Exemple #1
0
def extract_model_chunk():
    bin_file = open(SETTINGS["gif_path"] + "/modelchunk.bin", "rb")

    data = get_formatted_data(bin_file, "wdb", "modelchunk")
    bin_file.seek(0)
    for binn in data["bins"]:
        end_bin_offset = get_raw(binn["end_bin_offset"], bin_file)
        size_of_item = end_bin_offset - bin_file.tell()

        write_file = open(SETTINGS["bin_path"] + "/" + get_raw(binn["bin_name"], bin_file) + ".bin", "wb")
        write_file.truncate()

        write_file.write(bin_file.read(size_of_item))
        write_file.close()

        bin_file.seek(end_bin_offset)
Exemple #2
0
def export_mtl(data, path, bin_file):
    """
    Create a material file to be used by wavefront object files.

    Creates a mtl file using some default values plus the paths to the
    materials used by the object files.
    """
    global MATERIALS
    file = open(path + "/" + get_raw(data["file_name"], bin_file) + ".mtl",
                "w")
    file.truncate()

    for material in MATERIALS:
        file.write("newmtl " + str(material) + "\n")
        file.write("Ns 0\n")
        file.write("Ka 0.0\n")
        file.write("Kd 1.0\n")
        file.write("Ks 0.0\n")
        file.write("d 1\n")
        file.write("illum 2\n")

        file.write("map_Ka " + str(material) + ".png\n")
        file.write("map_Kd " + str(material) + ".png\n")
        file.write("map_Ks " + str(material) + ".png\n")

    file.close()
Exemple #3
0
def export_mtl(data, path, bin_file):
    """
    Create a material file to be used by wavefront object files.

    Creates a mtl file using some default values plus the paths to the
    materials used by the object files.
    """
    global MATERIALS
    file = open(path + "/" + get_raw(data["file_name"], bin_file) + ".mtl", "w")
    file.truncate()

    for material in MATERIALS:
        file.write("newmtl " + str(material) + "\n")
        file.write("Ns 0\n")
        file.write("Ka 0.0\n")
        file.write("Kd 1.0\n")
        file.write("Ks 0.0\n")
        file.write("d 1\n")
        file.write("illum 2\n")

        file.write("map_Ka " + str(material) + ".png\n")
        file.write("map_Kd " + str(material) + ".png\n")
        file.write("map_Ks " + str(material) + ".png\n")

    file.close()
Exemple #4
0
    def __init__(self, name, data_string, offset, bin_file):
        self.name = name
        self.data_string = data_string
        self.offset = offset

        self.value = str(get_raw((data_string, offset), bin_file))

        self.update_rect()
Exemple #5
0
def extract_model_chunk():
    bin_file = open(SETTINGS["gif_path"] + "/modelchunk.bin", "rb")

    data = get_formatted_data(bin_file, "wdb", "modelchunk")
    bin_file.seek(0)
    for binn in data["bins"]:
        end_bin_offset = get_raw(binn["end_bin_offset"], bin_file)
        size_of_item = end_bin_offset - bin_file.tell()

        write_file = open(
            SETTINGS["bin_path"] + "/" + get_raw(binn["bin_name"], bin_file) +
            ".bin", "wb")
        write_file.truncate()

        write_file.write(bin_file.read(size_of_item))
        write_file.close()

        bin_file.seek(end_bin_offset)
Exemple #6
0
def extract_wdb():
    """
    Open WORLD.WDB and write all the sections as *.bin files.

    Open WORLD.WDB as specified by the structure found in the wdb format
    file. Export each section of the wdb as a bin file in the folder
    hiearchy found in the header of the wdb file.
    """
    bin_file = open(SETTINGS["wdb_path"], "rb")

    data = get_formatted_data(bin_file, "wdb", "wdb")

    for each_group in data["groups"]:
        group_title = get_raw(each_group["group_title"], bin_file)
        i = -1
        for each_subgroup in each_group["subgroups"]:
            i += 1
            subgroup_title = "sub" + str(i)
            for each_sub_item in each_subgroup["sub_items"]:
                sub_item_title = get_raw(each_sub_item["sub_item_title"],
                                         bin_file)
                item_offset = get_raw(each_sub_item["item_offset"], bin_file)
                size_of_item = get_raw(each_sub_item["size_of_item"], bin_file)

                directory = create_dir(SETTINGS["bin_path"] + "/" +
                                       group_title[:-1] + "/" +
                                       subgroup_title + "/")

                write_file = open(
                    directory + "/" + sub_item_title[:-1] + ".bin", "wb")
                write_file.truncate()

                bin_file.seek(item_offset)
                write_file.write(bin_file.read(size_of_item))
                write_file.close()

