#!/usr/bin/env python3
import csv, difflib, re, time, urllib.parse, urllib.request
from pathlib import Path
import xml.etree.ElementTree as ET

PAPERS = [
("23.11", "MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers"),
("24.05", "NeurCross: a neural approach to computing cross fields for quad mesh generation"),
("24.05", "MeshXL: neural coordinate field for generative 3D foundation models"),
("24.06", "MeshAnything: artist-created mesh generation with autoregressive transformers"),
("24.08", "MeshAnything V2: artist-created mesh generation with adjacent mesh tokenization"),
("24.09", "EdgeRunner: auto-regressive auto-encoder for artistic mesh generation"),
("24.11", "Scaling mesh generation via compressive tokenization"),
("24.12", "Meshtron: high-fidelity, artist-like 3D mesh generation at scale"),
("25.01", "Nautilus: locality-aware autoencoder for scalable mesh generation"),
("25.03", "TreeMeshGPT: artistic mesh generation with autoregressive tree sequencing"),
("25.03", "DeepMesh: auto-regressive artist-mesh creation with reinforcement learning"),
("25.03", "MeshCraft: exploring efficient and controllable mesh generation with flow-based DiTs"),
("25.05", "Mesh-RFT: enhancing mesh generation via fine-grained reinforcement fine-tuning"),
("25.06", "CrossGen: learning and generating cross fields for quad meshing"),
("25.07", "Topology-preserved auto-regressive mesh generation in the manner of weaving silk"),
("25.08", "VertexRegen: mesh generation with continuous level of detail"),
("25.08", "FastMesh: Efficient Artistic Mesh Generation via Component Decoupling"),
("25.09", "QuadGPT: Native Quadrilateral Mesh Generation with Autoregressive Models"),
("25.09", "ARMesh: autoregressive mesh generation via next-level-of-detail prediction"),
("25.09", "MeshMosaic: scaling artist mesh generation via local-to-global assembly"),
("25.10", "Topology sculptor, shape refiner: discrete diffusion model for high-fidelity 3D meshes generation"),
("25.12", "MeshRipple: structured autoregressive generation of artist-meshes"),
("26.03", "LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents"),
("26.03", "Mesh-pro: asynchronous advantage-guided ranking preference optimization for artist-style quadrilateral mesh generation"),
("26.03", "FACE: a face-based autoregressive representation for high-fidelity and efficient mesh generation"),
("26.03", "TopGen: learning structural layouts and cross-fields for quadrilateral mesh generation"),
("26.03", "TopoMesh: high-fidelity mesh autoencoding via topological unification"),
("26.04", "Strips as tokens: artist mesh generation with native UV segmentation"),
("26.04", "SQuadGen: generating simple quad layouts via chart distance fields"),
("26.05", "QuadLink: autoregressive quad-dominant mesh generation via point-relation learning"),
("26.06", "MeshFlow: Efficient Artistic Mesh Generation via MeshVAE and Flow-based Diffusion Transformer"),
("26.06", "TriFlow: generating artist-like 3D mesh topology via nearest-vertex vector fields"),
("26.06", "MeshFlow: mesh generation with equivariant flow matching"),
("26.06", "MeshWeaver: sparse-voxel-guided surface weaving for autoregressive mesh generation"),
("26.06", "Mesh BDF: barycentric dominance field for 3D native mesh generation"),
("26.07", "Nexus: native mesh generation with diffusion"),
("26.07", "LATO.2: factorized 3D mesh generation with vertex and topology flow"),
("26.07", "Meshy T2: fast native mesh generation with flow matching"),
]
NS={"a":"http://www.w3.org/2005/Atom"}
def norm(s): return re.sub(r"[^a-z0-9]+", " ", s.lower()).strip()
def slug(s): return re.sub(r"[^A-Za-z0-9._-]+", "_", s).strip("_")[:120]
def get(url):
    req=urllib.request.Request(url,headers={"User-Agent":"paper-archiver/1.0 contact=local"})
    with urllib.request.urlopen(req,timeout=60) as r:return r.read()

out=Path("papers/pdfs"); out.mkdir(parents=True,exist_ok=True)
rows=[]
for requested_date,title in PAPERS:
    q=urllib.parse.urlencode({"search_query":f'ti:"{title}"',"start":0,"max_results":5})
    try:
        root=ET.fromstring(get("https://export.arxiv.org/api/query?"+q))
        candidates=[]
        for e in root.findall("a:entry",NS):
            found=" ".join((e.findtext("a:title",default="",namespaces=NS)).split())
            score=difflib.SequenceMatcher(None,norm(title),norm(found)).ratio()
            candidates.append((score,e,found))
        candidates.sort(key=lambda x:x[0],reverse=True)
        if not candidates or candidates[0][0] < .72:
            rows.append([requested_date,title,"","","not_found","",f"best_score={candidates[0][0]:.3f}" if candidates else "no result"]); continue
        score,e,found=candidates[0]
        abs_url=e.findtext("a:id",default="",namespaces=NS).replace("http://","https://")
        aid=abs_url.split("/abs/")[-1].split("v")[0]
        published=e.findtext("a:published",default="",namespaces=NS)[:10]
        filename=f"{aid}_{slug(found)}.pdf"
        path=out/filename
        if not path.exists(): path.write_bytes(get(f"https://arxiv.org/pdf/{aid}"))
        rows.append([requested_date,title,found,published,"downloaded",f"https://arxiv.org/abs/{aid}",filename,f"match={score:.3f}"])
    except Exception as ex:
        rows.append([requested_date,title,"","","error","","",repr(ex)])
    # arXiv asks clients to avoid bursty access; PDF downloads dominate runtime.
    time.sleep(1)

manifest=Path("papers/manifest.csv")
with manifest.open("w",newline="",encoding="utf-8") as f:
    w=csv.writer(f); w.writerow(["requested_date","requested_title","verified_title","published","status","source_url","filename","note"]);w.writerows(rows)
print("\n".join(f"{r[4]:10} {r[0]} {r[1]} -> {r[2]}" for r in rows))