| 85f19de | Mark Kittisopikul | 23 June 2023, 18:05:02 UTC | Document JULIA_CPU_TARGET in environment-variables.md (#50147) | 23 June 2023, 18:05:02 UTC |
| c0f623d | Joe(y) Carpinelli | 22 June 2023, 21:34:20 UTC | Small fix to running-external-programs.md (#50220) The `run` method no longer throws an `ErrorException` on failure. It currently throws a `ProcessFailedException`. | 22 June 2023, 21:34:20 UTC |
| d138701 | Valentin Churavy | 22 June 2023, 19:58:57 UTC | Merge pull request #50264 from JuliaLang/kp/fix-full-build Fix `-full` tarball | 22 June 2023, 19:58:57 UTC |
| 7785db7 | pchintalapudi | 22 June 2023, 17:43:50 UTC | Check that we don't name LLVM constants (#50263) | 22 June 2023, 17:43:50 UTC |
| a8d76c6 | N5N3 | 22 June 2023, 16:20:54 UTC | typeintersect: fix `constraintkind` for non-covariant var (#50209) Co-authored-by: Jameson Nash <vtjnash+github@gmail.com> | 22 June 2023, 16:20:54 UTC |
| 07f0525 | Ian Butterworth | 22 June 2023, 16:20:10 UTC | Tweak cache pidlocking pt. 2 (#50254) | 22 June 2023, 16:20:10 UTC |
| ec33194 | Keno Fischer | 22 June 2023, 16:04:24 UTC | Remove type_lift_pass!/OptimizerLattice (#50257) This pass (which has become somehwat misnamed) was inserting error checks, both for undefined slots and undefrefs from sroa'd getfields (at least in theory - this broke at some point - see #50250). It accomplished this partly by using the OptimizerLattice, which adjoins the `MaybeUndef` lattice element to the ordinary incidence lattice. This lattice element indicates that the SSAValue may potentially have come from an undef slot and gives it special semantics inside `:isdefined` and `:undefcheck`. However, in our more recent formalization of lattices, this element is ill-defined. It is not valid to widen it, because doing so would change semantics and cause crashes. It would be possible to have a correct version of this element, but it would require inverting the meaning (i.e. having all types be maybe-undef and using a NotUndef lattice element). However, such a change would be expensive and not worth it. This has been causing me some headaches downstream when trying to use custom lattices and custom pass pipelines, so I had some extra motivation to do something about it. This PR just does away with all this complexity. SSA conversion and SROA now directly insert the requisite `:throw_undef_if_not` checks. This does increase the size of the IR somewhat earlier in the pipeline, but on the other hand it saves a full scan over the IR later in the pipeline, so it's probably a was overall. While we're here, we fix #50250 by properly inserting the requistie phi nest inside SROA. | 22 June 2023, 16:04:24 UTC |
| 330c79d | Lilith Orion Hafner | 22 June 2023, 15:39:15 UTC | Add note to `load_path` docstring stating that it is not safe to mutate its return value (#50233) | 22 June 2023, 15:39:15 UTC |
| ef6d900 | Jameson Nash | 22 June 2023, 14:17:35 UTC | codegen: handle dead code with unsafe_store of FCA pointers (#50164) Fix #50125 | 22 June 2023, 14:17:35 UTC |
| 8af82ea | K Pamnany | 21 June 2023, 18:43:53 UTC | Set `CMAKE_INSTALL_LIBDIR` for non-Windows platforms This makes it so libgit2.so and libssh2.so go into `usr/lib` where they should be, rather than into `usr/lib64` on some platforms. | 22 June 2023, 14:01:32 UTC |
| 49ac54a | K Pamnany | 21 June 2023, 18:06:56 UTC | Fix ITTAPI for offline build of LLVM | 22 June 2023, 14:01:16 UTC |
| ad120f4 | Max Horn | 22 June 2023, 13:45:44 UTC | Fix jl_gc_internal_obj_base_ptr segfault regression (#50231) The function `jl_gc_internal_obj_base_ptr` takes a pointer and tries to determine if it is a valid object pointer. As such it has to carefully validate all data it reads, and abort whenever there are obvious inconsistencies. This patch adds a check which aborts when `meta->osize` is zero, just before we perform a division-with-remainder by this value, thus avoiding a potential division-by-zero exception. This fixes a crash we are seeing in our code. The crash did not happen before PR #49644 was merged because back then there was a check for `meta->ages` not being zero, which apparently was enough to detect invalid values for `meta` (e.g. when `meta` points into a null page). | 22 June 2023, 13:45:44 UTC |
