[
https://issues.apache.org/jira/browse/TINKERPOP-3275?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18097997#comment-18097997
]
ASF GitHub Bot commented on TINKERPOP-3275:
-------------------------------------------
GumpacG opened a new pull request, #3539:
URL: https://github.com/apache/tinkerpop/pull/3539
## Summary
`gremlin-python`'s GraphBinary `BigIntIO` writer computed the byte length
from the
value's magnitude bit-length but encoded the bytes as signed
two's-complement. Negative
boundary values (e.g. `-129`, `-255`) raised `OverflowError`, and
positive/zero values
produced non-canonical bytes that diverge from the Java reference
(`BigInteger.toByteArray()`).
The fix computes the minimal signed two's-complement length, unifying the
positive, negative,
and zero paths. `BigDecimal` is fixed transitively (its unscaled value is
serialized as a
`BigInteger`).
## Before vs After
| Value | Before | After |
|---|---|---|
| `BigInteger(-129)` | `OverflowError` | `00 00 00 02 ff 7f` |
| `BigDecimal(-2.55)` (unscaled `-255`) | `OverflowError` | `00 00 00 02 ff
01` |
| `BigInteger(127)` | `00 00 00 02 00 7f` (non-minimal) | `00 00 00 01 7f` |
| `BigInteger(0)` | `00 00 00 00` (Java rejects) | `00 00 00 01 00` |
## Changes
- `gremlin-python/.../structure/io/graphbinaryV1.py` - rewrote
`BigIntIO.write_bigint`.
- `gremlin-python/.../tests/unit/io/test_graphbinaryV1.py` - added boundary
round-trip tests
and exact wire-byte assertions.
Assisted-by: Kiro:claude-opus-4.8
> gremlin-python graphbinary BigInt serializer incorrect for negative boundary
> values
> -----------------------------------------------------------------------------------
>
> Key: TINKERPOP-3275
> URL: https://issues.apache.org/jira/browse/TINKERPOP-3275
> Project: TinkerPop
> Issue Type: Bug
> Components: python
> Affects Versions: 3.7.6, 3.8.1
> Reporter: Ken Hu
> Priority: Major
>
> Caveat: I haven't fully verified this, just something AI told me when I was
> updating the "model" tests for gremlin-python. Just putting here so it can be
> checked later.
> gremlin-python can't serialize BigInt signed boundary values like -129.
> {code:java}
> The failure path is:
> GraphBinaryWriter.write_object()
> BigIntIO.dictify()
> BigIntIO.write_bigint()
> obj.to_bytes(length, byteorder='big', signed=True)
> {code}
> For -129, Python reports:
> (-129).bit_length() == 8
> The current writer computes:
> length = (obj.bit_length() + 7) // 8 # 1
> But one signed byte only represents -128 through 127, so -129 needs two bytes
> in two’s complement form. The serializer is using magnitude bit length as if
> it were signed encoded length.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
