Github user DaveBirdsall commented on a diff in the pull request:
https://github.com/apache/trafodion/pull/1748#discussion_r236024891
--- Diff: core/sqf/src/tm/tm.cpp ---
@@ -2789,7 +2789,7 @@ void tm_process_msg(BMS_SRE *pp_sre)
{
short lv_ret;
char la_send_buffer[4096];
- char la_recv_buffer[sizeof(Tm_Req_Msg_Type)];
+ char la_recv_buffer[pp_sre->sre_reqDataSize];
--- End diff --
It is not obvious to me why this change solves the problem. For example, at
line 2808 below, we check pp_src->src_reqDataSize to see if it is too big to
fit in la_recv_buffer, and if so, we dynamically allocate a buffer
la_recv_buffer_ddl for it. It looks like the remaining code in the method makes
assumptions that certain message types are always shorter than
sizeof(Tm_Req_Msg_Type) though; but then the logic would read the message into
the wrong buffer and it would fail in some other way. Could you provide an
explanation of why it solves the problem?
Also, I'm wondering if we could get rid of dynamically allocating
la_recv_buffer_ddl and just use la_recv_buffer, by using dynamic array sizing
as you have done here. That would simplify the logic and reduce heap pressure.
@zcorrea, what do you think?
---