| de7670e | Neal Gafter | 22 June 2023, 12:58:54 UTC | Correct the documentation for `pkgdir`. (#50255) | 22 June 2023, 12:58:54 UTC |
| a34ff20 | Shuhei Kadowaki | 22 June 2023, 06:15:33 UTC | allow `Meta.parse` to take optional `filename` argument (#50224) Code analysis tools such as JET.jl use `Meta.parse` to get a Julia code representation from textual data, or to display diagnostic messages when the input data is invalid. In such scenario it would be beneficial if these diagnostic messages could refer to the original file name of the input data. This commit allows `Meta.parse` to accept the optional `filename::AbstractString` argument for this purpose. If an error occurs, the `filename` can be included in the diagnostic information. The default value is set to `"none"`, thus preserving existing behavior. | 22 June 2023, 06:15:33 UTC |
| fbb8d6c | Claire Foster | 22 June 2023, 00:28:30 UTC | Bump JuliaSyntax to 0.4.5 (#50253) Fixes an issue with error formatting in Meta.parse with nontrivial starting indices (#50245) | 22 June 2023, 00:28:30 UTC |
| e7d00e1 | Jameson Nash | 21 June 2023, 23:41:04 UTC | [LibGit2] fix wrong definition of ConfigStruct (#50247) | 21 June 2023, 23:41:04 UTC |
| 8d0d012 | Elliot Saba | 21 June 2023, 16:13:02 UTC | Merge pull request #50208 from giordano/mg/external-llvm-patches [build] Allow automatically applying Julia's patches to LLVM source | 21 June 2023, 16:13:02 UTC |
| 4f0216a | Cody Tapscott | 21 June 2023, 11:26:54 UTC | Clear specsigflags correctly for copied code instance (#50238) As provided by Jameson to fix an issue encountered in a downstream project. Co-authored-by: Jameson Nash <vtjnash@gmail.com> | 21 June 2023, 11:26:54 UTC |
| 94dd5cf | Shuhei Kadowaki | 21 June 2023, 11:26:36 UTC | compiler: add few more type annotations (#50242) | 21 June 2023, 11:26:36 UTC |
| 95749c3 | Keno Fischer | 21 June 2023, 00:52:53 UTC | Avoid creating invalid PhiNodes in IR passes (#50235) As of #50158, irverify catches cases where PhiNodes show up in the middle of a basic block (which is illegal). Unfortunately, it turns out there were two cases in Base, where we created just such code: 1. When cfg_simplify! merged basic blocks, it didn't bother to delete (resp, replace by the one incoming edge) the PhiNodes in the basic block it was merging. 2. In irinterp we try to delete instructions that result in constants. This is not legal if the instruction is a PhiNode. The second of these is somewhat unfortunate, but any subsequent compaction will of course take care of it, so I don't think it's a huge issue to just disable the replacement. | 21 June 2023, 00:52:53 UTC |
| 71a6a32 | Elliot Saba | 20 June 2023, 21:51:58 UTC | Merge pull request #50240 from fingolfin/patch-1 LICENSE.md: update copyright years | 20 June 2023, 21:51:58 UTC |
| 5098646 | Max Horn | 20 June 2023, 21:47:24 UTC | LICENSE.md: update copyright years | 20 June 2023, 21:47:24 UTC |
| a40cb96 | Jameson Nash | 20 June 2023, 20:00:20 UTC | Merge pull request #50162 from topolarity/dlfind-search-order Lookup libraries in `libjulia-*` before `jl_exe_handle` | 20 June 2023, 20:00:20 UTC |
| 4195a0e | Elliot Saba | 20 June 2023, 19:24:21 UTC | Merge pull request #50232 from JuliaLang/sf/dont_precompile_multiple_modules_into_a_ji_file Throw precompilation error if dependency load failure during incremental precompilation | 20 June 2023, 19:24:21 UTC |
