Fix resultBuilt(dirty mechanism) in hls-graph #4238
Conversation
| deps <- readIORef deps | ||
| let changed = if runChanged == ChangedRecomputeDiff then built else maybe built resultChanged result | ||
| built' = if runChanged /= ChangedNothing then built else changed |
There was a problem hiding this comment.
set the new build time to changed might update the build time to a older one
| @@ -187,7 +187,6 @@ instance NFData RunMode where rnf x = x `seq` () | |||
| -- | How the output of a rule has changed. | |||
| data RunChanged | |||
| = ChangedNothing -- ^ Nothing has changed. | |||
| | ChangedStore -- ^ The stored value has changed, but in a way that should be considered identical (used rarely). | |||
There was a problem hiding this comment.
I do not think it is ever used or would be used, but we could add back if there is a need for this.
|
Do we have any tests for this mechanism? |
|
this seems to make sense. how did you discover this? |
|
Actually, I've discovered the Yep, I'll see if some test could be added. |
…dirty-mechanism-of-hls-graph' into 4237-clarify-the-dirty-mechanism-of-hls-graph
|
I've added two test, one from the lower level and one from higher level to demonstrate the bug and ensure no regression. |
|
I'd like @wz1000 to check the tests make sense just to be sure! |
|
nag @wz1000 |
soulomoon
added
the
performance
|
Let's merge it if no one object to this. I've been using this branch and no regression yet. And the bench reuslt look pretty promising |
Fix #4237
resultBuiltmight updated to an older value, which make a rule result that should be clean back to dirty.It might trigger un-needed recomputation. Fixing it could minimize the recomputation further.