Backports for 1.10.1 #52755
Backports for 1.10.1 #52755
Conversation
|
Looks like this branch isn't rebased against the latest |
Since the values stored in the parent corresponding to the structural zeros of a tridiagonal matrix aren't well-defined, using it in `ldiv!` is a footgun that may lead to heisenbugs (one seen in https://buildkite.com/julialang/julia-master/builds/31285#018c7cc7-6c77-41ac-a01b-1c7d14cb1b15). This PR changes it to using the tridiagonal matrix directly in `ldiv!`, which should lead to predictable results, and be bug-free. The failing tests for #52571 pass locally with this change. (cherry picked from commit ef549ae)
Make these consistent with what's shown by `julia --help`. Fixes JuliaLang/www.julialang.org#1997. (cherry picked from commit e7e8b89)
Fixes #52404 @nilshg I opted to make it one line as it fits. ``` julia> versioninfo() Julia Version 1.11.0-DEV.1011 Commit bb7091c* (2023-12-04 14:58 UTC) Platform Info: OS: macOS (arm64-apple-darwin23.0.0) CPU: 10 × Apple M2 Pro WORD_SIZE: 64 LLVM: libLLVM-15.0.7 (ORCJIT, apple-m1) Threads: 1 default, 0 interactive, 1 GC (on 6 virtual cores) Environment: JULIA_EDITOR = code ``` (cherry picked from commit e96c13a)
KristofferC
force-pushed
the
backports-release-1.10
branch
from
582a228
to
5e2b620
Compare
|
Thanks, should be fixed. |
This comment was marked as off-topic.
This comment was marked as off-topic.
Fix #52613 by making the TypeVar special case in inference check for vararg first. There's nothing the special case can really do with vararg anyway, so fall back to the ordinary abstract call handling.
aviatesk
force-pushed
the
backports-release-1.10
branch
from
5e2b620
to
6eb535d
Compare
Fixes #52686 Fixes timholy/Revise.jl#770 --------- Co-authored-by: Jameson Nash <vtjnash@gmail.com> Co-authored-by: Kristoffer Carlsson <kcarlsson89@gmail.com> Co-authored-by: Shuhei Kadowaki <40514306+aviatesk@users.noreply.github.com>
|
Also backport [OpenBLAS] Ugrade to v0.3.26 #52762 (it fixes a bug, and e.g. better for Apple M1, can be 2.8x faster)?! [Also maybe backport [SuiteSparse_jll] Update to v7.4.0 from 7.2.1? I'm not sure it's an upgradable stdlib by now, or OpenBLAS, is it standard dependence to upgrade these dependencies or only if not upgradable?]
I suppose the bar is fixing a bug and the only I see are in v7.2.2:
And I'm not sure it fixes a bug (that applies), this in 7.4:
but of interest:
|
IMO, no, this isn't backportable. The JLL for 7.4.0 was built with a Julia 1.11 compat entry, so the JLL won't install on 1.10 right now.
We don't build with OpenMP, so I don't see this bugfix as being useful for us.
The SparseArrays.jl wrappers (and underlying SuiteSparse solver wrappers) don't support any of the single-precision parts yet.
This is a build system change, and the SuiteSparse we use is already built against LBT, so you can try BLIS/FLAME by setting the redirects in LBT appropriately in 7.2.1 that we already ship in 1.10. |
The fallback code that was written for #41099 is causing unintended issues with some inlined stack frames (one previous #51405, new #52709), since the main piece, linetable storage and lookup, was removed in #50546. Probably better to strip it all back to how it was previously, until it can all be revisited more fully. Should be backported to 1.10.