| 9032926 | Elliot Saba | 20 June 2023, 17:05:58 UTC | Throw precompilation error if dependency load failure during incremental precompilation In rare cases, if we fail to load a dependency during precompilation, we can fall-through to the "load locally" fallthrough in `_require()`. However, if this happens during incremental precompilation, this ends up emitting `.ji` files that have multiple modules embedded within, which can cause massive precompilation issues further down the chain, as dependencies which try to load our `.ji` file themselves get corrupted. This catches the error at the source, refusing to generate such a `.ji` file in the first place. | 20 June 2023, 17:05:58 UTC |
| a8ef873 | Elliot Saba | 20 June 2023, 16:59:54 UTC | Merge pull request #50230 from JuliaLang/kf/dontreversecache Don't reverse! load_path cache | 20 June 2023, 16:59:54 UTC |
| e2b0f66 | Cody Tapscott | 20 June 2023, 14:57:04 UTC | Merge branch 'master' into dlfind-search-order | 20 June 2023, 14:57:04 UTC |
| d48e17b | Keno Fischer | 20 June 2023, 14:50:59 UTC | Don't reverse! load_path cache As noted in https://github.com/JuliaLang/julia/pull/50119#discussion_r1235329616. | 20 June 2023, 14:50:59 UTC |
| a74e9f9 | Cody Tapscott | 20 June 2023, 06:05:04 UTC | Install `libclang_rt.asan-<arch>` properly | 20 June 2023, 06:05:04 UTC |
| 0da46e2 | pchintalapudi | 20 June 2023, 02:23:18 UTC | Name LLVM variables from codegen (#50094) | 20 June 2023, 02:23:18 UTC |
| 427b123 | Oscar Smith | 19 June 2023, 23:16:16 UTC | Merge pull request #46410 from dalum/master Add mutating `stat!` function for non-allocating filesystem `stat` | 19 June 2023, 23:16:16 UTC |
| 82ab124 | Oscar Smith | 19 June 2023, 23:08:52 UTC | Merge pull request #46372 from JuliaLang/cjf/juliasyntax-stdlib Enable JuliaSyntax.jl as an alternative Julia parser | 19 June 2023, 23:08:52 UTC |
| d8a4810 | Oscar Smith | 19 June 2023, 23:08:31 UTC | Merge branch 'master' into cjf/juliasyntax-stdlib | 19 June 2023, 23:08:31 UTC |
| 690a5f6 | Keno Fischer | 19 June 2023, 19:23:17 UTC | Thread through lattice in a couple more places (#50216) Now that the lattice code has stabilized, we should probably consider getting rid of the fallback methods to catch all the remaining places where we're relying on it (and to prevent any new ones from being introduced), but for now, just fix two cases I happened to run into. | 19 June 2023, 19:23:17 UTC |
| da9cc1a | pchintalapudi | 19 June 2023, 16:45:41 UTC | Expose the newpm pipeline to a C api (#50210) | 19 June 2023, 16:45:41 UTC |
| f555b4b | Ian Butterworth | 19 June 2023, 16:03:23 UTC | Tweak cache pidlocking (#50214) | 19 June 2023, 16:03:23 UTC |
| 90521dd | pchintalapudi | 19 June 2023, 03:54:21 UTC | Emphasize the default base of multiversioning in the documentation, and discourage the implicit default base (#50211) | 19 June 2023, 03:54:21 UTC |
| 41e8d7f | Mosè Giordano | 18 June 2023, 16:42:04 UTC | [build] Allow automatically applying Julia's patches to LLVM source | 18 June 2023, 17:22:49 UTC |
| 9501540 | Keno Fischer | 18 June 2023, 00:14:32 UTC | Merge pull request #50201 from JuliaLang/kf/metherrworldage Pass through world age for kwargs MethodError | 18 June 2023, 00:14:32 UTC |
| fde8692 | Oscar Smith | 17 June 2023, 21:30:37 UTC | Merge branch 'master' into cjf/juliasyntax-stdlib | 17 June 2023, 21:30:37 UTC |
| 65523e4 | pchintalapudi | 17 June 2023, 20:46:45 UTC | Add some JIT docs (#50168) | 17 June 2023, 20:46:45 UTC |
