adjustments to the latest master #284
Conversation
aviatesk
force-pushed
the
avi/1.12
branch
2 times, most recently
from
55a6d44
to
9883872
Compare
| # This broke some time between 1.10 and 1.11-DEV.10001 | ||
| @test isa(sin′′, Core.OpaqueClosure{Tuple{Float64}, Float64}) broken=VERSION>=v"1.11-" |
There was a problem hiding this comment.
I don't know too much why these broken test cases have been fixed, but I guess it was probably because the previous (::∂☆recurse)(...) implementation has the serious flaw (i.e. the generator uses function that hasn't been defined at the time when the generator function is defined)?
aviatesk
force-pushed
the
avi/1.12
branch
2 times, most recently
from
132f9bb
to
6eda0db
Compare
aviatesk
force-pushed
the
avi/1.12
branch
4 times, most recently
from
6764904
to
44a4308
Compare
aviatesk
changed the title
| if interp !== nothing | ||
| cis.inferred = true | ||
| @static if VERSION ≥ v"1.12.0-DEV.15" | ||
| rettype = Any # ci.rettype # TODO revisit |
There was a problem hiding this comment.
ci no longer has rettype.
There was a problem hiding this comment.
We probably need further tweak on this code, but maybe later.
src/codegen/reverse.jl
Outdated
| else | ||
| ci.inferred = true | ||
| rettype = ci.rettype | ||
| end | ||
| ocm = ccall(:jl_new_opaque_closure_from_code_info, Any, (Any, Any, Any, Any, Any, Cint, Any, Cint, Cint, Any), |
There was a problem hiding this comment.
The signature of this c function changed
There was a problem hiding this comment.
That said, this should ideally use the Core.OpaqueClosure constructor
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #284 +/- ##
==========================================
- Coverage 55.29% 54.83% -0.46%
==========================================
Files 28 27 -1
Lines 2917 2945 +28
==========================================
+ Hits 1613 1615 +2
- Misses 1304 1330 +26 ☔ View full report in Codecov by Sentry. |
|
I will give them reviews quickly. |
|
It's really nice to have tests passing on the stable version! |
| @@ -6,36 +6,34 @@ module stage2_fwd | |||
| @test sin′(1.0) == cos(1.0) | |||
| end | |||
| let sin′′ = Diffractor.dontuse_nth_order_forward_stage2(Tuple{typeof(mysin), Float64}, 2) | |||
| # This broke some time between 1.10 and 1.11-DEV.10001 | |||
| @test isa(sin′′, Core.OpaqueClosure{Tuple{Float64}, Float64}) broken=VERSION>=v"1.11-" | |||
There was a problem hiding this comment.
These test cases about inferred OC types appear to work fine in v1.12, but they're still broken on v1.11. We might need to consider backporting some changes, or tweak the broken condition here.
Many adjustments...