    #write gif chunk to be extracted by extract_gif_chunk()
    directory = create_dir(SETTINGS["gif_path"])
    write_file = open(directory + "/gifchunk.bin", "wb")
    bin_file.seek(data["gif_chunk_size"][1] + 4)
    write_file.write(bin_file.read(get_raw(data["gif_chunk_size"], bin_file)))
    write_file.close()

    #write model chunk to be extracted by extract_model_chunk()
    directory = create_dir(SETTINGS["gif_path"])
    write_file = open(directory + "/modelchunk.bin", "wb")
    bin_file.seek(data["model_chunk_size"][1] + 4)
    write_file.write(bin_file.read(get_raw(data["model_chunk_size"],
                                           bin_file)))
    write_file.close()
Exemple #7
0
def export_gif(image, path, bin_file, pretext=""):
    """
    Create a PNG image based on a GIF image.

    The GIF data is specified in the dictionary image and contains formatted
    data based on the wdb structure. The PNG image is saved to path.
    """
    gif_name = get_raw(image["gif_name"], bin_file)
    width = get_raw(image["width"], bin_file)
    height = get_raw(image["height"], bin_file)
    num_colors = get_raw(image["num_colors"], bin_file)

    colors = []
    for color in image["colors"]:
        r = get_raw(color["r"], bin_file)
        g = get_raw(color["g"], bin_file)
        b = get_raw(color["b"], bin_file)
        colors.append((r, g, b))

    rows = []
    y = 0
    for row in image["rows"]:
        x = 0
        rows.append([])
        for pixel in row["pixels"]:
            color_index = get_raw(pixel["color_index"], bin_file)
            c = colors[color_index]
            for i in range(3):
                rows[y].append(c[i])

            x += 1
        y += 1

    #write png
    f = open(path + "/" + pretext + gif_name[:-4] + ".png", "wb")
    f.truncate()
    w = png.Writer(width, height)
    w.write(f, rows)
    f.close()
Exemple #8
0
def export_gif(image, path, bin_file, pretext=""):
    """
    Create a PNG image based on a GIF image.

    The GIF data is specified in the dictionary image and contains formatted
    data based on the wdb structure. The PNG image is saved to path.
    """
    gif_name = get_raw(image["gif_name"], bin_file)
    width = get_raw(image["width"], bin_file)
    height = get_raw(image["height"], bin_file)
    num_colors = get_raw(image["num_colors"], bin_file)

    colors = []
    for color in image["colors"]:
        r = get_raw(color["r"], bin_file)
        g = get_raw(color["g"], bin_file)
        b = get_raw(color["b"], bin_file)
        colors.append((r, g, b))

    rows = []
    y = 0
    for row in image["rows"]:
        x = 0
        rows.append([])
        for pixel in row["pixels"]:
            color_index = get_raw(pixel["color_index"], bin_file)
            c = colors[color_index]
            for i in range(3):
                rows[y].append(c[i])

            x += 1
        y += 1

    #write png
    f = open(path + "/" + pretext + gif_name[:-4] + ".png", "wb")
    f.truncate()
    w = png.Writer(width, height)
    w.write(f, rows)
    f.close()
Exemple #9
0
def extract_wdb():
    """
    Open WORLD.WDB and write all the sections as *.bin files.

    Open WORLD.WDB as specified by the structure found in the wdb format
    file. Export each section of the wdb as a bin file in the folder
    hiearchy found in the header of the wdb file.
    """
    bin_file = open(SETTINGS["wdb_path"], "rb")

    data = get_formatted_data(bin_file, "wdb", "wdb")

    for each_group in data["groups"]:
        group_title = get_raw(each_group["group_title"], bin_file)
        i = -1
        for each_subgroup in each_group["subgroups"]:
            i += 1
            subgroup_title = "sub" + str(i)
            for each_sub_item in each_subgroup["sub_items"]:
                sub_item_title = get_raw(each_sub_item["sub_item_title"], bin_file)
                item_offset = get_raw(each_sub_item["item_offset"], bin_file)
                size_of_item = get_raw(each_sub_item["size_of_item"], bin_file)

                directory = create_dir(SETTINGS["bin_path"] + "/" + group_title[:-1] + "/" + subgroup_title + "/")

                write_file = open(directory + "/" + sub_item_title[:-1] + ".bin", "wb")
                write_file.truncate()

                bin_file.seek(item_offset)
                write_file.write(bin_file.read(size_of_item))
                write_file.close()

