design: avoid creating a name:value mapping each time we evaluate expression

[zyang-tudb]

Currently we have to create a name:value mapping when evaluating expression against each row. Then every time we get a certain column value we have to look up this map. This is inefficient and unnecessary:

TuDB/lynx/src/main/scala/org/grapheco/lynx/physical.scala

Line 277 in a4a3392

eval(expr)(ec.withVars(df.schema.map(_._1).zip(record).toMap)) match {

[zyang-tudb]

@LianxinGao @xuanying2020

[LianxinGao]

Currently we have to create a name:value mapping when evaluating expression against each row. Then every time we get a certain column value we have to look up this map. This is inefficient and unnecessary:

TuDB/lynx/src/main/scala/org/grapheco/lynx/physical.scala

Line 277 in a4a3392

eval(expr)(ec.withVars(df.schema.map(_._1).zip(record).toMap)) match {

Yes, in pr #227 , I noticed this problem. It is a key function, fix it first?

[zyang-tudb]

It's not a easy problem if you think about it. We might need to revamp the way we evaluate expression.