app.py

import os
import tempfile
import uuid
import zipfile
import shutil
from lxml import etree
import gradio as gr
from concurrent.futures import ThreadPoolExecutor
from functools import lru_cache

from huggingface_hub import InferenceClient

##############################################
# Hugging Face Inference Client Initialization
##############################################

hf_token = os.getenv("HF_TOKEN")
if not hf_token:
    raise ValueError("Hugging Face token not found. Please set it in the environment.")
client = InferenceClient(token=hf_token)
llama_model_id = "google/gemma-2-27b-it"  # or any model you prefer
client.model = llama_model_id  # Set model once

##############################################
# Define Supported Languages
##############################################

languages = {
    "Arabic": "ar",
    "Bulgarian": "bg",
    "Catalan": "ca",
    "Czech": "cs",
    "German": "de",
    "English": "en",
    "Spanish": "es",
    "Estonian": "et",
    "Finnish": "fi",
    "French": "fr",
    "Hungarian": "hu",
    "Italian": "it",
    "Lithuanian": "lt",
    "Latvian": "lv",
    "Dutch": "nl",
    "Polish": "pl",
    "Portuguese": "pt",
    "Romanian": "ro",
    "Russian": "ru",
    "Slovak": "sk",
    "Slovenian": "sl",
    "Swedish": "sv",
    "Turkish": "tr",
    "Ukrainian": "uk",
    "Vietnamese": "vi",
    # Add more languages as needed
}

sorted_languages = sorted(languages.items())

##############################################
# Single Translation Helper Function with Dynamic Language Prompt
##############################################

def single_translate(target_lang, text):
    # Map language codes to names using the languages dictionary
    language_names = {v: k for k, v in languages.items()}
    language = language_names.get(target_lang, "Arabic")  # Default to Arabic if not found

    # Construct prompt dynamically based on the selected language
    prompt = f"You are a professional Translator: don't say a greeting or anything just jump into the translation: Make sure you don't output anything else other than the {language} translation requested for the upcoming text:{text} also dont use curly brackets or such just the translation."
    try:
        output = client.text_generation(prompt)
        # Handle different possible response types and return the translated text directly
        if isinstance(output, str):
            return output.strip()
        elif isinstance(output, list) and len(output) > 0:
            first_item = output[0]
            if isinstance(first_item, dict) and "generated_text" in first_item:
                return first_item["generated_text"].strip()
            else:
                return str(output).strip()
        else:
            return str(output).strip()
    except Exception as e:
        return f"Translation error: {e}"

##############################################
# Parallel Translation Function
##############################################

@lru_cache(maxsize=10000)
def translate_texts(target_lang, texts):
    translations = []
    max_workers = 8  # Adjust based on performance and API constraints
    with ThreadPoolExecutor(max_workers=max_workers) as executor:
        futures = [executor.submit(single_translate, target_lang, text) for text in texts]
        for future in futures:
            translations.append(future.result())
    return translations

##############################################
# PPTX Processing Functions
##############################################

def get_slide_size_from_presentation(xml_root):
    fallback_width = 914400 * 10  
    fallback_height = 914400 * 7   
    nsmap = {"p": "http://schemas.openxmlformats.org/presentationml/2006/main"}
    sldSz_elems = xml_root.xpath("//p:presentation/p:sldSz", namespaces=nsmap)
    if sldSz_elems:
        sldSz = sldSz_elems[0]
        cx_attr = sldSz.get("cx")
        cy_attr = sldSz.get("cy")
        if cx_attr and cy_attr:
            try:
                slide_width_emu = int(cx_attr)
                slide_height_emu = int(cy_attr)
                return (slide_width_emu, slide_height_emu)
            except:
                pass
    return (fallback_width, fallback_height)

def mirror_x_around_center(off_x, ext_cx, slide_width):
    shape_center = off_x + (ext_cx / 2.0)
    slide_center = slide_width / 2.0
    displacement = shape_center - slide_center
    mirrored_center = slide_center - displacement
    new_off_x = mirrored_center - (ext_cx / 2.0)
    return int(new_off_x)

