Backports for 1.2-RC1 #31727
Backports for 1.2-RC1 #31727
Conversation
(cherry picked from commit b08f120)
Using a WeakRef meant we might not actually `bind` the result. If nobody was still holding a reference to put contents into the Condition, we would simply garbage collect it, and then never need to close it. Since that does not seem to be the intent, instead move to just keeping a strong reference (alternatively, we would have to switch to using `stream_wait` with ref-counting, but that seems suboptimal for several reasons.). fix #31507 (cherry picked from commit 29f61cd)
* edit 1.2 release notes * Run NEWS-update.jl (cherry picked from commit ccd9574)
(cherry picked from commit bbb8eae)
(cherry picked from commit 20834c3)
(cherry picked from commit 3e6a893)
(cherry picked from commit 50ec39c)
|
Backported the non-feature part of #31489 |
|
@nanosoldier |
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan |
|
Added #31687, merged both commits into one. |
|
Most Pkg regressions fixed. Need fix to #32084 and then we should rerun PkgEval. @nanosoldier |
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan |
|
Looks pretty good overall. The sparse matmul memory regressions maybe deserve investigation, as do the |
|
BigFloat regressions are #31759. |
|
are right, thanks. |
It is possible when marking task objects in the GC for the associated jl_ptls_t reference to be NULL. To determine whether a task is a root task, we therefore also have to check if that reference is valid. (cherry picked from commit 25c33e4)
|
@nanosoldier |
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan |
This restricts the element type of the input vector to be `<:Real` and only converts in case the promoted type is concrete. Fixes #32092. (cherry picked from commit 587cb82)
|
@staticfloat, could you double check my backport for your commit above this comment |
|
I think we should rerun PkgEval now @ararslan |
One minor issue, I have directly pushed the fix; it wasn't critical, but it's better this way. |
|
Newest PkgEval run regressions:
Logs are here: https://gist.github.com/ararslan/fd6edd0d6d3b61babab8cb7fe5b661f0. |
ararslan
removed
the
needs pkgeval
|
Merged and rc1 tagged. |
24 participants
Backported PRs:
freein subtyping #31668 - add a missingfreein subtypingiterate#31762 - fix Regression in iterate on Dicts vs 1.1 #31760, regression in Dictiterateschedule#31812 - fix incorrect error path inschedulefieldtype_tfunc#31848 - fix Fatal error in type inference in StatsModels #31803, covariance handling infieldtype_tfuncjl_code_requires_compiler#31967 - fix Regression with --compile=min for 1.2/master compared to 1.0/1.1 #31965, bug in interpreter stacktraces andjl_code_requires_compilerType{Vec}andType{<:Vec{T}}#31997 - fix ambiguity error with non-applicable methods #31993, intersection ofType{Vec}andType{<:Vec{T}}isafast path for typevars with lower bounds #32040 - fixisafast path for typevars with lower boundsEventfrom Base back to Threads #32043 - move export ofEventfrom Base back to Threadsview(a,:) .+= view(b,:)#31300USE_BINARYBUILDER_xxxwhenUSE_SYSTEM_xxxis set #31725 - Automatically disableUSE_BINARYBUILDER_xxxwhenUSE_SYSTEM_xxxis setNon-merged PRs with backport label: