Musa-Cpp-Lib-V2/lib/OS/OS_Filesystem.cpp

// So NTFS (and most systems) sort the tree by default in lexicographical descending order.
// For lookups, this often isn't that useful if you only know substrings of the path.


// struct BTNode {
//   u16 key_count;
//   BTNode* keys;
//   u16 allocated;
// };

// struct B_Tree {
//   BTNode* root;
//   Allocator allocator;
// };

// A compact collection of data with sorting indices
// Maybe we can make B+ trees for sorting according to 
// size and modtime. 

// It really doesn't make sense to store data in memory as a B-tree except 
// if we need ordered insertions and deletes.
// 

// Returns offset
force_inline u32 AddString_NoCount (Serializer* serializer, u8* data, u8 count) { // #TODO: , bool null_terminate=false
  u32 original_count = (u32)serializer->count;
  u8* current_point = &serializer->data[original_count];
  
  s64 final_count = serializer->allocated + (count * sizeof(u8));
  
  if (serializer->allocated < final_count) {
    array_reserve(*serializer, final_count);
  }
  
  memcpy(current_point, data, count * sizeof(u8));
  serializer->count += count * sizeof(u8);
  
  return original_count;
}

constexpr s64 DFS_Preallocation_Count = 4194304; // 2^22

// template <typename Length_Type> 
struct DFS_Array {
  Serializer* wstrings; // is a ArenaArray<u8>
  Serializer* strings;
  
  ArenaArray<u32>* offsets; // offsets into strings->data
  ArenaArray<u8>*  lengths; // this type may vary <hmmm> Not sure if I should make it a template argument. Seems yucky.
  ArenaArray<u64>* modtimes;
  ArenaArray<u64>* sizes;
  
  ArenaArray<s32>* parent_indices;
  
  // #Temporary arrays for linking files/dirs to their parent directory, if present.
  ArenaArray<u32>* record_ids;
  ArenaArray<u32>* parent_ids;
};

struct DFS_Value {
  wstring wpath;
  u64 modtime;
  s64 size;
  s32 parent_index;
  string full_path;
};

struct Dense_FS { // Link to OS_Drive
  DFS_Array paths;
  DFS_Array files;
  
  ArenaTable<u32, s32> path_table; // <entry_id, array_offset>.
  
  OS_Drive* drive; // backlink for reference. 
};

// Need the drive letter too!
string get_full_path (string drive_label, Dense_FS* dfs, s32 first_parent_index, string file_name) {
  s32 parent_index = first_parent_index;

  DFS_Array* paths = &dfs->paths;

  auto_release_temp();
  Array<string> path_list_reverse;
  path_list_reverse.allocator = temp();
  
  array_add(path_list_reverse, file_name);
  
  while (parent_index > 0) { // -1 means we reached the drive letter (root)
    u32 path_offset = (*paths->offsets)[parent_index];
    u8* path_data_ptr = (u8*)((paths->strings->data) + path_offset);
    string next_dir = {(s64)(*paths->lengths)[parent_index], path_data_ptr};
    array_add(path_list_reverse, next_dir);
    
    parent_index = (*paths->parent_indices)[parent_index];
  }
  
  // Drive letter with colon:
  Assert(drive_label.count >= 2);
  string drive_letter_with_colon = {2, drive_label.data};
  
  array_add(path_list_reverse, drive_letter_with_colon);
  
  // reset_string_builder(sb, true);
  String_Builder* sb = new_string_builder();
  
  for (s64 i = path_list_reverse.count-1; i >= 0; i -= 1) {
    append(sb, path_list_reverse[i]);
    if (i != 0) { append(sb, "/"); }
  }
  
  // return copy_string(string_view(sb));
  return builder_to_string(sb);
}

string get_full_path_from_index (OS_Drive* drive, DFS_Array* dfsa, s64 i) {
  u32 path_offset = (*dfsa->offsets)[i];
  u8* path_data_ptr = (u8*)((dfsa->strings->data) + path_offset);
  string file_name = {(*dfsa->lengths)[i], path_data_ptr};
  return get_full_path(drive->label, drive->data, (*dfsa->parent_indices)[i], file_name);
}

