, actual_size: usize) -> u32 {
    let data = output.as_slice();
    let file_size = actual_size.min(data.len()); // Don't exceed mmap bounds
                                                 // Safe: checksum_offset is a small PE header offset that always fits in usize
    let checksum_offset_usize = usize::try_from(checksum_offset).unwrap_or(usize::MAX);

    let mut sum: u64 = 0;

    // Process 16-bit words directly from the memory-mapped file
    let mut i = 0;
    while i + 1 < file_size {
        // Skip the checksum field (4 bytes = 2 words)
        if i >= checksum_offset_usize && i < checksum_offset_usize + 4 {
            i += 2;
            continue;
        }

        let word = u16::from_le_bytes([data[i], data[i + 1]]);
        sum += u64::from(word);
        i += 2;
    }

    // Handle odd byte at the end of file (if any) - pad with zero
    if i < file_size {
        // Only include if not in checksum field
        if i < checksum_offset_usize || i >= checksum_offset_usize + 4 {
            sum += u64::from(data[i]);
        }
    }

    // Fold the sum to 16 bits (add carry to low 16 bits)
    while sum > 0xFFFF {
        sum = (sum & 0xFFFF) + (sum >> 16);
    }

    // Add file size - safe: sum is folded to fit in u16, and file_size fits in u32 on all platforms
    #[allow(clippy::cast_possible_truncation)]
    let checksum = (sum as u32) + (file_size as u32);

    checksum
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::cilassembly::writer::generator::PeGenerator;
    use crate::cilassembly::CilAssembly;
    use crate::CilAssemblyView;
    use std::path::Path;
    use tempfile::NamedTempFile;

    #[test]
    fn test_checksum_excludes_checksum_field() {
        // Create a small test output
        let temp_file = NamedTempFile::new().expect("Failed to create temp file");
        let mut output = Output::create(temp_file.path(), 256).expect("Failed to create output");

        // Fill with known pattern
        let pattern: Vec<u8> = (0..256).map(|i| i as u8).collect();
        output.write_at(0, &pattern).expect("Failed to write data");

        // Calculate checksum with exclusion at offset 64, using full 256 bytes as actual size
        let checksum1 = calculate_pe_checksum(&output, 64, 256);

        // Modify bytes at checksum offset
        output
            .write_at(64, &[0xFF, 0xFF, 0xFF, 0xFF])
            .expect("Failed to modify checksum area");

        // Recalculate - should be the same since checksum area is excluded
        let checksum2 = calculate_pe_checksum(&output, 64, 256);

        assert_eq!(
            checksum1, checksum2,
            "Checksum should be the same regardless of checksum field content"
        );
    }

    #[test]
    fn test_apply_fixups_integration() {
        // Load a test assembly
        let view = CilAssemblyView::from_path(Path::new("tests/samples/crafted_2.exe"))
            .expect("Failed to load test assembly");
        let assembly: CilAssembly = view.to_owned();

        // Generate to a temp file
        let temp_file = NamedTempFile::new().expect("Failed to create temp file");
        let generator = PeGenerator::new(&assembly);
        generator
            .to_file(temp_file.path())
            .expect("PE generation should succeed");

        // Reload and verify
        let reloaded = CilAssemblyView::from_path(temp_file.path())
            .expect("Should be able to reload generated PE");

        // Verify basic structure is intact
        assert!(
            reloaded.tables().is_some(),
            "Reloaded PE should have tables"
        );
    }
}