    #write gif chunk to be extracted by extract_gif_chunk()
    directory = create_dir(SETTINGS["gif_path"])
    write_file = open(directory + "/gifchunk.bin", "wb")
    bin_file.seek(data["gif_chunk_size"][1]+4)
    write_file.write(bin_file.read(get_raw(data["gif_chunk_size"], bin_file)))
    write_file.close()

    #write model chunk to be extracted by extract_model_chunk()
    directory = create_dir(SETTINGS["gif_path"])
    write_file = open(directory + "/modelchunk.bin", "wb")
    bin_file.seek(data["model_chunk_size"][1]+4)
    write_file.write(bin_file.read(get_raw(data["model_chunk_size"], bin_file)))
    write_file.close()
Exemple #10
0
def export_obj(data, model, bin_file, filename):
    """
    Create a wavefront object file based on the data and model dictionaries.

    The dictionaries data and model are based on the wdb format taken from
    get_formatted_data() in formatter.py. The object file is exported to
    the path specified by filename.
    """

    global MATERIALS
    obj_file = open(filename + ".obj", "w")
    obj_file.truncate()

    obj_file.write("mtllib " + get_raw(data["file_name"], bin_file) + ".mtl\n")

    scale = 1.0
    #WRITE OBJ VERTS, NORMALS, COORDINATES
    for vertex in model["vertices"]:
        obj_file.write("v " + str(get_raw(vertex["x"], bin_file)*scale) + " " + str(-get_raw(vertex["y"], bin_file)*scale) + " " + str(get_raw(vertex["z"], bin_file)*scale) + "\n")

    for text_coord in model["coordinates"]:
        obj_file.write("vt " + str(((get_raw(text_coord["u"], bin_file)*scale)-scale)/scale) + " " + str(((-get_raw(text_coord["v"], bin_file)*scale)-scale)/scale) + "\n")

    for normal in model["normals"]:
        obj_file.write("vn " + str(get_raw(normal["x"], bin_file)*scale) + " " + str(-get_raw(normal["y"], bin_file)*scale) + " " + str(get_raw(normal["z"], bin_file)*scale) + "\n")


    #INTERPRET AND WRITE INDICES
    part_index = 0
    for part in model["parts"]:
        tri_normal = None
        triangles = []

        vert_definitions = []
        text_definitions = []

        if len(part["coordinate_indices"]) > 0 and len(part["coordinate_indices"]) != len(part["indices"]):
            trace("This shouldn't happen!!!!!")

        for index in range(len(part["indices"])):
            vert_indices = []
            text_indices = []
            for axis in range(len(part["indices"][index]["axis"])):
                first = get_raw(part["indices"][index]["axis"][axis]["first"], bin_file)
                second = get_raw(part["indices"][index]["axis"][axis]["second"], bin_file)
                if len(part["coordinate_indices"]) > 0:
                    coordinate_index = get_raw(part["coordinate_indices"][index]["axis"][axis]["coordinate_index"], bin_file)
                else:
                    coordinate_index = 1

                #index definition
                if second >= 32768:
                    vert_definitions.append(first)
                    text_definitions.append(coordinate_index)

                    vert_indices.append(first)
                    text_indices.append(coordinate_index)

                    tri_normal = second - 32768
                #index to a definition, not a definition
                else:
                    vert_indices.append(vert_definitions[first])
                    if len(part["coordinate_indices"]) > 0:
                        text_indices.append(text_definitions[first])
                    else:
                        text_indices.append(1)

            #Triangle format: (verts, normal, coords)
            triangles.append([vert_indices, tri_normal, text_indices])

        red = get_raw(part["part_color_red"], bin_file)
        green = get_raw(part["part_color_green"], bin_file)
        blue = get_raw(part["part_color_blue"], bin_file)

        color = (red, green, blue)

        texture_name = get_raw(part["texture_name"], bin_file)[:-4].replace(" ", "")
        material_name = get_raw(part["material_name"], bin_file).replace(" ", "")

        if texture_name == "":
            name_to_use = material_name
        else:
            name_to_use = texture_name

        dont_add = False
        for each_material in MATERIALS:
            if each_material == name_to_use:
                dont_add = True
        if dont_add == False:
            MATERIALS[name_to_use] = color


