start-gui.sh et améliorations diverses

copienator/commands/gemini_for_enonce.py

@@ -22,6 +22,7 @@ from copienator import (
     workspace_from_args,
 )
 from copienator.platform import validate_windows_labels
+from copienator.filesystem import staged_directory
 from copienator.utils import compile_to_pdf
 
 
@@ -73,6 +74,9 @@ class RubricItem(BaseModel):
 class GroupRubrics(BaseModel):
     rubrics: list[RubricItem]
 
+class LabelGroups(BaseModel):
+    groups: list[list[str]]
+
 PROMPT_4 = """Je te fournis les questions, le contexte éventuel, et les corrections pour un groupe de questions d'un examen.
 Ta tâche :
 Établir un barème de correction détaillé pour CHAQUE question.
@@ -176,6 +180,104 @@ def find_file(folder: Path, base_name: str) -> Path | None:
             return path
     return None
 
+
+def generate_rubrics(client, group_context_text: str) -> dict[str, str]:
+    """Use the same rubric request for full and selective statement generation."""
+    response = client.models.generate_content(
+        model=MODEL_ID,
+        contents=[types.Content(role="user", parts=[
+            types.Part.from_text(text=PROMPT_4),
+            types.Part.from_text(text=f"--- CONTENU DU GROUPE ---\n{group_context_text}"),
+        ])],
+        config=types.GenerateContentConfig(
+            system_instruction="Rédige tous les barèmes et consignes de notation en français. Conserve les clés JSON, les labels et les formules mathématiques.",
+            temperature=0.2,
+            response_mime_type="application/json",
+            response_json_schema=GroupRubrics.model_json_schema(),
+        ),
+    )
+    rubrics = GroupRubrics.model_validate_json(response.text).rubrics
+    if len({item.label for item in rubrics}) != len(rubrics):
+        raise ValueError("Gemini returned duplicate rubric labels")
+    return {item.label: item.rubric_content for item in rubrics}
+
+
+def validate_groups(groups: list[list[str]], labels: list[str]) -> None:
+    flattened = [label for group in groups for label in group]
+    if (not groups or any(not group for group in groups)
+            or len(flattened) != len(set(flattened)) or set(flattened) != set(labels)):
+        raise CliError("Groups must contain every existing label exactly once")
+
+
+def refine_existing(workspace: EvaluationWorkspace, mode: str, *, api_client=None) -> ExitCode:
+    """Regroup or replace rubrics without regenerating statements or solutions."""
+    labels = workspace.read_labels()
+    if not labels or len(labels) != len(set(labels)):
+        raise CliError("Generate unique question labels before refining the statement")
+    validate_windows_labels(labels)
+    questions = {}
+    for label in labels:
+        safe_label = label.replace("/", "_")
+        parts = []
+        for directory, title in (("Text2", "Question"), ("Sol2", "Correction")):
+            path = workspace.root / directory / f"{safe_label}.tex"
+            if not path.is_file():
+                raise CliError(f"Missing {path}; generate statements and solutions first")
+            parts.append(f"{title} [{label}]:\n{path.read_text(encoding='utf-8')}")
+        questions[label] = "\n".join(parts)
+    context = utils.enonce_total(workspace.root)
+    if api_client is None:
+        if not api_key:
+            raise CliError("GEMINI_API_KEY is not configured")
+        api_client = genai.Client(api_key=api_key)
+
+    if mode == "groups":
+        response = api_client.models.generate_content(
+            model=MODEL_ID,
+            contents=[types.Content(role="user", parts=[types.Part.from_text(text=(
+                "Regroupe ces questions d’examen en groupes cohérents pour la correction "
+                "et l’annotation, selon leurs dépendances et leur contexte commun. "
+                "Ne mélange pas des exercices différents. Conserve l’ordre des questions. "
+                "Chaque label doit apparaître exactement une fois, sans modification. "
+                "Renvoie uniquement un objet JSON groups contenant des listes de labels.\n\n"
+                + context + "\n\n" + "\n\n".join(questions.values())
+            ))])],
+            config=types.GenerateContentConfig(
+                temperature=0.1, response_mime_type="application/json",
+                response_json_schema=LabelGroups.model_json_schema(),
+            ),
+        )
+        groups = LabelGroups.model_validate_json(response.text).groups
+        validate_groups(groups, labels)
+        atomic_write_text(workspace.label_groups_file,
+                          "".join(", ".join(group) + "\n" for group in groups))
+        print(f"Updated label_groups: {len(groups)} Gemini groups.")
+    elif mode == "persp":
+        if not workspace.label_groups_file.is_file():
+            raise CliError("Generate label_groups before generating rubrics")
+        groups = [[label.strip() for label in line.split(",") if label.strip()]
+                  for line in workspace.label_groups_file.read_text(encoding="utf-8").splitlines()
+                  if line.strip()]
+        validate_groups(groups, labels)
+        with staged_directory(workspace.root / "Persp") as staging:
+            for group in groups:
+                print(f"Generating rubric (Persp) for group: {', '.join(group)}...")
+                group_content = (
+                    "Contexte général de l’examen, fourni uniquement pour comprendre les questions :\n"
+                    + context + "\n\nProduis des barèmes UNIQUEMENT pour les labels suivants : "
+                    + ", ".join(group) + "\n\n" + "\n\n".join(questions[label] for label in group)
+                )
+                rubrics = generate_rubrics(api_client, group_content)
+                if set(rubrics) != set(group) or any(not value.strip() for value in rubrics.values()):
+                    raise CliError("Incomplete or unexpected Gemini rubrics; previous Persp preserved")
+                for label, rubric in rubrics.items():
+                    (staging / label.replace("/", "_")).write_text(
+                        f"{label}\n{rubric}", encoding="utf-8")
+        print("Replaced Persp with Gemini rubrics.")
+    else:
+        raise ValueError(f"Unknown statement refinement: {mode}")
+    return ExitCode.SUCCESS
+
 def process_exam(
     workspace: EvaluationWorkspace,
     restart: bool = False,
@@ -701,32 +803,9 @@ def process_exam(
 