…52836) Investigating into #52763, I've found that `AbstractInterpreters` which enables the `aggressive_constprop` option, such as `REPLInterpreter`, might perform const-prop' even if the result of a non-const call is `LimitedAccuracy`. This can lead to an unintended infinite loop with a custom aggressive const-prop' implementation. This commit restricts const-prop' for such cases where the non-const call result is limited to avoid the issue. This fix is conservative, but given that accurate inference is mostly impossible when there are unresolvable cycles (which is indicated by limited result), aggressive const-prop' isn't necessary for such cases, and I don't anticipate this leading to any obvious regression. fix #52763
…ead289a (#52886) Stdlib: Downloads URL: https://github.com/JuliaLang/Downloads.jl.git Stdlib branch: release-1.10 Julia branch: backports-release-1.10 Old commit: f97c72f New commit: ead289a Julia version: 1.10.0 Downloads version: 1.6.0(It's okay that it doesn't match) Bump invoked by: @IanButterworth Powered by: [BumpStdlibs.jl](https://github.com/JuliaLang/BumpStdlibs.jl) Diff: JuliaLang/Downloads.jl@f97c72f...ead289a ``` $ git log --oneline f97c72f..ead289a ead289a Close Multi timers atexit. Add 1.6 CI (#234) 8a614d5 Skip flakey "concurrent requests" tests on windows (#228) 246504e add a small precompile workload (#226) 3ed0f08 Document how to bypass the 20-second timeout (#222) ``` Co-authored-by: Dilum Aluthge <dilum@aluthge.com>
I think when we backed off the new heuristics for 1.10 this got missed in the meanwhile. Should fix #52923
N5N3
force-pushed
the
backports-release-1.10
branch
from
b1c1d24
to
47999ab
Compare
|
Notable failures: REPL NetworkOptions |
…5539d (#53028) Stdlib: Pkg URL: https://github.com/JuliaLang/Pkg.jl.git Stdlib branch: release-1.10 Julia branch: backports-release-1.10 Old commit: 11cf00df7 New commit: 70525539d Julia version: 1.10.0 Pkg version: 1.10.0 Bump invoked by: @IanButterworth Powered by: [BumpStdlibs.jl](https://github.com/JuliaLang/BumpStdlibs.jl) Diff: JuliaLang/Pkg.jl@11cf00d...7052553 ``` $ git log --oneline 11cf00df7..70525539d 70525539d Merge pull request #3770 from JuliaLang/backports-release-1.10 729ebe1e3 Avoid deleting existing artifacts when ignoring hashes. (#3768) ``` Co-authored-by: Dilum Aluthge <dilum@aluthge.com>
(cherry picked from commit fb76136)
KristofferC
force-pushed
the
backports-release-1.10
branch
from
ee45196
to
cfbff10
Compare
|
Is the OpenBLAS 0.3.26 backport possible? We have it configured with a Julia compat of 1.11 in the Yggdrasil build recipe, so I don't think the JLL will be compatible with Julia 1.10 unless we do another change in Yggdrasil first. @giordano thoughts on this? |
|
Dunno, compatibility can be fixed in the registry, that isn't a big deal, I don't mind but I think in the past other devs weren't excited about backporting upgrades of JLLs, although OpenBLAS is a pretty stable package. |
|
This looked like a simple patch update to a JLL. Why is there a julia version dependency on it? |
OpenMathLib/OpenBLAS@v0.3.23...v0.3.26
Because jlls need to specify julia compat like any other package? And we usually put as compat the julia version where we expect that build to be used, but the registry can be fixed if necessary, it isn't a problem. |
Naively, it seems better to put it for the lowest julia version where it can be used. Anyway, I'll drop it for now. |
KristofferC
force-pushed
the
backports-release-1.10
branch
from
82e0057
to
cfbff10
Compare
) https://github.com/JuliaLang/julia/blob/8a69745bdcb06409ab7e4fc84718f34d7d54a7f9/base/lock.jl#L33-L50 this probably also needs a fix, and maybe other places as well (cherry picked from commit 91ec2bb)
There is no HTML entity `⇔` and accordingly all the usual browsers (Chrome, Firefox, Safari) don't render it. My guess is that there is confusion because GitHub supports `⇔` and shows it as ⇔ (and `↔` as ↔). But here is for example what Firefox shows: <img width="853" alt="Screen Shot 2023-11-08 at 14 02 01" src="https://github.com/JuliaLang/julia/assets/241512/af7d9e27-a5c3-46c3-94d9-82ed52653cb4"> This patch could be backported to 1.9 and 1.10. (cherry picked from commit 81c4f8e)
|
@nanosoldier |
|
The package evaluation job you requested has completed - possible new issues were detected. |
|
JuliaPackaging/Yggdrasil#7202 is merged so we now should be able to update openblas. this is pretty important because it fixes issues on M1 chips and fixes the cutoff point for gemv using multiple threads which previously could case a ~100x regression on chips with many cores. |
It is the OpenBLAS@0.3.23+2 package specifically that should be backported directly to this branch (that PR did merge both the 0.3.26 for current master and the update for 0.3.23 for here). |
(cherry picked from commit 4919dd7)
The earlier test relied on the order of the method match list returned by `_methods_by_ftype`. However, particularly in cases where the `ambig=true` option is used, the match list isn't deterministic. This commit modifies it, similar to other pieces of code, so that it `any` to ensure the code does not depend on the implementation details of `ml_matches`. Fixes #52739.
(cherry picked from commit 919c390)
KristofferC
force-pushed
the
backports-release-1.10
branch
from
3f5f846
to
c644e89
Compare
|
This branch does not compile on Cygwin on Windows without #51698 |
This was inefficient, though not wrong. Fixes #53020
KristofferC
force-pushed
the
backports-release-1.10
branch
from
1fbca96
to
6407c33
Compare
This avoid that these files are picked up during julia's build process, and are instead only used to link pkgimages, as intended. Co-authored-by: Tim Besard <tim.besard@gmail.com> (cherry picked from commit 4ef353c)
|
@nanosoldier |
|
The package evaluation job you requested has completed - possible new issues were detected. |
Backported PRs:
cwstringfor Windows env variables #52758jl_types_egalinequiv_field_types#52748QuoteNodewrapping for captured variables #52878Typeunwrapping ininstanceof_tfunc#52228widen_diagonalbug for nestedUnionAll#52924joint_effectscalculation #53076⇔by↔in documentation #52078Non-merged PRs with backport label: