Optimize away profunctor classes when profiling (#324) #325
Conversation
| -- marked INLINE for GHC to optimize away profunctor classes when profiling. | ||
| -- See https://github.com/well-typed/optics/issues/324 for more details. | ||
| getOptic (Optic o) = o | ||
| {-# INLINE getOptic #-} |
There was a problem hiding this comment.
doesn't {-# INLINE getOptic #-} just work when getOptic (selector) function is defined with record syntax? (That's not clear from comment, and maybe worth GHC issue & link).
There was a problem hiding this comment.
I think the inline should not be necessary, since newtypes are advertised as "zero cost" wrappers. So one would expect those to be automatically inline. Afaik the newtypes are compiled via coercions. If this gets in the way of optimizations, this feels even more like something GHC could improve.
There was a problem hiding this comment.
doesn't {-# INLINE getOptic #-} just work when getOptic (selector) function is defined with record syntax?
It doesn't seem possible to do that then.
I think the inline should not be necessary, since newtypes are advertised as "zero cost" wrappers.
Yes, it's incredibly bizarre. However, removing the INLINE pragma doesn't improve the situation.
I think it might be connected to this documentation bit:
-fprof-auto
All bindings not marked INLINE, whether exported or not, top level or nested, will be given automatic SCC annotations. Functions marked INLINE must be given a cost centre manually.
|
The newtype unwrap function is the issue? 👀 I will check if this works for me! |
|
I'll merge after the weekend unless anyone objects. |
Fixes #324.Almost.