Backports for 1.11.3

[KristofferC]

Backported PRs:

• subtype: fast path for Type == TypeVar #56640
• Fix test report alignment #56789
• fix exp(weirdNaN) #56784
• Fix generate_precompile statement grouping & avoid defining new func #56317
• Fix partially_inline for unreachable #56787

Need manual backport:

• Fix trampoline assembly for build on clang 18 on apple silicon #54634
• add logic to prefer loading modules that are already loaded #55908
• rearrange jl_delete_thread to be thread-safe #56097
• inference: fix inference error from constructing invalid TypeVar #56264
• InteractiveUtils.jl: fixes issue where subtypes resolves bindings and causes deprecation warnings  #56306
• loading: clean up more concurrency issues #56329
• irinterp: set IR_FLAG_REFINED for narrowed PhiNodes #56391
• Allow taking Matrix slices without an extra allocation #56236
• count bytes allocated through malloc more precisely #55223
• lowering: don't reverse handler order in (pop-handler-list ...) #55871
• precompile: don't waste memory on useless inferred code #56749
• gc: improve mallocarrays locality #56801

Contains multiple commits, manual intervention needed:

• fix FileWatching designs and add workaround for a stat bug on Apple #55877
• Make String(::Memory) copy #54457
• recommend explicit using Foo: Foo, ... in package code (was: "using considered harmful") #42080
• precompile: fix performance issues with IO #56370
• fix jl_mutex_lock deadlock under rr #56644
• codegen: make more judicious use of global rooting #56625

Non-merged PRs with backport label:

• Export opaque closure constructor #56808
• Make Pairs public #56794
• Extend Base.rationalize instead of defining new function Base.MathConstants.rationalize #56793
• Update version that version-named manifests works from #56709
• Fix and test lcm([1//2, 1//2]) == 1//1 #56423
• xref UnionAll in the doc string of where #56411
• REPL: complete imports when there are multiple macros before #56281
• also redirect JL_STDERR etc. when redirecting to devnull #55958
• Make threadcall gc safe #55956
• Check for conflicting @ccallable name before JIT registration #55813
• Set stdlib sources as read-only during installation #55534
• propagate the terminal's displaysize to the IOContext used by the REPL #55499
• Allow for generically extracting unannotated string #55458
• Make AnnotateChar equality consider annotations #55457
• isfile_casesensitive fixes on Windows #55220
• propertynames for SVD respects private argument #55169
• tweak how filtering is done for what packages should be precompiled #53957
• prevent code loading from lookin in the versioned environment when building Julia #51479
• More doctests for Sockets and capitalization fix #50813
• improve docs for @inbounds and Base.@propagate_inbounds #50157

[maleadt]

This also needs to include LinearAlgebra.jl back-ports: https://github.com/JuliaLang/LinearAlgebra.jl/issues?q=state%3Aclosed%20label%3A%22backport%201.11%22
Or should those be made into commits here against release-1.11 (where the stdlib hasn't been moved out)?

[KristofferC]

Or should those be made into commits here against release-1.11 (where the stdlib hasn't been moved out)?

I've been thinking about it and I think the easiest is just to cherry pick these commits onto the julia repo onto either a backports-linearalgebra-1.11 branch or directly to backports-1.11.

[IanButterworth]

@nanosoldier runtests(vs=“release-1.10”)

[maleadt]

@IanButterworth You typo'd the syntax, branches need to be specified using :: https://github.com/JuliaCI/Nanosoldier.jl?tab=readme-ov-file#trigger-syntax. In any case, it shouldn't be required here, as PkgEval uses the merge-base, which should be release-1.10 already.

@nanosoldier runtests()

[IanButterworth]

Ah ok, but this is a 1.11 backports PR. I wanted to check it against 1.10 (because stuff in the blacklist hasn't been tested vs 1.10 because of that bug)

[DilumAluthge]

I don't see the Nanosoldier commit status here, which leads me to believe that either the webhook was never delivered, or that the Nanosoldier server didn't accept the job.

[DilumAluthge]

@nanosoldier runtests()

[nanosoldier]

The package evaluation job you requested has completed - possible new issues were detected.
The full report is available.

[IanButterworth]

@nanosoldier runtests(vs=":release-1.10")

[IanButterworth]

@maleadt please cancel the duplicate run invoked here #56741 (comment) (or let me know how to if I can)

The latest request I did is the correct spec. We want to check against 1.10 state

[maleadt]

please cancel the duplicate run invoked here #56741 (comment) (or let me know how to if I can)

It is currently not possible to cancel Nanosoldier invocations. Only thing I can to is restart the service at a strategic time. I'll check tomorrow when this job starts.