[
https://issues.apache.org/jira/browse/THRIFT-6308?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jens Geyer updated THRIFT-6308:
-------------------------------
Fix Version/s: (was: 0.26.0)
> Align the meaning of a zero/unset wire-length limit across bindings
> -------------------------------------------------------------------
>
> Key: THRIFT-6308
> URL: https://issues.apache.org/jira/browse/THRIFT-6308
> Project: Thrift
> Issue Type: Improvement
> Components: D - Library, JavaME - Library, Lua - Library, Smalltalk
> - Library
> Reporter: Jens Geyer
> Priority: Minor
>
> h2. Observation
> The configurable string/container/frame length limits behave inconsistently
> across bindings:
> * D: a non-positive limit disables the check ({{limit > 0 && size > limit}});
> zero is a documented escape hatch, and zero/negative is the value most likely
> to be passed by accident.
> * Java ME, Smalltalk: setting a non-positive limit raises immediately.
> * Lua: the base check is always in force, with no way to disable it.
> The error raised on an over-limit value also differs (Lua
> {{INVALID_FRAME_SIZE}}; D {{CORRUPTED_DATA}} / {{SIZE_LIMIT}}; Smalltalk
> {{sizeLimit}} / {{negativeSize}}), so a cross-binding caller cannot rely on a
> single code.
> h2. Suggestion
> Settle on one meaning for a zero/unset limit and one error-code family, and
> align the bindings. Separately, Smalltalk exposes a string cap but no
> container-size cap; adding one would match the other bindings (its container
> reads are lazy, one element at a time, so this is consistency rather than a
> pressing gap).
> _Drafted with AI assistance (Claude Opus 4.8); reviewed and filed by Jens
> Geyer._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
