fix: export only what the platform declares

scripts/common.py

@@ -159,7 +159,11 @@ def load_platform_config(platform_name: str, platforms_dir: str = "platforms") -
                         key = (gf.get("name"), gf.get("destination", gf.get("name")))
                         dest_lower = gf.get("destination", gf.get("name", "")).lower()
                         if key not in existing and dest_lower not in existing_lower:
-                            system.setdefault("files", []).append(gf)
+                            # Marked as ours: the platform's own file does not
+                            # carry it, and an export must not count it kept.
+                            system.setdefault("files", []).append(
+                                {**gf, "shared_group": group_name}
+                            )
                             existing.add(key)
 
     # Merge metadata from _registry.yml. The registry is our curated source;

scripts/export_native.py

@@ -205,7 +205,34 @@ def load_inputs(
         scraped = load_platform_config(platform, platforms_dir)
     except (FileNotFoundError, OSError):
         scraped = None
-    return truth, scraped
+    return truth, _native_layer(scraped)
+
+
+def _native_layer(config: dict | None) -> dict | None:
+    """The platform config without what we add to it.
+
+    A shared group's files join the config for the pack, but the platform's
+    file does not carry them: counted kept, they were written into
+    libretro's System.dat (N88EXT0-3.ROM) as if libretro declared them. What
+    a core needs from a group reaches the export through the truth, as an
+    addition the format's gates apply to. A scraper's mirror of an entry
+    under another system (RomM's tg16 firmware under turbografx-cd), and a
+    file our scraper adds from the cores' source (Dolphin's DSP ROMs in
+    RetroArch's config), are not in the platform's file either.
+    """
+    if config is None:
+        return None
+    systems = {
+        sys_id: {
+            **system,
+            "files": [
+                f for f in system.get("files", [])
+                if not ({"shared_group", "mirror_of", "curated"} & f.keys())
+            ],
+        }
+        for sys_id, system in config.get("systems", {}).items()
+    }
+    return {**config, "systems": systems}
 
 
 def refresh_cache(

scripts/scraper/base_scraper.py

@@ -35,6 +35,9 @@ class BiosRequirement:
     # is not always derivable from ours: libretro writes iplromco.dat bare
     # but ep128emu/roms/cpc464.rom with its directory.
     native_path: str | None = None
+    # The native system whose entry this one repeats under another system.
+    # The platform's own file does not carry the copy: an export skips it.
+    mirror_of: str | None = None
     # Fields the platform declares that have no equivalent in our model.
     # Kept verbatim so the native file can be written back unchanged.
     native: dict[str, object] = field(default_factory=dict)
@@ -95,6 +98,8 @@ def requirement_entry(req: BiosRequirement) -> dict:
         entry["native_system"] = req.native_id
     if req.native_path and req.native_path != req.name:
         entry["native_path"] = req.native_path
+    if req.mirror_of:
+        entry["mirror_of"] = req.mirror_of
     for key in sorted(req.native):
         value = req.native[key]
         if value not in (None, "", [], {}):

scripts/scraper/libretro_scraper.py

@@ -279,9 +279,14 @@ class Scraper(BaseScraper):
                 "docs": "https://docs.libretro.com/library/xrick/",
             },
         }
+        # Every entry below is ours, not System.dat's: marked curated so an
+        # export of System.dat does not take it for libretro's.
         for sys_id, sys_data in EXTRA_SYSTEMS.items():
             if sys_id not in systems:
-                systems[sys_id] = sys_data
+                systems[sys_id] = {
+                    **sys_data,
+                    "files": [{**f, "curated": True} for f in sys_data["files"]],
+                }
 
         # Arcade BIOS present in the repo but absent from System.dat.
         # FBNeo expects them in system/ or system/fbneo/.
@@ -310,7 +315,7 @@ class Scraper(BaseScraper):
             existing = {f["name"] for f in systems["arcade"].get("files", [])}
             for ef in EXTRA_ARCADE_FILES:
                 if ef["name"] not in existing:
-                    systems["arcade"]["files"].append(ef)
+                    systems["arcade"]["files"].append({**ef, "curated": True})
 
         # segasp.zip for Sega System SP (Flycast)
         if "sega-dreamcast-arcade" in systems:
@@ -323,6 +328,7 @@ class Scraper(BaseScraper):
                         "name": "segasp.zip",
                         "destination": "dc/segasp.zip",
                         "required": True,
+                        "curated": True,
                     }
                 )
 
@@ -442,7 +448,7 @@ class Scraper(BaseScraper):
                 for ef in extra_files:
                     key = (ef["name"], ef.get("destination", ef["name"]))
                     if key not in existing:
-                        systems[sys_id]["files"].append(ef)
+                        systems[sys_id]["files"].append({**ef, "curated": True})
                         existing.add(key)
 
         # ep128emu shared group for Enterprise

scripts/scraper/romm_scraper.py

@@ -175,6 +175,7 @@ class Scraper(BaseScraper):
                         destination=f"{slug}/{filename}",
                         required=True,
                         native_id=slug,
+                        mirror_of=None if slug == igdb_slug else igdb_slug,
                     )
                 )