Improve LLM and CSV processing

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

drhider/obfuscator.py

@@ -17,10 +17,19 @@ from . import extractor
 from . import scanner
 from . import replacer
 from . import builder
+from .names import resolve_name
 
 log = logging.getLogger("drhider")
 
 
+class LLMIncompleteError(Exception):
+    """Processing cannot safely return files when one or more LLM chunks failed."""
+
+    def __init__(self, llm_errors: int):
+        self.llm_errors = llm_errors
+        super().__init__("LLM processing incomplete")
+
+
 def _dedupe_file_names(
     files: List[Tuple[str, bytes, str]]
 ) -> List[Tuple[str, bytes, str]]:
@@ -39,17 +48,9 @@ def _dedupe_file_names(
     out: List[Tuple[str, bytes, str]] = []
 
     for fname, content, ctype in files:
-        candidate = fname
-        if candidate in seen:
-            if seen[candidate] == content:
-                continue  # точный дубль — пропускаем
-            base, ext = os.path.splitext(fname)
-            n = 2
-            while f"{base}_{n}{ext}" in seen:
-                n += 1
-            candidate = f"{base}_{n}{ext}"
-        seen[candidate] = content
-        out.append((candidate, content, ctype))
+        candidate = resolve_name(fname, content, seen)
+        if candidate is not None:
+            out.append((candidate, content, ctype))
 
     return out
 
@@ -177,13 +178,18 @@ class TwoPassObfuscator:
             cancelled_llm = cancelled_extract
             if not cancelled_llm and self._llm_client:
                 try:
-                    scanner.scan_llm_ner(
+                    llm_failures = scanner.scan_llm_ner(
                         all_texts, self._mapping, self._llm_client, self._counters,
                         cancel_event=cancel_event, file_progress=_fp,
                     )
-                except scanner.CancelRequested:
+                except scanner.CancelRequested as e:
+                    if e.llm_failures:
+                        raise LLMIncompleteError(e.llm_failures)
                     log.info("obfuscate: LLM cancelled, llm_done=%d", len(llm_done))
                     cancelled_llm = True
+                else:
+                    if llm_failures:
+                        raise LLMIncompleteError(llm_failures)
 
             # Предсортировать ключи один раз (по убыванию длины)
             self._sorted_keys = sorted(
@@ -220,7 +226,18 @@ class TwoPassObfuscator:
                 t0 = time.time()
                 obf_content = content  # По умолчанию — без изменений
 
-                if fname.endswith('.doc'):
+                if fname.lower().endswith('.csv'):
+                    try:
+                        obf_content = extractor.obfuscate_csv(
+                            content, self._mapping, self._sorted_keys, self._compiled_re
+                        )
+                    except Exception as e:
+                        log.warning("obfuscate: skip CSV %r during replacement: %r", fname, e)
+                        skipped.add(fname)
+                        if progress_cb:
+                            progress_cb("done", i, total, display_name, 0.0)
+                        continue
+                elif fname.endswith('.doc'):
                     # .doc — конвертируется через liberta, обрабатывается как .docx
                     md_text = all_texts.get(fname, '')
                     replaced = replacer.apply_replacements(