Personal tex compilation fixes

copienator/commands/gemini_for_enonce.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import argparse
+import hashlib
 import io
 import json
 import re
@@ -1016,24 +1017,35 @@ def write_exam_outputs(workspace: EvaluationWorkspace, grouped_extraction: Group
         save_missing(path, '\n\n'.join(contents))
 
     print(f"Wrote {len(grouped_extraction.groups)} groups; checking PDF compilation...")
-    # ==========================================
-    # PDF COMPILATION (4 Threads)
-    # ==========================================
-    all_tex_files = list(text2_dir.glob("*.tex")) + list(sol2_dir.glob("*.tex"))
+    compilation_status = compile_existing_pdfs(workspace)
+    return ExitCode.PARTIAL if processing_errors or compilation_status != ExitCode.SUCCESS else ExitCode.SUCCESS
+
+
+def compile_existing_pdfs(workspace: EvaluationWorkspace) -> ExitCode:
+    """Compile existing fragments without reading checkpoints or calling Gemini."""
+    folder = workspace.root
+    all_tex_files = list(workspace.rendered_text_dir.glob("*.tex")) + list(workspace.rendered_solution_dir.glob("*.tex"))
 
     def compile_worker(tex_file: Path) -> str | None:
         """Helper to read content and call the utility function."""
         try:
             pdf_path = tex_file.with_suffix('.pdf')
             status_path = folder / 'Cache' / 'statement_pdf_status' / tex_file.parent.name / (tex_file.stem + '.json')
-            failed_before = status_path.exists() and not json.loads(status_path.read_text()).get('complete')
+            source_path = (workspace.correction_source_file if tex_file.parent.name == 'Sol2'
+                           else workspace.statement_source_file)
+            if not source_path.is_file():
+                source_path = None
+            fingerprint = hashlib.sha256((utils.fragment_latex_header(source_path)
+                                          + utils.LATEX_AFTER + str(source_path)).encode()).hexdigest()
+            status = json.loads(status_path.read_text()) if status_path.exists() else {}
+            failed_before = not status.get('complete') or status.get('preamble_hash') != fingerprint
             if (pdf_path.is_file() and pdf_path.stat().st_size > 0
                     and pdf_path.stat().st_mtime_ns >= tex_file.stat().st_mtime_ns and not failed_before):
                 return None
-            atomic_write_json(status_path, {'complete': False})
+            atomic_write_json(status_path, {'complete': False, 'preamble_hash': fingerprint})
             content = tex_file.read_text(encoding='utf-8')
-            compile_to_pdf(content, pdf_path, strict=True)
-            atomic_write_json(status_path, {'complete': True})
+            compile_to_pdf(content, pdf_path, strict=True, source_path=source_path)
+            atomic_write_json(status_path, {'complete': True, 'preamble_hash': fingerprint})
         except Exception as e:  # noqa: BLE001 - compiler worker boundary
             return f"Error compiling {tex_file.name}: {e}"
         return None
@@ -1045,8 +1057,7 @@ def write_exam_outputs(workspace: EvaluationWorkspace, grouped_extraction: Group
         ]
     for error in compile_errors:
         report_error(folder, 'PDF compilation', error)
-    processing_errors.extend(compile_errors)
-    return ExitCode.PARTIAL if processing_errors else ExitCode.SUCCESS
+    return ExitCode.PARTIAL if compile_errors else ExitCode.SUCCESS
 
 
 def build_parser() -> argparse.ArgumentParser:
@@ -1056,6 +1067,8 @@ def build_parser() -> argparse.ArgumentParser:
                          help="Regroup existing questions with Gemini; update only label_groups")
     actions.add_argument("--persp-only", action="store_true",
                          help="Generate only missing Persp rubrics for existing groups")
+    actions.add_argument("--compile-only", action="store_true",
+                         help="Retry existing fragment PDFs using source preambles; no Gemini requests")
     parser.add_argument('--delete-persp', action='store_true',
                         help='Delete current Persp before generation (requires --persp-only)')
     parser.add_argument(
@@ -1071,6 +1084,10 @@ def main(argv: Sequence[str] | None = None) -> int:
     def run(args):
         if args.delete_persp and not args.persp_only:
             raise CliError('--delete-persp requires --persp-only', ExitCode.INVALID_ARGUMENTS)
+        if args.restart and (args.compile_only or args.groups_only or args.persp_only):
+            raise CliError('--restart cannot be combined with a partial action', ExitCode.INVALID_ARGUMENTS)
+        if args.compile_only:
+            return compile_existing_pdfs(workspace_from_args(args))
         if args.groups_only or args.persp_only:
             kwargs = {'delete_persp': True} if args.delete_persp else {}
             return refine_existing(workspace_from_args(args),

copienator/utils.py

@@ -149,9 +149,43 @@ def report_latex_warning(output_pdf_path) -> None:
     print(f"{LATEX_WARNING_MARKER}{Path(output_pdf_path).resolve()}")
 