        #WRITE INDICES TO OBJ
        obj_file.write("o " + str(part_index) + "\n")

        #try to use texture_name, otherwise material_name
        if texture_name == "":
            obj_file.write("usemtl " + str(material_name) + "\n")
        else:
            obj_file.write("usemtl " + str(texture_name) + "\n")

        for each_triangle in triangles:
            obj_file.write("f ")
            #three iterations for 3 points in a triangle, NOT 3 dimensions (x,y,z)!
            for i in range(3):
                #make sure there are texture coordinates
                if len(each_triangle[2]) > 0:
                    #index/coordinate/normal
                    obj_file.write(str(each_triangle[0][i]+1) + "/" + str(each_triangle[2][i]+1) + "/" + str(each_triangle[1]+1))
                else:
                    #index/1/normal
                    obj_file.write(str(each_triangle[0][i]+1) + "//" + str(each_triangle[1]+1))
                obj_file.write(" ")

            obj_file.write("\n")

        part_index += 1

    obj_file.close()
Exemple #11
0
def extract_pattern(file_path, pattern):
    """
    Attempt to extract data from a .bin file and return the progress made.

    There are five steps to extraction:
     * Interpret data from the wdb using the wdb format file
     * Export the object files using the data
     * Export textures
     * Export materials
     * Export the material file mtl
    These steps are represented by the list of five items returned by the
    function that signify if that step was successful. '_' means success,
    and 'X' means failure.
    """
    global MATERIALS
    global STATS
    bin_file = open(file_path, "rb")

    #reset materials to empty
    MATERIALS = {}

    progress = ["X", "X", "X", "X", "X"]

    #INTERPRET FORMAT FROM FILE
    try:
        data = get_formatted_data(bin_file, "wdb", pattern)
        #trace(str(data))
        progress[0] = "_"
    except:
        trace_error()
        return progress

    #EXPORT OBJ
    try:
        file_name = get_raw(data["file_name"], bin_file)

        #trace(data)
        for component in data["components"]:
            component_name = get_raw(component["component_name"], bin_file)
            if "models" in component:
                model_index = len(component["models"])
                for model in component["models"]:
                    model_index -= 1
                    #determine whether or not to export this LOD model based on SETTINGS
                    export = False
                    if SETTINGS["highest_lod_only"]:
                        if model_index == 0:
                            export = True
                    else:
                        export = True

                    if export:
                        if SETTINGS["highest_lod_only"] and not SETTINGS["lod_labels"]:
                            end_string = ""
                        else:
                            end_string = "_lod" + str(model_index)
                        file_path = file_path.replace("\\", "/")
                        obj_path = create_dir(SETTINGS["obj_path"] + "/" + file_path[file_path.find("/", 3):file_path.rfind("/")] + "/" + file_name + "/")
                        export_obj(data, model, bin_file, obj_path + "/" + component_name + end_string)
            else:
                #no models in this component, only the component header
                pass
        progress[1] = "_"
    except:
        trace_error()
        return progress


    found_materials = []
    #EXPORT TEXTURES
    try:
        num_images = get_raw(data["num_images"], bin_file)

        #export textures embedded in this bin group as .png
        for image in data["images"]:
            #normal gif
            export_gif(image, obj_path, bin_file)
            #special hidden gif, only seen on isle and isle_hi gifs
            if "extra_images" in image:
                image["extra_images"][0]["gif_name"] = image["gif_name"]
                export_gif(image["extra_images"][0], obj_path, bin_file, pretext="hidden_")
            found_materials.append(get_raw(image["gif_name"], bin_file)[:-4])
        progress[2] = "_"
    except:
        trace_error()
        return progress

    #EXPORT MATERIALS
    try:
        #export materials without textures as .png, just their rgb on a 4x4 texture
        for material in MATERIALS:
            if material not in found_materials:
                found_materials.append(material)
                #write 4x4 png of color c
                c = MATERIALS[material]
                rows = []
                for row in range(4):
                    rows.append([])
                    for pixel in range(4):
                        rows[row].append(c[0])
                        rows[row].append(c[1])
                        rows[row].append(c[2])
                f = open(obj_path + "/" + material + ".png", "wb")
                f.truncate()
                w = png.Writer(4, 4)
                w.write(f, rows)
                f.close()