| 60bf0b6 | Claire Foster | 24 May 2023, 21:10:29 UTC | Enable JuliaSyntax.jl as the defult parser * Vendor JuliaSyntax into Base via deps directory * Install JuliaSyntax as the Julia parser unless the environment variable JULIA_USE_NEW_PARSER=0 is set. * Add a function to set the Core._parse binding. Required because we'd like to set the binding during Base.__init__. This can be done with `Core.eval` but that doesn't work well in incremental compilation mode. Also accommodate JuliaSyntax within tests: * When JuliaSyntax is enabled, ignore error messages in parser tests which are tested separately upstream - error messages are inherently expressed a bit differently when they go alongside full source location info. * Accommodate a small number of incompatibilities where in JuliaSyntax - `import .Mod.x as (a.b)` is a syntax not lowering error - `f(2x for x=1:10, y` is `Expr(:incomplete)` not `Expr(:error)` - `incomplete_tag` is more precise for `:block` vs `:other` - `global const` without an assignment is a syntax error, in keeping with plain `const` without assignment being a syntax error (not lowering error). * Adjust a few tests to be more precise about testing lowering vs the parser. * Make Meta.parse doctest compatible with JuliaSyntax errors | 17 June 2023, 05:45:42 UTC |
| 2d68286 | Claire Foster | 27 May 2023, 21:23:10 UTC | Fix string escaping in REPL completion of paths REPL completion of paths within strings need to be escaped according to the usual escaping rules, and delimited by the starting " rather than whitespace. This differs from completion of paths within cmd backticks which need to be escaped according to shell escaping rules. Separate these cases and fix string escaping. This was found because JuliaSyntax emits an Expr(:error) rather than Expr(:incomplete) for paths inside strings with invalid escape sequences before whitespace. | 17 June 2023, 05:45:42 UTC |
| 964f0d6 | Claire Foster | 28 May 2023, 07:12:42 UTC | Pass Int rather than UInt as lengths to core parser hook This is more consistent with the way we're likely call it from the Julia side via Meta.parse(). | 17 June 2023, 05:25:12 UTC |
| 8caba95 | Claire Foster | 26 May 2023, 20:33:00 UTC | Show top level location for any `exc` in `eval(Expr(:error, exc))` Previously this only worked when `exc` was a `String`. | 17 June 2023, 05:25:12 UTC |
| 1a6cd97 | Claire Foster | 24 May 2023, 23:13:33 UTC | Only wrap Strings in Meta.ParseError String errors come from the flisp parser as Expr(:error). But other than that, allow the parser library to choose its own error type. | 17 June 2023, 05:25:12 UTC |
| 9e7bb12 | Claire Foster | 24 May 2023, 21:11:01 UTC | Add `Meta.ParseError` detail field Here we add a `detail` field to `Meta.ParseError`, but retain the `msg::String` field for compatibility. `showerror(::ParseError)` defers to the `detail` field if it's present. This allows us to still throw `Meta.ParseError` from `Meta.parse` for compatibility, but allow more expressivity when `detail` is set to an exception type like `JuliaSyntax.ParseError`. | 17 June 2023, 05:25:12 UTC |
| b6902ae | Claire Foster | 24 May 2023, 21:10:58 UTC | Make incomplete_tag extensible This allows `incomplete_tag` to work when Expr(:incomplete) holds a Meta.ParseError as its child rather than a plain string, as it will when JuliaSyntax is enabled. | 17 June 2023, 05:25:12 UTC |
| fd1cec2 | Keno Fischer | 17 June 2023, 05:03:37 UTC | Improve effects for Base.fieldindex (#50199) Split out the error path into a function with separate effects assumptions, so that constant propagation on `err` can conclude that the `err=false` case does not throw. Fixes #50198. | 17 June 2023, 05:03:37 UTC |
| 18dd7a2 | Keno Fischer | 16 June 2023, 22:50:28 UTC | Don't color Any... red if the method actually matched Fixes the drive-by observation I made in #50200. | 16 June 2023, 22:50:28 UTC |
| f0881ef | Keno Fischer | 16 June 2023, 22:38:00 UTC | Pass through world age for kwargs MethodError Fixes #50200 | 16 June 2023, 22:38:39 UTC |
| ba251e8 | Lilith Orion Hafner | 16 June 2023, 20:49:24 UTC | Fix sorting bugs (esp `MissingOptimization`) that come up when using SortingAlgorithms.TimSort (#50171) | 16 June 2023, 20:49:24 UTC |