DFS_Value get_value (Dense_FS* dfs, DFS_Array* dfsa, s64 i) {
  u32 path_offset = (*dfsa->offsets)[i];
  u8* path_data_ptr = (u8*)((dfsa->strings->data) + path_offset);
  // u16* path_wide_ptr = (u16*)((dfsa->wstrings->data) + path_offset);
  // wstring path_wide = {(*dfsa->lengths)[i], path_wide_ptr};
  DFS_Value dfsv;
  dfsv.modtime       = (*dfsa->modtimes)[i];
  dfsv.size          = (s64)(*dfsa->sizes)[i];
  dfsv.parent_index  = (*dfsa->parent_indices)[i];
  // #TODO: I cna just use strings.
  // string path_utf8 = wide_to_utf8(path_wide.data, (s32)path_wide.count);
  // dfsv.full_path     = get_full_path_from_offset(dfs, dfsv.parent_index, path_utf8);
  dfsv.full_path = {(*dfsa->lengths)[i], path_data_ptr};
  return dfsv;
}

s64 item_count (DFS_Array* dfsa) {
  return dfsa->offsets->count;
}

void initialize (DFS_Array* dfsa) {
  Assert(dfsa != nullptr);
  dfsa->wstrings = new_serializer(Arena_Reserve::Size_2G);
  dfsa->strings = new_serializer(Arena_Reserve::Size_2G);
  
  dfsa->offsets  = arena_array_new<u32>(DFS_Preallocation_Count, Arena_Reserve::Size_2G);
  dfsa->lengths  = arena_array_new<u8> (DFS_Preallocation_Count, Arena_Reserve::Size_2G);
  dfsa->modtimes = arena_array_new<u64>(DFS_Preallocation_Count, Arena_Reserve::Size_2G);
  dfsa->sizes    = arena_array_new<u64>(DFS_Preallocation_Count, Arena_Reserve::Size_2G);
  
  dfsa->record_ids = arena_array_new<u32>(DFS_Preallocation_Count, Arena_Reserve::Size_2G);
  dfsa->parent_ids = arena_array_new<u32>(DFS_Preallocation_Count, Arena_Reserve::Size_2G);
  
  dfsa->parent_indices = arena_array_new<s32>(DFS_Preallocation_Count, Arena_Reserve::Size_2G);
}

void release (DFS_Array* dfsa) {
  free_serializer(dfsa->wstrings);
  free_serializer(dfsa->strings);

  if (is_valid(dfsa->offsets))  { arena_array_free(*dfsa->offsets);  }
  if (is_valid(dfsa->lengths))  { arena_array_free(*dfsa->lengths);  }
  if (is_valid(dfsa->modtimes)) { arena_array_free(*dfsa->modtimes); }
  if (is_valid(dfsa->sizes))    { arena_array_free(*dfsa->sizes);    }

  if (is_valid(dfsa->record_ids)) { arena_array_free(*dfsa->record_ids); }
  if (is_valid(dfsa->parent_ids)) { arena_array_free(*dfsa->parent_ids); }

  if (is_valid(dfsa->parent_indices)) { arena_array_free(*dfsa->parent_indices); }
  
  zero_struct(dfsa);
}

force_inline void release (Dense_FS* dfs) {
  Timed_Block_Print("release: Dense_FS*");
  dfs->drive = nullptr; // just a link  
  
  release(&dfs->paths);
  release(&dfs->files);
  
  if (table_is_valid(&dfs->path_table)) {
    table_release(&dfs->path_table);
  }
}

internal void Dense_FS_initialize (Dense_FS* dfs) {
  initialize(&dfs->paths);
  initialize(&dfs->files);
  
  table_init(&dfs->path_table, 1048576); // 2^20
  // dfs->path_table.hash_function = table_hash_function_fnv1a; // default.
  dfs->path_table.hash_function = sdbm_hash;
  dfs->path_table.compare_function = u32_keys_match;
}

void initialize (Dense_FS* dfs, OS_Drive* drive) {
  Assert(drive != nullptr); Assert(dfs != nullptr);
  // Is there a less stupid way of doing this?
  dfs->drive = drive;
  drive->data = dfs;
  
  Dense_FS_initialize(dfs);
}

s32 find_previous_index (Dense_FS* dfs, u32 record_id, bool* success) {
  s32 result = -1;

  (*success) = table_find(&dfs->path_table, record_id, &result);

  return result;
}

void cleanup_after_enumeration(Dense_FS* dfs) {
  table_release(&dfs->path_table);
  reset_struct(&dfs->path_table);
  
  arena_array_free(*dfs->paths.record_ids);
  arena_array_free(*dfs->paths.parent_ids);
  arena_array_free(*dfs->files.record_ids);
  arena_array_free(*dfs->files.parent_ids);
  
  // Unfortunately, we have to do this whenever we free, because we're using raw pointers.
  dfs->paths.record_ids = nullptr;
  dfs->paths.parent_ids = nullptr;
  dfs->files.record_ids = nullptr;
  dfs->files.parent_ids = nullptr;
}