An N value with an exponent near the i32 limits panics the number validator in debug builds and gets the wrong error in release builds
维护者通常 4 天内回复
评估
这个 Issue 还没有评估数据。
描述
Summary
validate_and_normalize_number in crates/core/src/validation/number.rs does its exponent arithmetic in unchecked i32. An N value whose exponent is near the i32 limits, such as 10E2147483647 or 1.5E-2147483648, overflows it. In a build with overflow checks on (the default for debug builds and cargo test) the request's worker thread panics and the client's connection closes with no response. Release builds keep Cargo's default of overflow checks off (the workspace Cargo.toml has no [profile.release] section), so the arithmetic wraps, and 10E2147483647 is refused as Number underflow, where Amazon DynamoDB answers Number overflow.
Measured (DynamoDB in us-east-1, 2026-10-27 08:42 to 08:44 UTC, each request sent twice, both rounds the same)
PutItem with a non-key attribute {"N": <value>}:
| Value | DynamoDB | ExtendDB main abf8082f, debug build | ExtendDB 0.1.14 dev image, release-slim profile (same number.rs as main) |
|---|---|---|---|
10E2147483647 |
ValidationException 1 validation error detected: Number overflow. Attempting to store a number with magnitude larger than supported range |
connection closed, panicked at crates/core/src/validation/number.rs:86:15: attempt to add with overflow |
ValidationException Number underflow. Attempting to store a number with magnitude smaller than supported range |
1.5E-2147483648 |
ValidationException 1 validation error detected: Number underflow. Attempting to store a number with magnitude smaller than supported range |
connection closed, panicked at crates/core/src/validation/number.rs:57:24: attempt to subtract with overflow |
ValidationException Number underflow. ... |
1E2147483647 |
Number overflow. ... |
Number overflow. ... |
Number overflow. ... |
1E99999999999999999999 |
The parameter cannot be converted to a numeric value: 1E99999999999999999999 |
the same | the same |
1E126, 1E-131, 1E125 (controls) |
overflow, underflow, accepted | the same | the same |
The server keeps serving after the panic; only that request is lost.
DynamoDB starts each of these messages with 1 validation error detected: and ExtendDB starts none of them with it, controls included. That difference is not part of this issue.
Cause
The exponent is parsed as i32 (line 26). Three sums can then leave the i32 range:
- line 57,
explicit_exp - frac_part.len() as i32 - line 86,
point_offset + trailing_zeros as i32 - line 90,
sig_trimmed.len() as i32 - 1 + point_offset + trailing_zeros as i32
AttributeValue's Deserialize calls this function on each N value it decodes (crates/core/src/types/attribute_value.rs:107), so other operations that carry an N value reach it too. I measured PutItem only.
Exponent range (PutItem, twice each: DynamoDB 2026-10-27 08:50 to 08:52 UTC, ExtendDB main 09:40 UTC)
DynamoDB reads the exponent as a signed 64-bit integer. ExtendDB main parses it as an i32 and answers The parameter cannot be converted to a numeric value for anything wider.
| Value | DynamoDB | ExtendDB main |
|---|---|---|
1E2147483648, 1E4294967296, 1E9223372036854775807 |
Number overflow. ... |
cannot be converted |
1E-2147483649, 1E-9223372036854775808 |
Number underflow. ... |
cannot be converted |
1E9223372036854775808, 1E-9223372036854775809 |
The parameter cannot be converted to a numeric value: <value> |
the same |
DynamoDB also range checks a zero on its exponent less its fraction digits (0E126 is Number overflow, 0.0E-130 is Number underflow), where ExtendDB's release builds accept every zero whose exponent fits in an i32 as 0 (a debug build panics on one such as 0.0E-2147483648, which the fix accepts as 0). That part depends on #618 and is left to it.
Fix
Parse the exponent as i64 and do the three sums in i128, where none of them can overflow. The order of magnitude then decides between Number overflow and Number underflow; that is DynamoDB's answer for every out-of-range value above except 0E126 and 0.0E-130, which are left to #618. A zero whose exponent does not fit in an i32 is range checked too, since main already refused those.
Found while reviewing #394, whose manifest parser also calls this function on untrusted input.
- 主要语言
- Rust
- 星标
- 517
- 派生
- 60
- 平均合并
- 4 天 10 小时
- 30 天内合并 PR
- 37
环境准备
- 提供 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
ExtendDB/extenddb 的其他 Issue
-
难度 2/5 3-5 天 新手友好度 65/100
维护者通常 4 天内回复
-
documentation
难度 2/5 1-3 小时 新手友好度 84/100
ExtendDB/extenddb#658 · 1 条评论 ·
维护者通常 4 天内回复
-
backend:postgresql bug
难度 2/5 1-3 小时 新手友好度 70/100
维护者通常 4 天内回复
-
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 4 天内回复
-
backend:mongodb bug
难度 2/5 1-3 小时 新手友好度 77/100
维护者通常 4 天内回复
查看 ExtendDB/extenddb 的全部 Issue
相似的 Issue
-
bug
难度 2/5 1-3 小时 新手友好度 84/100
dekobon/big-code-analysis#1729 ·
维护者通常 1 天内回复
-
bug help wanted
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 3 天内回复
-
难度 2/5 1-3 小时 新手友好度 88/100
akitaonrails/ai-memory#1888 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
peteonrails/voxtype#941 ·
维护者通常 4 天内回复
-
bug priority:medium triage:maintainable
难度 2/5 1-3 小时 新手友好度 82/100
ccusage/ccusage#2000 · 3 条评论 ·
维护者通常 1 天内回复