def reflect_translate_align_xml(xml_path, slide_width_emu, target_lang, do_mirroring=True, do_text_translation=True, do_right_align=True):
    nsmap = {
        "p": "http://schemas.openxmlformats.org/presentationml/2006/main",
        "a": "http://schemas.openxmlformats.org/drawingml/2006/main"
    }
    tree = etree.parse(xml_path)
    root_elem = tree.getroot()
    changed = False

    if do_mirroring:
        xfrm_elems = root_elem.xpath(
            ".//p:spPr/a:xfrm | .//p:pic/p:spPr/a:xfrm | .//p:cxnSp/p:spPr/a:xfrm",
            namespaces=nsmap
        )
        for xfrm in xfrm_elems:
            off_elem = xfrm.find("a:off", nsmap)
            ext_elem = xfrm.find("a:ext", nsmap)
            if off_elem is not None and ext_elem is not None:
                try:
                    off_x = int(off_elem.get("x", "0"))
                    ext_cx = int(ext_elem.get("cx", "0"))
                    new_off_x = mirror_x_around_center(off_x, ext_cx, slide_width_emu)
                    off_elem.set("x", str(new_off_x))
                    changed = True
                except:
                    pass

    if do_text_translation:
        t_elems = root_elem.xpath(".//a:t", namespaces=nsmap)
        texts = [a_t.text.strip() for a_t in t_elems if a_t.text and a_t.text.strip()]
        unique_texts = list(set(texts))
        translations = translate_texts(target_lang, tuple(unique_texts)) if texts else []

        translation_map = dict(zip(unique_texts, translations))
        for a_t in t_elems:
            original_text = a_t.text.strip() if a_t.text else ""
            if original_text in translation_map:
                a_t.text = translation_map.get(original_text, original_text)
                changed = True

    if do_right_align:
        pPr_elems = root_elem.xpath(".//a:p/a:pPr", namespaces=nsmap)
        for pPr in pPr_elems:
            pPr.set("algn", "r")
            changed = True

    if changed:
        tree.write(xml_path, xml_declaration=True, encoding="UTF-8", standalone="yes")
    return changed

def direct_raw_translate_reflect_pptx_with_progress(pptx_path, target_lang, do_mirroring):
    try:
        temp_dir = os.path.join(tempfile.gettempdir(), f"raw_{uuid.uuid4().hex}")
        os.makedirs(temp_dir, exist_ok=True)
        yield (None, "10% - Extracting PPTX contents...")

        with zipfile.ZipFile(pptx_path, "r") as z_in:
            z_in.extractall(temp_dir)

        presentation_xml = os.path.join(temp_dir, "ppt", "presentation.xml")
        slide_width_emu, slide_height_emu = (9144000, 6858000)
        if os.path.exists(presentation_xml):
            pres_tree = etree.parse(presentation_xml)
            pres_root = pres_tree.getroot()
            slide_width_emu, slide_height_emu = get_slide_size_from_presentation(pres_root)
        yield (None, "20% - Retrieved slide size.")

        xml_files = []
        for sub_dir in ["slideMasters", "slideLayouts", "slides"]:
            dir_path = os.path.join(temp_dir, "ppt", sub_dir)
            if os.path.isdir(dir_path):
                for fname in os.listdir(dir_path):
                    if fname.endswith(".xml"):
                        xml_files.append(os.path.join(dir_path, fname))
        total_files = len(xml_files)
        yield (None, f"30% - Found {total_files} XML files to process.")

        with ThreadPoolExecutor() as executor:
            for idx, _ in enumerate(executor.map(lambda path: reflect_translate_align_xml(
                path, slide_width_emu, target_lang, do_mirroring, do_right_align=do_mirroring), xml_files), start=1):
                progress_percent = 30 + (50 * idx / total_files)
                yield (None, f"{int(progress_percent)}% - Processed {idx}/{total_files} XML files.")

        yield (None, "80% - Fixing color schemes...")

