feat(core): ensure all module names are interned

core/src/all_modules.py.mako

@@ -0,0 +1,67 @@
+# generated from all_modules.py.mako
+# do not edit manually!
+# flake8: noqa
+# fmt: off
+# isort:skip_file
+<%
+from pathlib import Path
+from itertools import chain
+
+THIS = Path(local.filename).resolve()
+SRCDIR = THIS.parent
+
+PATTERNS = (
+    "*.py",
+    "storage/**/*.py",
+    "trezor/**/*.py",
+    "apps/**/*.py",
+)
+
+pyfiles = chain.from_iterable(sorted(SRCDIR.glob(p)) for p in PATTERNS)
+%>\
+from trezor.utils import halt
+
+# this module should not be part of the build, its purpose is only to add missed Qstrings
+halt("Tried to import excluded module.")
+
+# explanation:
+# uPy collects string literals and symbol names from all frozen modules, and converts
+# them to qstrings for certain usages. In particular, it appears that qualified names
+# of modules in sys.modules must be qstrings. However, the collection process is
+# imperfect. If `apps.common.mnemonic` is always imported as `from ..common import mnemonic`,
+# the string "apps.common.mnemonic" never appears in source code, is never collected,
+# but then is generated and interned at runtime.
+# A similar thing happens in reverse: if module `storage.cache` is always imported as
+# this name, then "storage.cache" is collected but neither "storage" nor "cache" alone.
+# Which is a problem, because "cache" is a symbol that is added to `storage`'s dict.
+#
+# We need to avoid run-time interning as much as possible, because it creates
+# uncollectable garbage in the GC arena.
+#
+# Below, every module is listed both as import (which collects the qualified name)
+# and as a symbol (which collects each individual component).
+# In addition, we list the alphabet, because apparently one-character strings are always
+# interned, and some operation somewhere (rendering?) is reading strings character by
+# character.
+
+% for pyfile in pyfiles:
+<%
+pyfile = pyfile.relative_to(SRCDIR)
+if pyfile.name == "__init__.py":
+    import_name = str(pyfile.parent)
+else:
+    import_name = str(pyfile.with_suffix(""))
+import_name = import_name.replace("/", ".")
+%>\
+${import_name}
+import ${import_name}
+% endfor
+
+# generate full alphabet
+<%
+ALPHABET = "abcdefghijklmnopqrstuvwxyz"
+%>\
+% for letter in ALPHABET:
+${letter}
+${letter.upper()}
+% endfor