fix: count data trees in the free space check

install.ps1

@@ -16,7 +16,7 @@
     # and Invoke-WebRequest throws on a cut transfer before the hash is read:
     # the site's build rewrites this line there (scripts/sitebootstrap.py).
     $mismatchAdvice = "The download was most likely cut short: try again. Within minutes of an update of this project the two files can come from caches of different ages. Waiting five minutes settles that."
-    $defaultInstallSha256 = "e6f1825bd04698851d98e30ad9ff016d3fbf28ea5db2d65e5d7e25820026cd4a"
+    $defaultInstallSha256 = "7940e0785cce9f2bc1665c691ec0d5a122d43e01d60113e6220364d1c0b256ba"
     $maximumInstallerBytes = 2MB
     $installerExit = 0
     $installer = if ($PSScriptRoot) { Join-Path $PSScriptRoot "install.py" } else { $null }

install.py

@@ -7377,6 +7377,22 @@ def data_directory_state(entry: dict, bios_path: Path,
     return root, needed, present
 
 
+def data_space_bytes(data_states: list[tuple[dict, tuple[Path, dict[str, dict], int]]]) -> int:
+    """What the data trees take on the drive while they are rebuilt: every
+    file they lack, plus the largest source archive, which sits in its tree
+    until that tree is read. Only the BIOS files were held against the free
+    space: 18.8 KB passed on a card with 20 MB left, and 81.8 MB followed.
+    Nothing is written for them offline without a pack."""
+    if LOCAL.pack is None and _offline_only():
+        return 0
+    files = sum(meta.get("size", 0) for _entry, (_root, needed, _present) in data_states
+                for meta in needed.values())
+    if LOCAL.pack is not None:
+        return files
+    return files + max((data_fetch_bytes(entry, state[1]) for entry, state in data_states
+                        if state[1]), default=0)
+
+
 def data_fetch_bytes(entry: dict, needed: dict[str, dict]) -> int:
     """What getting a tree's missing files costs: its source archive, which
     comes whole for one file, or from a pack the files themselves. A list
@@ -15997,6 +16013,7 @@ def _install(args: argparse.Namespace, real_stdout: TextIO) -> None:
                 print(line)
             continue
 
+        space_checked = False
         if to_download:
             fetched = [f for f in to_download if id(f) not in copies]
             if _offline_only() and fetched:
@@ -16093,9 +16110,11 @@ def _install(args: argparse.Namespace, real_stdout: TextIO) -> None:
             print("  Press Ctrl+C to stop. Running the same command again "
                   "carries on from here.")
             # Checked before the folder is created: a run stopped for space
-            # left an empty BIOS folder behind.
-            _check_free_space(bios_path, dl_size, plat_name, 1 if total_downloaded else 2,
-                              on_its_card(plat_name, bios_path))
+            # left an empty BIOS folder behind. The data trees count too,
+            # written after the files on the same drive.
+            _check_free_space(bios_path, dl_size + data_space_bytes(data_states), plat_name,
+                              1 if total_downloaded else 2, on_its_card(plat_name, bios_path))
+            space_checked = True
             created = not bios_path.exists()
             try:
                 bios_path.mkdir(parents=True, exist_ok=True)
@@ -16173,6 +16192,12 @@ def _install(args: argparse.Namespace, real_stdout: TextIO) -> None:
                     total_errors += 1
                 else:
                     writable.append((entry, state))
+            if writable and not space_checked:
+                # With every BIOS file in place, the trees alone: checked
+                # before the first archive is written.
+                _check_free_space(bios_path, data_space_bytes(writable), plat_name,
+                                  1 if total_downloaded else 2,
+                                  on_its_card(plat_name, bios_path))
             if writable:
                 print("\nData directories:")
             for entry, state in writable:

install.sh

@@ -23,7 +23,7 @@ DEFAULT_INSTALL_URL="https://raw.githubusercontent.com/Abdess/retrobios/main/ins
 # curl or wget stop on a cut transfer before the hash is read: the site's
 # build rewrites this line there (scripts/sitebootstrap.py).
 MISMATCH_ADVICE="The download was most likely cut short: try again. Within minutes of an update of this project the two files can come from caches of different ages. Waiting five minutes settles that."
-DEFAULT_INSTALL_SHA256="e6f1825bd04698851d98e30ad9ff016d3fbf28ea5db2d65e5d7e25820026cd4a"
+DEFAULT_INSTALL_SHA256="7940e0785cce9f2bc1665c691ec0d5a122d43e01d60113e6220364d1c0b256ba"
 MAX_INSTALLER_BYTES=2097152
 
 cleanup() {

install/launchers/index.json

@@ -27,8 +27,8 @@
         {
           "dest": "roms/ports/retrobios/install.py",
           "repo_path": "install.py",
-          "size": 1118187,
-          "sha256": "e6f1825bd04698851d98e30ad9ff016d3fbf28ea5db2d65e5d7e25820026cd4a",
+          "size": 1119606,
+          "sha256": "7940e0785cce9f2bc1665c691ec0d5a122d43e01d60113e6220364d1c0b256ba",
           "executable": false
         },
         {
@@ -71,8 +71,8 @@
     "muos": {
       "archive": {
         "dest": "ARCHIVE/RetroBIOS.muxapp",
-        "size": 1171814,
-        "sha256": "848576a731e09308fb897e760acf62c566523b881b1eca4f2818bacc259747ad"
+        "size": 1173233,
+        "sha256": "43ff8a0cfeefb5b044aee1d31ebbbfcd75ba679f9f67bf96603e706202fb1397"
       },
       "members": [
         {
@@ -92,8 +92,8 @@
         {
           "name": "RetroBIOS/install.py",
           "repo_path": "install.py",
-          "size": 1118187,
-          "sha256": "e6f1825bd04698851d98e30ad9ff016d3fbf28ea5db2d65e5d7e25820026cd4a",
+          "size": 1119606,
+          "sha256": "7940e0785cce9f2bc1665c691ec0d5a122d43e01d60113e6220364d1c0b256ba",
           "executable": false
         },
         {
@@ -166,8 +166,8 @@
         {
           "dest": ".config/modules/retrobios/install.py",
           "repo_path": "install.py",
-          "size": 1118187,
-          "sha256": "e6f1825bd04698851d98e30ad9ff016d3fbf28ea5db2d65e5d7e25820026cd4a",
+          "size": 1119606,
+          "sha256": "7940e0785cce9f2bc1665c691ec0d5a122d43e01d60113e6220364d1c0b256ba",
           "executable": false
         },
         {

tests/test_install.py

@@ -411,6 +411,84 @@ class TestFreeSpaceGuard(unittest.TestCase):
         self.assertLess(check, source.index("bios_path.mkdir(parents=True, exist_ok=True)"))
 