        ppt_dir = os.path.join(temp_dir, "ppt")
        for root_dir, dirs, files in os.walk(ppt_dir):
            for file in files:
                if file.endswith(".xml"):
                    xml_path = os.path.join(root_dir, file)
                    tree = etree.parse(xml_path)
                    root_elem = tree.getroot()
                    nsmap = {"a": "http://schemas.openxmlformats.org/drawingml/2006/main"}
                    scheme_clr_elems = root_elem.xpath(".//a:schemeClr", namespaces=nsmap)
                    changed2 = False
                    for sc in scheme_clr_elems:
                        val_attr = sc.get("val")
                        if not val_attr or val_attr.lower() == "none":
                            srgb_elem = etree.Element("{http://schemas.openxmlformats.org/drawingml/2006/main}srgbClr")
                            srgb_elem.set("val", "000000")
                            parent = sc.getparent()
                            parent.replace(sc, srgb_elem)
                            changed2 = True
                    if changed2:
                        tree.write(xml_path, xml_declaration=True, encoding="UTF-8", standalone="yes")

        yield (None, "90% - Re-zipping the PPTX file...")

        translated_path = os.path.join(tempfile.gettempdir(), f"translated_{uuid.uuid4().hex}.pptx")
        with zipfile.ZipFile(translated_path, "w", zipfile.ZIP_DEFLATED) as z_out:
            for root_dir, dirs, files in os.walk(temp_dir):
                for file in files:
                    full_path = os.path.join(root_dir, file)
                    rel_path = os.path.relpath(full_path, temp_dir)
                    z_out.write(full_path, arcname=rel_path)

        shutil.rmtree(temp_dir, ignore_errors=True)
        yield (translated_path, "✅ Processing complete.")
    except Exception as e:
        yield (None, f"❌ Error: {str(e)}")

def raw_xml_reflect_and_translate(file, target_lang, do_mirroring):
    if not file:
        yield (None, "❌ No file uploaded.")
        return
    try:
        yield from direct_raw_translate_reflect_pptx_with_progress(file, target_lang, do_mirroring)
    except Exception as e:
        yield (None, f"❌ Error: {str(e)}")

##############################################
# GRADIO DEMO: REFLECT + TRANSLATE + RIGHT-ALIGN
##############################################

with gr.Blocks() as demo:
    gr.Markdown("<h1>tarGEM 💎</h1>")
    gr.Markdown("""
    <h3>PowerPoint Reflection & Translation Tool</h3>
    <p>Enhance your presentations with mirrored layouts, translated text, and aligned bullet points.</p>
    <ul>
        <li>💎 <strong>Mirror</strong> shapes for balanced designs</li>
        <li>🔄 <strong>Translate</strong> text seamlessly</li>
        <li>📌 <strong>Align</strong> to the right using RTL checkbox</li>
        <li>🎨 <strong>Keep your original</strong> font style, slide background and color schemes</li>
    </ul>
    """)

    ppt_input = gr.File(label="📂 Upload PPTX File", file_types=[".pptx"], type="filepath")

    language_dropdown = gr.Dropdown(
        label="🌐 Select Target Language",
        choices=[(name, code) for name, code in sorted_languages],
        value="ar",
        type="value",
        interactive=True
    )

    mirror_checkbox = gr.Checkbox(
        label="🔄 Mirror Shapes for RTL",
        value=True,
        interactive=True
    )

    reflect_button = gr.Button("Process PPTX")

    output_file = gr.File(label="🔄 Download Processed PPTX", interactive=False)

    status_box = gr.Textbox(label="Status", interactive=False)

    reflect_button.click(
        fn=raw_xml_reflect_and_translate,
        inputs=[ppt_input, language_dropdown, mirror_checkbox],
        outputs=[output_file, status_box],
        queue=True
    )

    gr.Markdown("""
    <p style="text-align: center; margin-top: 20px;">
        💎 Thank you for using tarGEM!
    </p>
    """)

demo.launch(share=True)