New module loading by bettio · Pull Request #2229 · atomvm/AtomVM

bettio · 2026-03-23T23:24:09Z

ESP32 | 1498203 | atomvm-esp32.bin | debug, -Og, OLD loader
ESP32 | 1418775 | atomvm-esp32.bin | debug, -Og, NEW loader

ESP32 | 1476281 | atomvm-esp32.bin | perf, -O2, OLD loader
ESP32 | 1402729 | atomvm-esp32.bin | perf, -O2, NEW loader

These changes are made under both the "Apache 2.0" and the "GNU Lesser General
Public License 2.1 or later" license terms (dual license).

SPDX-License-Identifier: Apache-2.0 OR LGPL-2.1-or-later

Using X macros allows us to turn opcodes.h simple definitions into an opcode table with multiple columns. Signed-off-by: Davide Bettio <davide@uninstall.it>

Replace duplicated decode logic in opcodesswitch.h with a new code loader in module.c driven by opcode signatures declared in opcodes.def. Each opcode declares a compact signature string (e.g. "Af", "jtbssd") that describes its argument encoding. The loader iterates the signature and dispatches to the appropriate DECODE_* macro for each character, removing per-opcode inline decode duplication and reducing binary size. Variable-length arguments use a [X] loop syntax that handles extended list tags automatically, eliminating most custom handler functions. The signature alphabet is inspired to OTP's beam_makeops conventions (s, d, f, j, t, I, A, Q, e, b, F, etc.) so signatures are readable by anyone familiar with BEAM internals. Opcodes removed in commit 94f6f0b are marked with X_OPCODE_REMOVED. Signed-off-by: Davide Bettio <davide@uninstall.it>

Code loading is now handled by parse_core_chunk() in module.c using opcode signatures. Remove the IMPL_CODE_LOADER compilation pass from opcodesswitch.h, leaving it as execution-only. - Remove all #ifdef IMPL_CODE_LOADER blocks (decode macros, TRACE, UNUSED, module_add_label, module_insert_line_ref_offset, entry points) - Remove #ifdef IMPL_EXECUTE_LOOP guards (now always true) - Remove redundant USED_BY_TRACE for variables already used in execution code Signed-off-by: Davide Bettio <davide@uninstall.it>

petermm · 2026-03-25T20:18:02Z

Oracle Review of the 3 Commits (opcodes.def X-macro, signature-driven loader, IMPL_CODE_LOADER removal)

High Severity

Unchecked opcode-table indexing (module.c:731-739) — no bounds check on opcodes_signatures[opcode] before indexing; opcode > 185 causes OOB read before the NULL check runs.
No code-buffer end checks (module.c:724-895) — the loader walks pc without a code_end pointer, so truncated/corrupt bytecode can read past the CODE chunk.
Label range not validated (module.c:679-685) — module_add_label() blindly writes mod->labels[index] with no check that index < num_labels.

Medium Severity

FP register off-by-one (module.c:324-335) — accepts freg == MAX_REG which is out of range (valid: 0..MAX_REG-1).
List signature parser has no nesting/missing ] guard (module.c:746-778) — walks to ] without bounds; missing ] in a signature would overrun.

Low Severity

'c'/'f' signature types not enforced — documented semantics differ from actual parser behavior.
Uninitialized label in 'm' trace path — UB when tracing atom-or-label decodes.

What Looks Good

The core refactor is sound; all opcodes spot-checked have correct signatures matching the executor.
X-macro approach is a clear maintainability win (single source of truth).
Special handlers for OP_LABEL, OP_LINE, OP_FMOVE, OP_BS_MATCH are correct.

pguyot · 2026-03-25T21:14:19Z

src/libAtomVM/module.c

+                break;
+            default:
+                fprintf(stderr, "bs_match loader: unknown command %i\n", (int) term_to_atom_index(command));
+                abort();


We want AVM_ABORT() here.

pguyot · 2026-03-25T21:14:25Z

src/libAtomVM/module.c

+        if (opcode_signature == NULL) {
+            fprintf(stderr, "missing opcode: %i\n", (int) opcode);
+            fflush(stdout);
+            abort();


We want AVM_ABORT() here.

pguyot · 2026-03-25T21:14:33Z

src/libAtomVM/module.c

+                default: {
+                    fprintf(stderr, "unknown signature: %c\n", opcode_signature[arg_index]);
+                    fflush(stdout);
+                    abort();


We want AVM_ABORT() here.

pguyot · 2026-03-25T21:15:11Z

src/libAtomVM/module.c

+
+        if (opcode_signature == NULL) {
+            fprintf(stderr, "missing opcode: %i\n", (int) opcode);
+            fflush(stdout);


flush stdout but fprint stderr.

pguyot · 2026-03-25T21:15:26Z

src/libAtomVM/module.c

+
+                default: {
+                    fprintf(stderr, "unknown signature: %c\n", opcode_signature[arg_index]);
+                    fflush(stdout);


flush stdout but fprintf stderr

pguyot · 2026-03-25T21:16:35Z

src/libAtomVM/module.c

+        }                                     \
+    }
+
+#define DECODE_DEST_REGISTER_GC_SAFE(dreg, decode_pc) \


It seems unused now.

pguyot · 2026-03-25T21:18:04Z

src/libAtomVM/module.c

+
+#ifdef ENABLE_TRACE
+
+// About X macro: https://en.wikipedia.org/wiki/X_macro


We don't need to repeat this.

pguyot · 2026-03-25T21:18:08Z

src/libAtomVM/module.c

+    module_insert_line_ref_offset(mod, line_refs, line_ref, offset);
+}
+
+// About X macro: https://en.wikipedia.org/wiki/X_macro


We don't need to repeat this.

pguyot · 2026-03-25T21:19:29Z

src/libAtomVM/module.c

+    const uint8_t *new_encoded = encoded;
+    uint32_t len;
+    DECODE_LITERAL(len, new_encoded);
+    // TODO: check this: actually should be enough: len = *(new_encoded)++ >> 4;


We have a lot of aborts, should we have one here?

pguyot · 2026-03-25T21:20:32Z

src/libAtomVM/module.c


+union opcode_arg
+{
+    uint32_t u32_arg;


Do we need a union for this?

Turn opcodes.h into opcodes.def

3478d97

Using X macros allows us to turn opcodes.h simple definitions into an opcode table with multiple columns. Signed-off-by: Davide Bettio <davide@uninstall.it>

bettio force-pushed the new-module-loading branch 2 times, most recently from 146cf6d to 1f51a95 Compare March 25, 2026 13:38

bettio added 2 commits March 25, 2026 19:04

bettio force-pushed the new-module-loading branch from 1f51a95 to 7ce8135 Compare March 25, 2026 18:07

bettio changed the title ~~WIP: New module loading~~ New module loading Mar 25, 2026

bettio marked this pull request as ready for review March 25, 2026 18:33

pguyot reviewed Mar 25, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

New module loading#2229

New module loading#2229
bettio wants to merge 3 commits intoatomvm:release-0.7from
bettio:new-module-loading

bettio commented Mar 23, 2026 •

edited

Loading

Uh oh!

petermm commented Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

pguyot Mar 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants


		#ifdef ENABLE_TRACE

		// About X macro: https://en.wikipedia.org/wiki/X_macro

Conversation

bettio commented Mar 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

petermm commented Mar 25, 2026

Oracle Review of the 3 Commits (opcodes.def X-macro, signature-driven loader, IMPL_CODE_LOADER removal)

High Severity

Medium Severity

Low Severity

What Looks Good

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

bettio commented Mar 23, 2026 •

edited

Loading