         group_context_text = "\n\n---\n\n".join(group_text_parts)
 
-        contents_4 = [
-            types.Content(
-                role="user",
-                parts=[
-                    types.Part.from_text(text=PROMPT_4),
-                    types.Part.from_text(text=f"--- CONTENU DU GROUPE ---\n{group_context_text}"),
-                ],
-            )
-        ]
-
-        config_4 = types.GenerateContentConfig(
-            system_instruction="Rédige tous les barèmes et consignes de notation en français. Conserve les clés JSON, les labels et les formules mathématiques.",
-            temperature=0.2,
-            response_mime_type="application/json",
-            response_json_schema=GroupRubrics.model_json_schema(),
-        )
-
         print(f"Generating rubric (Persp) for group: {', '.join(labels)}...")
         try:
-            response_r = client.models.generate_content(
-                model=MODEL_ID,
-                contents=contents_4,
-                config=config_4
-            )
-            rubrics_data = GroupRubrics.model_validate_json(response_r.text)
-            rubrics_map = {r.label: r.rubric_content for r in rubrics_data.rubrics}
+            rubrics_map = generate_rubrics(client, group_context_text)
         except Exception as e:  # noqa: BLE001 - remote API boundary
             print(f"Error generating rubric for group {labels[0]}: {e}")
             processing_errors.append(str(e))
@@ -822,12 +901,23 @@ def process_exam(
         workspace.labels_file,
         "".join(f"{label}\n" for label in labels_list),
     )
+    atomic_write_text(
+        workspace.label_groups_file,
+        "".join(", ".join(item.label for item in group if isinstance(item, QuestionItem)) + "\n"
+                for group in grouped_extraction.groups
+                if any(isinstance(item, QuestionItem) for item in group)),
+    )
 
     return ExitCode.PARTIAL if processing_errors else ExitCode.SUCCESS
 
 
 def build_parser() -> argparse.ArgumentParser:
     parser = evaluation_parser("Extract exam and solution code via Gemini")
+    actions = parser.add_mutually_exclusive_group()
+    actions.add_argument("--groups-only", action="store_true",
+                         help="Regroup existing questions with Gemini; update only label_groups")
+    actions.add_argument("--persp-only", action="store_true",
+                         help="Replace only Persp with Gemini rubrics for existing groups")
     parser.add_argument(
         "--restart",
         action="store_true",
@@ -841,7 +931,9 @@ def main(argv: Sequence[str] | None = None) -> int:
     return execute(
         parser,
         argv,
-        lambda args: process_exam(
+        lambda args: refine_existing(workspace_from_args(args),
+                                     "groups" if args.groups_only else "persp")
+        if args.groups_only or args.persp_only else process_exam(
             workspace_from_args(args),
             restart=args.restart,
         ),