        #statistics for materials
        for material in found_materials:
            found_duplicate = False
            for row in STATS["csv"]["materials"]:
                if row[0] == material:
                    found_duplicate = True
                    row[2] += 1
            if not found_duplicate:
                STATS["csv"]["materials"].append([material, "No", 1])


        progress[3] = "_"
    except:
        trace_error()
        return progress

    #EXPORT MTL FILE
    try:
        export_mtl(data, obj_path, bin_file)
        progress[4] = "_"
    except:
        trace_error()
        return progress

    return progress
Exemple #12
0
def export_obj(data, model, bin_file, filename):
    """
    Create a wavefront object file based on the data and model dictionaries.

    The dictionaries data and model are based on the wdb format taken from
    get_formatted_data() in formatter.py. The object file is exported to
    the path specified by filename.
    """

    global MATERIALS
    obj_file = open(filename + ".obj", "w")
    obj_file.truncate()

    obj_file.write("mtllib " + get_raw(data["file_name"], bin_file) + ".mtl\n")

    scale = 1.0
    #WRITE OBJ VERTS, NORMALS, COORDINATES
    for vertex in model["vertices"]:
        obj_file.write("v " + str(get_raw(vertex["x"], bin_file) * scale) +
                       " " + str(-get_raw(vertex["y"], bin_file) * scale) +
                       " " + str(get_raw(vertex["z"], bin_file) * scale) +
                       "\n")

    for text_coord in model["coordinates"]:
        obj_file.write("vt " + str((
            (get_raw(text_coord["u"], bin_file) * scale) - scale) / scale) +
                       " " +
                       str(((-get_raw(text_coord["v"], bin_file) * scale) -
                            scale) / scale) + "\n")

    for normal in model["normals"]:
        obj_file.write("vn " + str(get_raw(normal["x"], bin_file) * scale) +
                       " " + str(-get_raw(normal["y"], bin_file) * scale) +
                       " " + str(get_raw(normal["z"], bin_file) * scale) +
                       "\n")

    #INTERPRET AND WRITE INDICES
    part_index = 0
    for part in model["parts"]:
        tri_normal = None
        triangles = []

        vert_definitions = []
        text_definitions = []

        if len(part["coordinate_indices"]) > 0 and len(
                part["coordinate_indices"]) != len(part["indices"]):
            trace("This shouldn't happen!!!!!")

        for index in range(len(part["indices"])):
            vert_indices = []
            text_indices = []
            for axis in range(len(part["indices"][index]["axis"])):
                first = get_raw(part["indices"][index]["axis"][axis]["first"],
                                bin_file)
                second = get_raw(
                    part["indices"][index]["axis"][axis]["second"], bin_file)
                if len(part["coordinate_indices"]) > 0:
                    coordinate_index = get_raw(
                        part["coordinate_indices"][index]["axis"][axis]
                        ["coordinate_index"], bin_file)
                else:
                    coordinate_index = 1

                #index definition
                if second >= 32768:
                    vert_definitions.append(first)
                    text_definitions.append(coordinate_index)

                    vert_indices.append(first)
                    text_indices.append(coordinate_index)

                    tri_normal = second - 32768
                #index to a definition, not a definition
                else:
                    vert_indices.append(vert_definitions[first])
                    if len(part["coordinate_indices"]) > 0:
                        text_indices.append(text_definitions[first])
                    else:
                        text_indices.append(1)

            #Triangle format: (verts, normal, coords)
            triangles.append([vert_indices, tri_normal, text_indices])

        red = get_raw(part["part_color_red"], bin_file)
        green = get_raw(part["part_color_green"], bin_file)
        blue = get_raw(part["part_color_blue"], bin_file)

        color = (red, green, blue)

        texture_name = get_raw(part["texture_name"],
                               bin_file)[:-4].replace(" ", "")
        material_name = get_raw(part["material_name"],
                                bin_file).replace(" ", "")

        if texture_name == "":
            name_to_use = material_name
        else:
            name_to_use = texture_name

        dont_add = False
        for each_material in MATERIALS:
            if each_material == name_to_use:
                dont_add = True
        if dont_add == False:
            MATERIALS[name_to_use] = color

        #WRITE INDICES TO OBJ
        obj_file.write("o " + str(part_index) + "\n")