| c5b0a6c | Ian Butterworth | 16 June 2023, 20:32:46 UTC | pidlock cache file precompilation (#49052) | 16 June 2023, 20:32:46 UTC |
| 631d187 | Gabriel Baraldi | 16 June 2023, 13:33:32 UTC | codegen: pass the pgcstack as an argument to specsig calls (#50093) The safepoint at function entry made it so that every function call did a relatively expensive load from the PTLS, we can instead pass the PTLS as an argument to functions making it significantly cheaper. Also use the swift calling conventions, that together with the `swiftself` attribute makes it so it's very likely the argument is kept in a register between calls. Fixes: https://github.com/JuliaLang/julia/issues/50068 | 16 June 2023, 13:33:32 UTC |
| 71c5115 | Tim Besard | 16 June 2023, 06:47:59 UTC | Export jl_method_set_source. (#49236) This makes it possible to experiment with an OpaqueClosure-like API, which uses Method objects, in the GPU stack. | 16 June 2023, 06:47:59 UTC |
| 432f300 | Valentin Churavy | 15 June 2023, 22:41:38 UTC | Remove xsaves from Zen cpuspec (#50186) | 15 June 2023, 22:41:38 UTC |
| 0aa93e3 | Igor Proskurin | 15 June 2023, 20:56:20 UTC | fix missing punctuation in docs (#50126) * fix missing punctuation * Update types.md --------- Co-authored-by: Viral B. Shah <ViralBShah@users.noreply.github.com> | 15 June 2023, 20:56:20 UTC |
| 9d1ac97 | Ian Butterworth | 15 June 2023, 18:14:45 UTC | Allow unquoted symbols for threadpool in `Threads.@spawn` (#50182) Co-authored-by: Julian Samaroo <jpsamaroo@gmail.com> | 15 June 2023, 18:14:45 UTC |
| 0b87d95 | Steve Kelly | 15 June 2023, 17:44:49 UTC | Use in-place operations where appropriate (#50119) This changes some uses of reverse/sort/filter to the in-place versions where appropriate. Should provide some minor memory savings. Co-authored-by: Lilith Orion Hafner <lilithhafner@gmail.com> Co-authored-by: Jakob Nybo Nissen <jakobnybonissen@gmail.com> | 15 June 2023, 17:44:49 UTC |
| 62e57f7 | Oscar Smith | 15 June 2023, 14:59:47 UTC | Merge branch 'master' into master | 15 June 2023, 14:59:47 UTC |
| 0f26966 | Keno Fischer | 15 June 2023, 06:07:56 UTC | Make sure Core.Compiler can throw kwarg mismatch errors (#50174) The _new_NamedTuple helper was in a Base-only branch, causing ``` julia> Core.eval(Core.Compiler, quote f(;a=1) = a end) f (generic function with 1 method) julia> Core.Compiler.f(;b=2) ERROR: UndefVarError: `_new_NamedTuple` not defined Stacktrace: [1] macro expansion @ Core.Compiler ./namedtuple.jl:0 [inlined] [2] structdiff(a::@NamedTuple{b::Int64}, b::Type{NamedTuple{(:a,)}}) @ Core.Compiler ./namedtuple.jl:421 [3] top-level scope @ REPL[2]:1 ``` After this change, we have the expected ``` julia> Core.eval(Core.Compiler, quote f(;a=1) = a end) f (generic function with 1 method) julia> Core.Compiler.f(;b=2) ERROR: MethodError: no method matching f(; b::Int64) Closest candidates are: f(; a) got unsupported keyword argument "b" @ Core REPL[13]:1 Stacktrace: [1] kwerr(kw::@NamedTuple{b::Int64}, args::Function) @ Core.Compiler ./error.jl:165 [2] top-level scope @ REPL[14]:1 ``` | 15 June 2023, 06:07:56 UTC |
| 5db2c27 | pchintalapudi | 15 June 2023, 03:47:49 UTC | Add marker passes to various pipeline stages (#50111) | 15 June 2023, 03:47:49 UTC |
| a595274 | Juan José González Oneto | 15 June 2023, 03:27:15 UTC | Fixed small typo in `map`'s docstring (#50173) | 15 June 2023, 03:27:15 UTC |
| 9e43494 | pchintalapudi | 14 June 2023, 22:37:12 UTC | Compile JITLink for every platform (#50099) | 14 June 2023, 22:37:12 UTC |
| abbe045 | Jakob Nybo Nissen | 14 June 2023, 20:14:58 UTC | FAQ: Clarify use of PROGRAM_FILE for importable scripts (#50139) In issue #45852, the proposed `@is_script` macro as a shorthand for `abspath(PROGRAM_FILE) == @__FILE__` was rejected, because it was agreed that the pattern of using the same file as a script and an importable library is questionable, so there is no reason to make it too idiomatic. For this reason, mention in the FAQ that while checking `PROGRAM_FILE` can be used to check if a file is run as a script, it is not recommended to have files double as both scripts and importable libraries. | 14 June 2023, 20:14:58 UTC |
