fix: merge identifiers across a book's formats

Identifiers are a collection, but the whole dict was replaced per format, so
the last file to report won outright — an EPUB declaring an ASIN, a Google id
and a Calibre id kept none of them once a PDF contributed one ISBN. Accumulate
instead, letting a declared identifier outrank one scraped off a page.

backend/src/chitai/services/metadata_extractor.py

@@ -124,15 +124,36 @@ class Extractor:
         # EPUB tends to give better metadata results over pdf
         sorted_files = sorted(files, key=lambda f: Extractor._get_file_priority(f))
 
+        # Identifiers accumulate across formats instead of replacing each other. Every
+        # other field is a single value where the later, better-trusted format simply
+        # wins, but identifiers are a *collection*: a book holding an EPUB and a PDF
+        # genuinely carries what both of them declare, and merging the dict wholesale
+        # threw away everything the earlier format found. An EPUB that declares an
+        # ASIN, a Google volume id and a Calibre id kept none of them once a PDF
+        # contributed a single ISBN.
+        identifiers: dict[str, str] = {}
+
         for file in sorted_files:
             match get_file_extension(file):
                 case "epub":
-                    metadata = metadata | await EpubExtractor.extract_metadata(file)
+                    extracted = await EpubExtractor.extract_metadata(file)
                 case "pdf":
-                    metadata = metadata | await PdfExtractor.extract_metadata(file)
+                    extracted = await PdfExtractor.extract_metadata(file)
                 case _:
                     break
 
+            # First writer wins per name, and the files are already ordered by how
+            # far their metadata can be trusted. A `dc:identifier` the publisher
+            # declared outranks an ISBN scraped out of a PDF's copyright page, which
+            # routinely prints the ISBNs of other formats and older editions too.
+            for name, value in (extracted.pop("identifiers", None) or {}).items():
+                identifiers.setdefault(name, value)
+
+            metadata = metadata | extracted
+
+        if identifiers:
+            metadata["identifiers"] = identifiers
+
         # Get metadata from file names
         for file in files:
             metadata = FilenameExtractor.extract_metadata(file) | metadata

backend/tests/unit/test_metadata_extractor.py

@@ -1,7 +1,7 @@
 import pytest
 from pathlib import Path
 from datetime import date
-from chitai.services.metadata_extractor import EpubExtractor
+from chitai.services.metadata_extractor import EpubExtractor, Extractor, PdfExtractor
 
 
 @pytest.mark.asyncio()
@@ -15,3 +15,62 @@ class TestEpubExtractor:
         assert metadata["authors"] == ["Herman Melville"]
         assert metadata["published_date"] == date(year=2001, month=7, day=1)
 
+
+
+EPUB = Path("tests/data_files/Metamorphosis - Franz Kafka.epub")
+PDF = Path("tests/data_files/Calculus Made Easy - Silvanus Thompson.pdf")
+
+
+@pytest.mark.asyncio()
+class TestIdentifierMerging:
+    """A book's formats each contribute identifiers; none of them replaces the rest."""
+
+    async def test_every_format_contributes(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        """
+        Identifiers are a collection, not a single value.
+
+        Merging the whole dict meant the last format to report won outright: an EPUB
+        declaring an ASIN, a Google volume id and a Calibre id kept none of them once
+        a PDF contributed one ISBN.
+        """
+
+        async def epub(_file):
+            return {
+                "title": "How Linux Works",
+                "identifiers": {"isbn-13": "9781718500419", "asin": "1718500408"},
+            }
+
+        async def pdf(_file):
+            return {"identifiers": {"isbn-13": "9781718500402", "isbn-10": "1593270356"}}
+
+        monkeypatch.setattr(EpubExtractor, "extract_metadata", epub)
+        monkeypatch.setattr(PdfExtractor, "extract_metadata", pdf)
+
+        metadata = await Extractor.extract_metadata(
+            [Path("How Linux Works.epub"), Path("How Linux Works.pdf")]
+        )
+
+        assert metadata["identifiers"] == {
+            # Declared by the publisher's toolchain, so it outranks the PDF's, which
+            # was scraped off a copyright page that also prints the print edition's.
+            "isbn-13": "9781718500419",
+            "asin": "1718500408",
+            "isbn-10": "1593270356",
+        }
+
+    async def test_a_second_format_that_finds_nothing_erases_nothing(self) -> None:
+        """The PDF fixture carries no ISBN, so it must leave the EPUB's alone."""
+        metadata = await Extractor.extract_metadata([EPUB, PDF])
+
+        assert metadata["identifiers"] == {"id": "http://www.gutenberg.org/5200"}
+
+    async def test_one_format_on_its_own_is_unaffected(self) -> None:
+        metadata = await Extractor.extract_metadata([EPUB])
+
+        assert metadata["identifiers"] == {"id": "http://www.gutenberg.org/5200"}
+
+    async def test_no_identifiers_anywhere_leaves_the_field_absent(self) -> None:
+        """An empty dict would count as extracted metadata and overwrite nothing."""
+        metadata = await Extractor.extract_metadata([PDF])
+
+        assert "identifiers" not in metadata