Thread

  1. Re: Fix an unnecessary cast calling elog in ExecHashJoinImpl

    Tender Wang <tndrwang@gmail.com> — 2025-10-17T04:20:37Z

    Xuneng Zhou <xunengzhou@gmail.com> 于2025年10月17日周五 10:31写道:
    
    > Hi Tender,
    >
    > On Thu, Oct 16, 2025 at 8:24 PM Tender Wang <tndrwang@gmail.com> wrote:
    > >
    > >
    > >
    > > Richard Guo <guofenglinux@gmail.com> 于2025年10月16日周四 17:53写道:
    > >>
    > >> On Thu, Oct 16, 2025 at 4:07 PM Xuneng Zhou <xunengzhou@gmail.com>
    > wrote:
    > >> > On Mon, Sep 1, 2025 at 9:26 AM Chao Li <li.evan.chao@gmail.com>
    > wrote:
    > >> > > On Aug 30, 2025, at 14:09, Tender Wang <tndrwang@gmail.com> wrote:
    > >> > > While debugging the HashJoin codes, I noticed below codes in
    > ExecHashJoinImpl():
    > >> > >
    > >> > > elog(ERROR, "unrecognized hashjoin state: %d",
    > >> > > (int) node->hj_JoinState);
    > >> > >
    > >> > > The type of hj_JoinState is already int, so the cast seems
    > unnecessary.
    > >> > > So I remove it in the attached patch.
    > >>
    > >> > > Yes, hj_JoinState is of type int, so the type cast to int is not
    > needed. The change looks good to me.
    > >>
    > >> > LGTM.
    > >>
    > >> I think we can remove this cast, although it's so trivial that it
    > >> doesn't seem to have any real impact.  A similar cast also exists for
    > >> mj_JoinState in ExecMergeJoin().
    > >>
    > >> (These values represent the state of the Join state machine for
    > >> HashJoin and MergeJoin respectively, so I kind of wonder if it might
    > >> be better to define them as enum rather than using int.)
    > >
    > >
    > > Make sense.
    > >
    > > Please check the  v2 patch.
    >
    > If we decide to converts HashJoin and MergeJoin state machine
    > definitions from #define macros to enum types, we might need to keep
    > the (int) casts from the elog() error messages:
    >
    > elog(ERROR, "unrecognized hashjoin state: %d", (int) node->hj_JoinState);
    > elog(ERROR, "unrecognized mergejoin state: %d", (int) node->mj_JoinState);
    >
    
    I searched the entire codebase and found that the accumStats() function is
    in pgbench.c,
    we have
    pg_fatal("unexpected error status: %d", estatus);
    and the estatus is enum type.
    So we can no need to use cast, too.
    
    
    > The enum comment has inconsistent indentation:
    >
    > +  /*
    > + * States of the ExecMergeJoin state machine
    > + */
    >
    
    Yeah, copy-paste issue. Fixed.
    
    
    -- 
    Thanks,
    Tender Wang