| 0fb86dd | Oscar Smith | 14 June 2023, 18:36:36 UTC | Fix `rationalize(::Complex)` test on 32 bit systems (#50163) * Fix `rationalize(::Complex)` test on 32 bit systems the test was relying on `Int===Int64` | 14 June 2023, 18:36:36 UTC |
| 834aad4 | Ashley Milsted | 14 June 2023, 14:15:41 UTC | Add (broken) tests for allocation in 5-arg mul!() (#49210) | 14 June 2023, 14:15:41 UTC |
| df09f67 | Keno Fischer | 14 June 2023, 09:06:07 UTC | irverify: Enforce invariant that PhiNodes are at the beginning of a BB (#50158) We have an invariant that all PhiNodes are at the beginning of a BasicBlock (only possible interrupted by a `nothing`) and we rely on this in various places for correctness. However, we did not actually verify this invariant. | 14 June 2023, 09:06:07 UTC |
| 1af6648 | Cody Tapscott | 14 June 2023, 01:31:42 UTC | Use `-shared-libasan` when linking w/ ASAN enabled. This flag is needed to ensure that `libclang_rt.asan-*.so` appears explicitly in the DT_NEEDED entries of libjulia-*. Without this entry, e.g. `dlsym(libjulia_internal_handle)` can end up finding symbols directly in libc.so.6, effectively bypassing the ASAN interceptors. | 14 June 2023, 02:36:17 UTC |
| 8a1b642 | Tanay Sharma | 14 June 2023, 00:31:55 UTC | Add method to rationalize `Rational` (#43427) * Add Method to Rationalize Rational and Integer Co-authored-by: Jeff Bezanson <jeff.bezanson@gmail.com> Co-authored-by: Oscar Smith <oscardssmith@gmail.com> | 14 June 2023, 00:31:55 UTC |
| 03c4bc1 | Diogo Netto | 14 June 2023, 00:29:16 UTC | Promote objects more eagerly (#49644) Simplifies generational behaviour | 14 June 2023, 00:29:16 UTC |
| 82c89c6 | Cody Tapscott | 13 June 2023, 22:17:21 UTC | Lookup libraries in `libjulia-*` before `jl_exe_handle` We do not use `dlvsym` to separate the symbols between multiple copies of libjulia, instead preferring to resolve symbols directly against the appropriate internal library handle. During bootstrapping, many internal symbols (e.g. `jl_fl_parse`) are available in the global EXE namespace, so we need to adapt our search order to resolve symbols in internal libraries first. With this fix, no sysimage symbols are resolved to `jl_exe_handle` (which is generally broken in Julia-in-Julia scenarios): ``` $ cat objdump_after.txt | grep libjulia_internal_handle | wc 1131 14703 145899 $ cat objdump_after.txt | grep jl_exe_handle | wc 0 0 0 ``` versus before: ``` $ cat objdump_before.txt | grep libjulia_internal_handle | wc 577 7501 74433 $ cat objdump_before.txt | grep jl_exe_handle | wc 554 7202 63710 ``` | 13 June 2023, 22:36:58 UTC |
| 84fc130 | Cody Tapscott | 13 June 2023, 19:06:19 UTC | Merge pull request #50083 from topolarity/fasttls-semaphore Prevent registering static FASTTLS with multiple libjulias | 13 June 2023, 19:06:19 UTC |
| 9d839f9 | Keno Fischer | 13 June 2023, 17:25:12 UTC | Refactor irinterp refinement logic (#50155) This continues the refactoring begun by #49340 to have irinterp consume the IR_FLAG_REFINED flag. This essentially has the same effect as the extra_reprocess bitset that irinterp takes, so we can remove that. However, there is a related issue where we would like to inform irinterp that we have *already* refined the type of a particular statement (likely using information not available to the irinterp) and would like it to just propagate that if possible. So bring back that extra bitset with a new name and these new semantics to make that possible. While I was working on this, I also noticed that the control hook I added in #48199 wasn't quite working as advertised. I don't currently need it, so rather than trying to work through an API without a concrete consumer, just nuke that hook for now. I do still think it'll be required at some point, but we can always add it back. | 13 June 2023, 17:25:12 UTC |
