simplify entry points to type inference #18591
Conversation
vtjnash
force-pushed
the
jn/codeinfo-2
branch
from
34a1c99 to
0e4cf64
Compare
|
@nanosoldier |
| return code.rettype | ||
| end | ||
| else | ||
| return code.rettype |
There was a problem hiding this comment.
the logic flow is a bit tough to follow, add more comments here?
There was a problem hiding this comment.
Maybe easier to read as
isa(inf, CodeInfo) || return code.rettype
if (inf::CodeInfo).inferred
return code.rettype
endThere was a problem hiding this comment.
agreed 3. I'm not sure why Github doesn't hide the comments anymore after the code is changed.
base/inference.jl
Outdated
| # don't infer it now, as attempting to force it now would be a bad idea (non terminating) | ||
| force_infer = false | ||
| if method.module == _topmod(method.module) || (isdefined(Main, :Base) && method.module == Main.Base) | ||
| # however, some gf have special tfunc and meaning they wouldn't have been inferred yet |
There was a problem hiding this comment.
is there a missing word here? "special tfunc and ___ meaning they" ?
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels |
vtjnash
force-pushed
the
jn/codeinfo-2
branch
from
0e4cf64 to
d0b1ac0
Compare
|
@nanosoldier |
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels |
this breaks up the typeinf_edge function into its subcomponents, allowing callers to exercise only the pieces they care about rather than returning all possible values the caller might care about in a Tuple also adds missing test-and-lock-and-test for threaded inference entry points
vtjnash
force-pushed
the
jn/codeinfo-2
branch
from
d0b1ac0 to
03cff8b
Compare
The logic had been switched since #18591.
this breaks up the typeinf_edge function into its subcomponents,
allowing callers to exercise only the pieces they care about
rather than returning all possible values the caller might care
about in a Tuple
also fixes missing test-and-lock-and-test for threaded inference
(backporting from #17057 WIP)