Draft: signals of an unreadable BO_ line join the previous frame

Status: draft, not posted. Whether it goes to ebroecker/canmatrix is the maintainer's decision. Everything from the issue title down to Notes for CanViewer is the text to post; the notes are not.

The reproduction and the Actual block are checked by app/src/lib/canmatrix/upstream-reports.test.ts, which runs the script under the canmatrix wheel CanViewer ships and compares its output with the block below.

---

Issue title: DBC import: when a BO_ line cannot be read, its signals are added to the previous frame

Summary

When the DBC reader cannot read a BO_ line, it prints an error for that line and goes on. The SG_ lines that follow are then added to the frame of the previous BO_ line, without any error or warning. The result is a frame with signals that belong to another frame, often overlapping its own.

The trigger seen most often in real files is a 29-bit identifier written without the extended-frame flag (bit 31), such as BO_ 2048 ... or BO_ 103596083 ...: ArbitrationId raises ArbitrationIdOutOfRange for a standard id above 0x7FF. Any other exception in the BO_ branch has the same effect.

Environment

• canmatrix 1.3.3, the latest release on PyPI (2026-10-21);
• CPython 3.13 on Linux, and Pyodide 314.0.7 (CPython 3.14) with the same wheel: identical output;
• the development branch at 8b54a41 (2026-10-28) prints the same: src/canmatrix/formats/dbc.py has not changed since 1.3.3.

Reproduction

import canmatrix.formats

DBC = b'''VERSION ""

BU_: ECU

BO_ 100 First: 8 ECU
 SG_ A : 0|8@1+ (1,0) [0|255] "" Vector__XXX

BO_ 2048 Second: 8 ECU
 SG_ B : 8|8@1+ (1,0) [0|255] "" Vector__XXX
'''

print("canmatrix", canmatrix.__version__)
db = canmatrix.formats.loads_flat(DBC, "dbc")
for frame in db.frames:
    print(frame.name, frame.arbitration_id.id, [s.name for s in frame.signals])

Expected

First keeps only its own signal A. Signal B belongs to Second: if Second cannot be read, B cannot be placed either, and should be reported (or dropped with an error), not attached to First. For example:

canmatrix 1.3.3
error with line no: 8
b'BO_ 2048 Second: 8 ECU\n'
error with line no: 9
b' SG_ B : 8|8@1+ (1,0) [0|255] "" Vector__XXX\n'
First 100 ['A']

Actual

canmatrix 1.3.3
error with line no: 8
b'BO_ 2048 Second: 8 ECU\n'
First 100 ['A', 'B']

B (bits 8 to 15 of Second) is now a signal of First, and nothing says so.

Cause

In load() in src/canmatrix/formats/dbc.py, frame holds the frame that SG_ lines are added to. The BO_ branch (line 561) assigns it only after the line was read:

frame = Frame(temp.group(2), arbitration_id=int(temp.group(1)),
              size=int(temp.group(3)), transmitters=temp.group(4).split())

When that raises (here in ArbitrationId.__attrs_post_init__, "ID out of range"), the bare except: at line 964 prints the error and the loop goes on with frame still pointing at the previous frame. The SG_ branch then calls frame.add_signal(temp_signal) (line 603) on it.

If the unreadable BO_ line is the first one in the file, frame is still None, add_signal raises, and every SG_ line of that message is reported. So the same input is reported or silently misplaced depending on where the message sits in the file.

Suggested fix

Clear frame before the BO_ line is read, so that the signals of a frame that could not be read fail the way they already do when it is the first:

--- a/src/canmatrix/formats/dbc.py
+++ b/src/canmatrix/formats/dbc.py
@@ -559,6 +559,8 @@
                 follow_up = _FollowUps.NEW_SYMBOLS
                 continue
             if decoded.startswith("BO_ "):
+                # if this line cannot be read, its SG_ lines must fail, not join the previous frame
+                frame = None
                 regexp = re.compile(r"^BO_ ([^\ ]+) ([^\ ]+) *: *([^\ ]+) ([^\ ]+)")
                 temp = regexp.match(decoded)
     #            db.frames.addFrame(Frame(temp.group(1), temp.group(2), temp.group(3), temp.group(4)))

With this patch applied to 1.3.3, the reproduction prints the Expected block above, and tests/test_dbc.py still passes with lxml installed (68 passed, as for unpatched 1.3.3 installed with pip install canmatrix[kcd]==1.3.3 pytest).

Whether such an id should be read at all is a separate question, and readers disagree: cantools 45.1.2 refuses the whole file ("Standard frame id 0x800 is more than 11 bits in message Second."), while some viewers take the id as written. This report is only about the misplaced signals.

---

Notes for CanViewer (not for posting)

• CanViewer reads .dbc with its own parser, which reads both messages of the reproduction (First with A, Second with id 2048 and B); upstream-reports.test.ts pins that. So the app is not affected.
• The bug shows in the mapper suite (app/src/lib/canmatrix/mapper.test.ts), which reads the 58 OEM DBC files through canmatrix as well. Its pinned baseline (mapper-dbc-baseline.json) has 71 BO_ lines canmatrix could not read in 6 files, and 275 signals on the wrong message in 3 of them: toyota_2017_ref_pt.dbc 259, vw_meb.dbc 14, chrysler_cusw.dbc 2 (for example line 182, BO_ 103596083 BSM_LEFT: 8 XXX). In gm_global_a_lowspeed.dbc all 13 BO_ lines are unreadable, so no frame is ever current, and its 27 SG_ lines are reported instead, as described above.
• Before posting: search the open and closed issues for "BO_" and "ArbitrationIdOutOfRange" for duplicates (not done yet). The opendbc files above (MIT) can be named as examples.