| 320e00d | Jameson Nash | 13 June 2023, 14:33:57 UTC | lowering: try to ignore special metadata nodes in tail position (#49963) A number of packages are relying on lowering removing or ignoring line number nodes that they place in incorrect position in the AST. Accommodate those packages accordingly. Fix #49920 | 13 June 2023, 14:33:57 UTC |
| 97f445e | Cody Tapscott | 13 June 2023, 14:30:57 UTC | Update cli/loader_lib.c From Jameson: > apparently on Windows, the _Atomic x syntax means something slightly different and is not correctly implemented Co-authored-by: Jameson Nash <vtjnash@gmail.com> | 13 June 2023, 14:30:57 UTC |
| 0d89d8b | Shuhei Kadowaki | 13 June 2023, 10:50:10 UTC | Test: support multiple assignments for `@testset let` (#50151) Nested `ContextTestset` is supported, so we can stack it when there are multiple assignments in a given `let` block. ```julia julia> @testset let logi = log(im), op = !iszero @test imag(logi) == π/2 @test op(real(logi)) end Test Failed at none:3 Expression: !(iszero(real(logi))) Context: logi = 0.0 + 1.5707963267948966im op = !iszero ERROR: There was an error during testing ``` | 13 June 2023, 10:50:10 UTC |
| f007c01 | Shuhei Kadowaki | 13 June 2023, 10:46:51 UTC | fix `array_builtin_common_nothrow` for `arrayref` (#50152) | 13 June 2023, 10:46:51 UTC |
| 7cd7a1b | Mosè Giordano | 13 June 2023, 06:54:21 UTC | Enable JITLink in aarch64 linux. (#49745) * Enable JITLink in aarch64 linux. * Simplify logic to enable JITLink * Do not enable JITLink on Aarch64 Linux with LLVM < 15 * Add NEWS entry and mention JITLink memory bug in ARM devdocs --------- Co-authored-by: Sunho Kim <ksunhokim123@gmail.com> | 13 June 2023, 06:54:21 UTC |
| 061401e | Elliot Saba | 13 June 2023, 04:27:13 UTC | Merge pull request #50150 from JuliaLang/kf/refkillissues Fix various refcount violations | 13 June 2023, 04:27:13 UTC |
| 0083a04 | Keno Fischer | 13 June 2023, 01:24:42 UTC | Fix refcount violations in new ifelse sroa lifting Fixes a refcount violation in #49882. Unlike PhiNodes, the code for ifelse was referencing the old arguments in the newly inserted nodes, so they were getting counted. However, it did not update the count when replacing these arguments later. Fix this by aligning with PhiNode and leaving the arguments unset, pushing them in later. | 13 June 2023, 01:27:16 UTC |
| c3cd3eb | Keno Fischer | 13 June 2023, 01:23:20 UTC | Fix refcount violation for pending_nodes These nodes get passed through renaming, but with SSAValues already renamed. Due to a long-standing bug, we were failing to refcount those SSAValues on insertion. Fix that bug. | 13 June 2023, 01:27:16 UTC |
| 970941c | Shuhei Kadowaki | 12 June 2023, 22:38:40 UTC | effects: allow concrete-eval when `--check-bounds=no` if proven "safe" (#50107) From version 1.9 onwards, when `--check-bounds=no` is used, concrete-eval is completely disabled. However, it appears `--check-bounds=no` is still being used within the community, causing issues like the one reported in JuliaArrays/StaticArrays.jl#1155. Although we should move forward to a direction of eliminating the flag in the future (#48245), for the time being, there are many requests to carry out a certain level of compiler optimization, even when this flag is enabled. This commit aims to allow concrete-eval "safely" even under `--check-bounds=no`. Specifically, when the method call being analyzed is `:nothrow`, it should be predominantly safe to concrete-eval it under this flag. Technically, however, even `:nothrow` methods could trigger undefined behavior, since `:nothrow` isn't a strict constraint and it's possible for users to annotate potentially risky methods with `Base.@assume_effects :nothrow`. Nonetheless, since this possibility is acknowledged in `Base.@assume_effects` documentation, I feel it's fair to relegate it to user responsibility. | 12 June 2023, 22:38:40 UTC |