 
+class TheDataTreesCountInTheFreeSpace(unittest.TestCase):
+    """The space check held only the BIOS files: 18.8 KB passed on a card
+    with 20 MB free, then 55.4 MB of archives came and 81.8 MB were written,
+    against the page's promise to refuse before writing anything."""
+
+    def _run(self, *frees_after_bios: int) -> list[tuple[int, str, list]]:
+        """One run per free size, in one folder: the second finds the BIOS
+        files the first installed."""
+        from types import SimpleNamespace
+
+        from test_install_selection import FakeRepository  # noqa: PLC0415
+
+        member = os.urandom(4096)
+        results = []
+        with FakeRepository() as repo, unittest.mock.patch.dict(os.environ):
+            listed = json.loads((repo.root / "install" / "retroarch.json").read_text())
+            listed["data_directories"] = [{
+                "ref": "tree", "dest": "tree", "type": "zip", "archive_size": 50_000_000,
+                "url": "https://example.invalid/tree.zip", "systems": ["nintendo-gba"],
+                "files": {"Sys/a.bin": {"sha1": hashlib.sha1(member).hexdigest(),
+                                        "size": len(member)}}}]
+            (repo.root / "install" / "retroarch.json").write_text(json.dumps(listed))
+            bios = sum(len(payload) for payload, *_ in FakeRepository.PAYLOADS.values())
+            base = repo.env()["RETROBIOS_BASE_URL"]
+            for after in frees_after_bios:
+                out = io.StringIO()
+                fetched: list = []
+                free = bios + after
+                with unittest.mock.patch.object(
+                        sys, "argv", ["install.py", "--platform", "retroarch", "--dest",
+                                      str(repo.dest), "--no-input"]), \
+                        unittest.mock.patch.object(install, "MANIFEST_URL",
+                                                   base + "/install/{platform}.json"), \
+                        unittest.mock.patch.object(install, "RAW_FILE_URL", base + "/{path}"), \
+                        unittest.mock.patch.object(install.shutil, "disk_usage",
+                                                   lambda _p, free=free: SimpleNamespace(free=free)), \
+                        unittest.mock.patch.object(
+                            install, "install_data_directory",
+                            lambda entry, *a, fetched=fetched, **k: fetched.append(entry["ref"]) or {
+                                "present": 0, "installed": 1, "changed": 0, "absent": 0,
+                                "error": "", "not_kept": []}), \
+                        contextlib.redirect_stdout(out), contextlib.redirect_stderr(out):
+                    try:
+                        install.main()
+                        code = 0
+                    except SystemExit as exc:
+                        code = exc.code
+                results.append((code, out.getvalue(), repo.installed() + fetched))
+        return results
+
+    def test_a_card_short_for_the_trees_is_refused_before_any_write(self):
+        ((code, said, written),) = self._run(20_000_000)
+        self.assertEqual(code, 2, said)
+        self.assertIn("Not enough space", said)
+        self.assertEqual(written, [])
+
+    def test_room_for_the_trees_goes_on(self):
+        ((code, said, written),) = self._run(60_000_000)
+        self.assertEqual(code, 0, said)
+        self.assertIn("tree", written)
+
+    def test_the_trees_alone_are_checked_too(self):
+        """Every BIOS file in place: the trees were written unchecked."""
+        _first, (code, said, written) = self._run(60_000_000, -1)
+        self.assertEqual(code, 2, said)
+        self.assertIn("Not enough space", said)
+        self.assertNotIn("tree", written)
+
+    def test_the_bytes_counted(self):
+        state = (Path("x"), {"a": {"size": 10}, "b": {"size": 5}}, 0)
+        entry = {"archive_size": 1000}
+        with unittest.mock.patch.object(install.LOCAL, "pack", None), \
+                unittest.mock.patch.object(install, "_offline_only", return_value=False):
+            self.assertEqual(install.data_space_bytes([(entry, state), (entry, state)]), 1030)
+        with unittest.mock.patch.object(install.LOCAL, "pack", object()):
+            self.assertEqual(install.data_space_bytes([(entry, state)]), 15)
+
+
 class AMissingDestIsSaid(unittest.TestCase):
     """--dest ~/.config/retroarh/system (a typo) was created with every
     missing parent and reported done, and --check on it said only

wiki/installer.md

@@ -445,7 +445,8 @@ The figures change with each release.
 The size depends on the platform and changes with each release. The run prints
 it before downloading, and asks first when the download reaches 1 GB. The
 installer also measures the free space and refuses before writing anything if
-there is not enough.
+there is not enough. The count holds the BIOS files, every data file still to
+come, and the largest source archive, which sits in its tree while it is read.
 
 `curl` or `wget`, and Python 3.8 or newer, are checked by the command itself:
 it stops with a message naming what is absent. PowerShell's execution policy