impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tim Armstrong (Code Review)" <>
Subject [Impala-ASF-CR] IMPALA-4809: Add Stability for codegen constants
Date Wed, 01 Feb 2017 02:19:30 GMT
Tim Armstrong has posted comments on this change.

Change subject: IMPALA-4809: Add Stability for codegen constants

Patch Set 1:

Commit Message:

PS1, Line 16: overloaded
> From the codegen perspective, we only care if a value is compilation time c
I think this would be useful once we want to do codegen caching and need to reason about whether
codegen output is re-usable.

Agree we can solve IMPALA-4809 with Expr::GetConstant() or the mechanism in this patch (if
that ever makes it in):
File be/src/exprs/expr.h:

Line 266:     RUNTIME_CONSTANT, // Does not change over an execution
Concretely, does this mean that it doesn't change within the scope of a process - it can be
different if you start another daemon or are on a different system?

> This seems odd to be in Expr class if an example of this class of constants
I think it would make sense to move this into a thrift enum - see my other comment about doing
the analysis in the FE.

Line 267:     QUERY_CONSTANT, // Does not change over a single query
We may also want to think about how it varies across different Impala daemons. E.g. does QUERY_CONSTANT
mean it's the same on all Impala daemons for that query, or just on the current daemon?

I think that is effectively a different dimension. I don't think we necessarily need to add
more complexity, just that we should document the assumptions.

E.g. one thing we've thought about hypothetically is compiling on one daemon then distributing
it others.

To view, visit
To unsubscribe, visit

Gerrit-MessageType: comment
Gerrit-Change-Id: I0431bcd84b908bcf130a6d618f57f8f7c5498428
Gerrit-PatchSet: 1
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Zach Amsden <>
Gerrit-Reviewer: Dan Hecht <>
Gerrit-Reviewer: Michael Ho
Gerrit-Reviewer: Michael Ho <>
Gerrit-Reviewer: Tim Armstrong <>
Gerrit-HasComments: Yes

View raw message