        #try to use texture_name, otherwise material_name
        if texture_name == "":
            obj_file.write("usemtl " + str(material_name) + "\n")
        else:
            obj_file.write("usemtl " + str(texture_name) + "\n")

        for each_triangle in triangles:
            obj_file.write("f ")
            #three iterations for 3 points in a triangle, NOT 3 dimensions (x,y,z)!
            for i in range(3):
                #make sure there are texture coordinates
                if len(each_triangle[2]) > 0:
                    #index/coordinate/normal
                    obj_file.write(
                        str(each_triangle[0][i] + 1) + "/" +
                        str(each_triangle[2][i] + 1) + "/" +
                        str(each_triangle[1] + 1))
                else:
                    #index/1/normal
                    obj_file.write(
                        str(each_triangle[0][i] + 1) + "//" +
                        str(each_triangle[1] + 1))
                obj_file.write(" ")

            obj_file.write("\n")

        part_index += 1

    obj_file.close()
Exemple #13
0
def extract_pattern(file_path, pattern):
    """
    Attempt to extract data from a .bin file and return the progress made.

    There are five steps to extraction:
     * Interpret data from the wdb using the wdb format file
     * Export the object files using the data
     * Export textures
     * Export materials
     * Export the material file mtl
    These steps are represented by the list of five items returned by the
    function that signify if that step was successful. '_' means success,
    and 'X' means failure.
    """
    global MATERIALS
    global STATS
    bin_file = open(file_path, "rb")

    #reset materials to empty
    MATERIALS = {}

    progress = ["X", "X", "X", "X", "X"]

    #INTERPRET FORMAT FROM FILE
    try:
        data = get_formatted_data(bin_file, "wdb", pattern)
        #trace(str(data))
        progress[0] = "_"
    except:
        trace_error()
        return progress

    #EXPORT OBJ
    try:
        file_name = get_raw(data["file_name"], bin_file)

        #trace(data)
        for component in data["components"]:
            component_name = get_raw(component["component_name"], bin_file)
            if "models" in component:
                model_index = len(component["models"])
                for model in component["models"]:
                    model_index -= 1
                    #determine whether or not to export this LOD model based on SETTINGS
                    export = False
                    if SETTINGS["highest_lod_only"]:
                        if model_index == 0:
                            export = True
                    else:
                        export = True

                    if export:
                        if SETTINGS["highest_lod_only"] and not SETTINGS[
                                "lod_labels"]:
                            end_string = ""
                        else:
                            end_string = "_lod" + str(model_index)
                        file_path = file_path.replace("\\", "/")
                        obj_path = create_dir(
                            SETTINGS["obj_path"] + "/" +
                            file_path[file_path.find("/", 3):file_path.
                                      rfind("/")] + "/" + file_name + "/")
                        export_obj(
                            data, model, bin_file,
                            obj_path + "/" + component_name + end_string)
            else:
                #no models in this component, only the component header
                pass
        progress[1] = "_"
    except:
        trace_error()
        return progress

    found_materials = []
    #EXPORT TEXTURES
    try:
        num_images = get_raw(data["num_images"], bin_file)

        #export textures embedded in this bin group as .png
        for image in data["images"]:
            #normal gif
            export_gif(image, obj_path, bin_file)
            #special hidden gif, only seen on isle and isle_hi gifs
            if "extra_images" in image:
                image["extra_images"][0]["gif_name"] = image["gif_name"]
                export_gif(image["extra_images"][0],
                           obj_path,
                           bin_file,
                           pretext="hidden_")
            found_materials.append(get_raw(image["gif_name"], bin_file)[:-4])
        progress[2] = "_"
    except:
        trace_error()
        return progress

    #EXPORT MATERIALS
    try:
        #export materials without textures as .png, just their rgb on a 4x4 texture
        for material in MATERIALS:
            if material not in found_materials:
                found_materials.append(material)
                #write 4x4 png of color c
                c = MATERIALS[material]
                rows = []
                for row in range(4):
                    rows.append([])
                    for pixel in range(4):
                        rows[row].append(c[0])
                        rows[row].append(c[1])
                        rows[row].append(c[2])
                f = open(obj_path + "/" + material + ".png", "wb")
                f.truncate()
                w = png.Writer(4, 4)
                w.write(f, rows)
                f.close()

