Repository navigation
fix(corekit): BitField packs every bit its namespace does not name as zero (shared root of the per-protocol reserved-bit issues) #1288
Description
Describe the bug
BitField.pre_process starts from an all-zero buffer (pcapkit/corekit/fields/strings.py:232) and only writes named subfields, so any uncovered bit is lost on pack. This is the shared root of #1199, #1207, #1211, #1212, #1221, #1235 and #1246. Counting BitFields in pcapkit/protocols/schema whose namespace leaves bits uncovered gives 70 (mh 28, tcp 14, hip 5, httpv2 4, hopopt 4, ipv6_opts 4, ipv4 3, sctp 3, pcapng 2, ipv6_frag 1, ipv6_route 1, l2tp 1).
To Reproduce
f = BitField(length=1, namespace={"x": (0, 1), "y": (7, 1)})
v = f.post_process(bytes.fromhex("ff"), {}) # {"x": 1, "y": 1}
f.pre_process(v, {}).hex() # "81", not "ff"Expected behavior
Either post_process keeps the uncovered bits under a hidden key that pre_process ORs back, or each site names them as reserved. A fix at the field fixes all the per-protocol reports at once.
Found by the corekit round-trip audit (#1202). The repro was re-checked by an independent verifier, and I re-ran it on main at 11ae68ebf.
Activity
JarryShaw commentedon Oct 7, 2026
Blocked until the per-protocol reserved-bit PRs land (#1226, #1253, #1254 and the IPv6 and HTTP/2 wave PRs). This issue then closes with an invariant test that every BitField namespace covers all of its bits.
JarryShaw commentedon Oct 8, 2026
#1226, #1253 and #1254 have merged. Still blocked until #1235 closes: its remaining HOPOPT/IPv6-Opts reserved-bit sites (MPL, IP_DFF, Quick-Start) have a worker on them now.
JarryShaw commentedon Oct 8, 2026
Unblocked: #1235 is closed (every site already round-trips on
ec2ba13a4). A worker is on the invariant test now.JarryShaw commentedon Oct 8, 2026
#1416 adds the invariant test. Its
KNOWN_GAPSallowlist holds the 27 sites still unfixed: HIP 3 (#1417) and MH 24 (#1418), both now being worked. This issue closes once that allowlist is empty.fix(mh): keep the reserved bits of 24 BitField flag namespaces on reb…
test(schema): empty the BitField KNOWN_GAPS allowlist (#1288)
JarryShaw commentedon Oct 8, 2026
Done: #1416's
KNOWN_GAPSallowlist is now empty onmain(949b0a900, test-only). #1419 (HIP) and #1421 (MH) merged after #1416, so the exact-match assertion was failing on02ee60a71. With the empty allowlist it passes (3 tests OK), and re-adding a single entry makes it fail again.