 
-def compile_to_pdf(text, output_pdf_path, *, strict=False):
+def fragment_latex_header(source_path=None):
+    """Keep source packages/macros while using the configured fragment layout."""
+    if source_path is None:
+        return LATEX_BEFORE
+    source = Path(source_path).read_text(encoding="utf-8")
+    # Ignore commented document commands when locating the preamble.
+    source = re.sub(r"(?m)(?<!\\)%[^\n]*", "", source)
+    begin = re.search(r"\\begin\s*\{document\}", source)
+    document_class = r"\\documentclass(?:\s*\[[^\]]*\])?\s*\{[^}]+\}"
+    configured_class = re.search(document_class, LATEX_BEFORE)
+    configured_begin = re.search(r"\\begin\s*\{document\}", LATEX_BEFORE)
+    if begin is None or configured_class is None or configured_begin is None:
+        return LATEX_BEFORE
+    preamble = source[:begin.start()]
+    if not re.search(document_class, preamble):
+        return LATEX_BEFORE
+    preamble = re.sub(document_class, lambda _: configured_class.group(), preamble, count=1)
+    # geometry overrides standalone's crop coordinates and can produce blank or
+    # clipped PDFs even when LaTeX succeeds. Keep other source packages intact.
+    def without_geometry(match):
+        packages = [name.strip() for name in match.group(2).split(',')
+                    if name.strip() != 'geometry']
+        return (r'\usepackage' + (match.group(1) or '') + '{' + ','.join(packages) + '}'
+                if packages else '')
+    preamble = re.sub(r'\\usepackage\s*(\[[^\]]*\])?\s*\{([^}]+)\}',
+                      without_geometry, preamble)
+    preamble = re.sub(r'\\(?:geometry|newgeometry)\s*\{[^}]*\}', '', preamble)
+    return preamble + LATEX_BEFORE[configured_begin.start():]
+
+
+def compile_to_pdf(text, output_pdf_path, *, strict=False, source_path=None):
     """Wraps text in standalone header/footer templates and compiles it to PDF."""
-    latex_template = f"{LATEX_BEFORE}{text}{LATEX_AFTER}"
+    latex_template = f"{fragment_latex_header(source_path)}{text}{LATEX_AFTER}"
+    env = os.environ.copy()
+    if source_path is not None:
+        source_dir = str(Path(source_path).resolve().parent)
+        env['TEXINPUTS'] = source_dir + os.pathsep + env.get('TEXINPUTS', '')
 
     with tempfile.TemporaryDirectory() as temp_dir:
         tex_filename = 'text.tex'
@@ -170,6 +204,7 @@ def compile_to_pdf(text, output_pdf_path, *, strict=False):
             result = subprocess.run(
                 ['pdflatex', '-interaction=nonstopmode', tex_filename],
                 cwd=temp_dir,
+                env=env,
                 stdout=subprocess.PIPE,
                 stderr=subprocess.STDOUT,
                 text=True,
@@ -180,6 +215,7 @@ def compile_to_pdf(text, output_pdf_path, *, strict=False):
                 result = subprocess.run(
                     ['pdflatex', '-interaction=nonstopmode', tex_filename],
                     cwd=temp_dir,
+                    env=env,
                     stdout=subprocess.PIPE,
                     stderr=subprocess.STDOUT,
                     text=True,
@@ -188,7 +224,7 @@ def compile_to_pdf(text, output_pdf_path, *, strict=False):
 
             generated_pdf = os.path.join(temp_dir, pdf_filename)
             pdf_generated = os.path.exists(generated_pdf)
-            if pdf_generated:
+            if pdf_generated and (not strict or result.returncode == 0):
                 shutil.move(generated_pdf, output_pdf_path)
 
             if result.returncode != 0:

copienator/workspace.py

@@ -65,8 +65,15 @@ class EvaluationWorkspace:
         )
         if not path.is_dir():
             return False
-        aliases = (f"{path.name}.pdf", f"{path.name}.tex", f"{path.name}c.tex")
-        return any((path / marker).exists() for marker in (*markers, *aliases))
+        workspace = EvaluationWorkspace(path)
+        inputs = (
+            candidate
+            for name in ("enonce.pdf", "enonce.tex", "correction.tex")
+            for candidate in workspace.input_candidates(name)
+        )
+        return any((path / marker).exists() for marker in markers) or any(
+            candidate.exists() for candidate in inputs
+        )
 
     @property
     def name(self) -> str:
@@ -75,9 +82,9 @@ class EvaluationWorkspace:
     def input_candidates(self, canonical_name: str) -> tuple[Path, ...]:
         """Return preferred paths for one of the three statement inputs."""
         aliases = {
-            "enonce.pdf": ("enonce.pdf", "énoncé.pdf", f"{self.name}.pdf"),
-            "enonce.tex": ("enonce.tex", f"{self.name}.tex"),
-            "correction.tex": ("correction.tex", f"{self.name}c.tex"),
+            "enonce.pdf": ("enonce.pdf", "énonce.pdf", "énoncé.pdf", f"{self.name}.pdf"),
+            "enonce.tex": ("enonce.tex", "énonce.tex", "énoncé.tex", f"{self.name}.tex"),
+            "correction.tex": ("correction.tex", "corrige.tex", "corrigé.tex", f"{self.name}c.tex"),
         }
         names = aliases.get(canonical_name, (canonical_name,))
         return tuple(self.root / name for name in names)