        #statistics for materials
        for material in found_materials:
            found_duplicate = False
            for row in STATS["csv"]["materials"]:
                if row[0] == material:
                    found_duplicate = True
                    row[2] += 1
            if not found_duplicate:
                STATS["csv"]["materials"].append([material, "No", 1])

        progress[3] = "_"
    except:
        trace_error()
        return progress

    #EXPORT MTL FILE
    try:
        export_mtl(data, obj_path, bin_file)
        progress[4] = "_"
    except:
        trace_error()
        return progress

    return progress
Exemple #14
0
def main():
    args = parser.parse_args()

    file_name = args.file.name
    bin_file = args.file

    data = get_formatted_data(bin_file, "crp", "crp")

    name_of_mod = get_raw(data.get("name_of_mod", ""), bin_file)
    if name_of_mod == "":
        name_of_mod = file_name[:-4]
    output_path = os.path.join(args.output_dir, name_of_mod.decode('utf-8'))
    if not os.path.exists(output_path):
        os.makedirs(output_path)

    end_header_offset = get_raw(data["end_header_offset"], bin_file)

    metadata = {}

    #go through each file found
    for file_header in data["file_headers"]:
        file_name = get_raw(file_header["file_name"], bin_file).decode('utf-8')
        offset_from_header = get_raw(file_header["offset_from_header"],
                                     bin_file)
        file_size = get_raw(file_header["file_size"], bin_file)

        #absolute_offset = offset_from_header+end_header_offset+1
        absolute_offset = offset_from_header + end_header_offset

        bin_file.seek(absolute_offset)
        try:
            id_string = str(unpack(bin_file, "s", 48)).lower()
        except:
            id_string = ""

        #manually search for PNG header in data
        png_header = [137, 80, 78, 71, 13, 10, 26, 10]
        bin_file.seek(absolute_offset)
        found_header = True
        for i in range(8):
            if unpack(bin_file, "B") != png_header[i]:
                found_header = False

        #TEXTURE2D
        if "unityengine.texture2d" in id_string:
            print("found texture2d")
            bin_file.seek(absolute_offset)

            dds_string = ""
            #find "DDS " in the file to mark the start of the dds
            while (True):
                value = unpack(bin_file, "B")
                if value == 68:
                    dds_string += "D"
                elif value == 83:
                    dds_string += "S"
                elif value == 32:
                    dds_string += " "
                else:
                    dds_string = ""

                if dds_string == "DDS ":
                    dds_offset = bin_file.tell() - 4

                    meta_offset = absolute_offset
                    meta_size = dds_offset - absolute_offset

                    final_path = os.path.join(output_path, file_name + '.dds')
                    final_offset = dds_offset
                    final_size = file_size - meta_size
                    break

        #STEAM PREVIEW PNG (AND RANDOM PNGS)
        elif "icolossalframework.importers.image" in id_string or found_header:
            print("found png")
            bin_file.seek(absolute_offset)

            png_string = ""
            #find "PNG" in the file to mark the start of the png
            while (True):
                value = unpack(bin_file, "B")
                if value == 137:
                    png_string += "89"
                elif value == 80:
                    png_string += "50"
                elif value == 78:
                    png_string += "4E"
                elif value == 71:
                    png_string += "47"
                else:
                    png_string = ""

                if png_string == "89504E47":
                    png_offset = bin_file.tell() - 4

                    meta_offset = absolute_offset
                    meta_size = png_offset - absolute_offset

                    final_path = os.path.join(output_path, file_name + ".png")
                    final_offset = png_offset
                    final_size = file_size - meta_size
                    break

        #GENERIC
        else:
            print("found generic")
            meta_offset = absolute_offset
            meta_size = 0

            final_path = os.path.join(output_path, file_name)
            final_offset = absolute_offset
            final_size = file_size

        #add metadata to the metadata dictionary
        if meta_size == 0:
            metadata[final_path] = ""
        else:
            bin_file.seek(meta_offset)
            metadata[final_path] = unpack(bin_file, "s",
                                          meta_size).decode('utf-8', 'ignore')

        #write file
        write_file = open(final_path, "wb")
        write_file.truncate()

        bin_file.seek(final_offset)
        write_file.write(bin_file.read(final_size))
        write_file.close()

    #save the metadata dictionary using json
    with open(os.path.join(output_path, "metadata.json"), "w") as f:
        json.dump(metadata, f, indent=4, sort_keys=True)