[
https://issues.apache.org/jira/browse/CALCITE-3414?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16952491#comment-16952491
]
Feng Zhu commented on CALCITE-3414:
-----------------------------------
[~danny0405], there are two reasons:
(1)some implementation in Types.castIfNecessary is not correct;
(2)we just need to copy a small part of unique logics, mainly
{color:#FF0000}_returnType instanceof RecordType_{color} and{color:#FF0000}
_Types.isAssignableFrom(returnType, type)_{color}, For other parts,
RexToLixTranslator is more powerful.
> Unify Expression'type cast and conversion as a robust one
> ---------------------------------------------------------
>
> Key: CALCITE-3414
> URL: https://issues.apache.org/jira/browse/CALCITE-3414
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.21.0
> Reporter: Feng Zhu
> Assignee: Feng Zhu
> Priority: Major
> Labels: pull-request-available
> Attachments: RexToLixTranslator.png, TypeConversion.txt, Types.png
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Current now, there are two functions in calcite that can be used to
> cast/convert Expression to a specific Type.
> *_Types.castIfNecessary_* and _*RexToLixTranslator.convert*_.
> We make a deep investigation on their implementations and demonstrate them as
> below.
> {color:#ff0000}
> !RexToLixTranslator.png!{color}
> {color:#ff0000}
> *RexToLixTranslator.convert*{color}
> !Types.png!
> {color:#ff0000}
> *Types.castIfNecessary*{color}
> It can be seen that:
> (1) They have a lot of overlaps;
> (2) *_RexToLixTranslator.cast_* can cover more cases with tools like
> _SqlFunctions_ and etc.
> (3) Both of them have limitations and may generate incorrect code, which is
> listed in attachment(TypeConversion.txt).
> Multiple choices usually bring confusion to developers and resulting to the
> misuse of them.
> For example, CALCITE-3245 exposes that Types.castIfNecessary cannot cast the
> Expression to BigDecimal.class.
> Fixing the issue in *_Types.castIfNecessary_* directly seems to be not a
> good idea.
> On one hand, it is not convenient to call _SqlFunctions_ in linq4j. One the
> other hand, it will brings duplicate with _*RexToLixTranslator.cast*_.
> However, due to some unique logic in _*Types.castIfNecessary*_, we cannot
> replace it as _*RexToLixTranslator.cast*_ neither.
> Therefore, it is a good idea to integrate implementations into
> RexToLixTranslator.cast.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