| bf9bbb2 | Sheehan Olver | 12 June 2023, 19:07:25 UTC | Add doc string for reflectorApply! (#50081) | 12 June 2023, 19:07:25 UTC |
| d69b1a2 | Christopher Rackauckas | 12 June 2023, 14:26:22 UTC | Add check call to getrf! (#50134) * Add check call to getrf! `lu!(A; check=false)` is supposed to disable the checking and leave it to the user: > When check = true, an error is thrown if the decomposition fails. When check = false, responsibility for checking the decomposition's validity (via issuccess) lies with the user. However, this is not quite true since `lu!` calls `getrf!` which internally does a check for `chkfinite` which does throw an error. This updates the `getrf!` function to have a `check` argument which is then used by `lu!` to fully disable the error throwing checks. * Update lapack.jl | 12 June 2023, 14:26:22 UTC |
| 75bda64 | Jeff Bezanson | 12 June 2023, 07:13:48 UTC | speed up expansion and lowering of ccall macro (#50077) | 12 June 2023, 07:13:48 UTC |
| a81398f | Mosè Giordano | 11 June 2023, 20:47:32 UTC | [CompilerSupportLibraries_jll] Upgrade to v1.0.5 (#50135) This build includes a fix from GCC 12.2 for a bug in libgomp 12.1 which was causing frequent crashes on Windows in third-party libraries. | 11 June 2023, 20:47:32 UTC |
| c58e508 | Martin Kunz | 10 June 2023, 16:12:48 UTC | fix(docs): typo in code-loading (#50123) | 10 June 2023, 16:12:48 UTC |
| a31c8ef | Shuhei Kadowaki | 10 June 2023, 04:12:00 UTC | inlining: fix the `joint_effects` calculation (#50117) There are cases when `joint_effects` already has `:nothrow` tainted while the matches are fully covered. | 10 June 2023, 04:12:00 UTC |
| 0919cba | Chris Hodapp | 09 June 2023, 20:30:51 UTC | Fix typo in calling-c-and-fortran-code.md (#50121) | 09 June 2023, 20:30:51 UTC |
| 59bf9e8 | Kristoffer Carlsson | 09 June 2023, 19:24:52 UTC | use a fixed world for code loading (#49525) | 09 June 2023, 19:24:52 UTC |
| feaea22 | Kristoffer Carlsson | 09 June 2023, 19:23:54 UTC | print the skipping bytes notification when skipping printing parts of a string in color (#50096) | 09 June 2023, 19:23:54 UTC |
| f8827f1 | Jishnu Bhattacharya | 09 June 2023, 14:46:41 UTC | Structured display for adjoint/transpose (#49979) * Structured display for adjoint/transpose * structured display for Symmetric/Hermitian * tests with Symmetric and Transpose | 09 June 2023, 14:46:41 UTC |
| 2584b59 | Jishnu Bhattacharya | 09 June 2023, 13:17:47 UTC | Fill zeros only if necessary in `Matrix(::Diagonal)` and friends (#50008) * delay filling zeros while convering (sym)tridiag to matrix * conditional zero fill for bidiagonal/diagonal | 09 June 2023, 13:17:47 UTC |
| 43d7f88 | Shuhei Kadowaki | 09 June 2023, 04:28:41 UTC | effects: audit `EFFECTS_UNKNOWN` usages (#50106) | 09 June 2023, 04:28:41 UTC |
| d041162 | Viral B. Shah | 09 June 2023, 01:42:26 UTC | Update mailmap (#50116) | 09 June 2023, 01:42:26 UTC |
| 7ceea13 | Erik Schnetter | 09 June 2023, 01:39:18 UTC | Update Unicode.jl (#50087) Correct confusing typo in documentation | 09 June 2023, 01:39:18 UTC |
| 746a15b | Lilith Orion Hafner | 08 June 2023, 21:52:57 UTC | Fail gracefully when attempting pair destructuring (#49368) `a => b = x` was interpreted as a function definition. Now it's an error unless you `import Base: =>` | 08 June 2023, 21:52:57 UTC |
| c4d162e | Max Horn | 08 June 2023, 15:55:41 UTC | mkpidlock: clarify that stale_age is given in seconds (#50014) | 08 June 2023, 15:55:41 UTC |
| 0a2d6fc | Gabriel Baraldi | 08 June 2023, 15:28:34 UTC | Don't use exchange in the hot path of the GC (#50021) * Don't use exchange in the hot path of the GC and save one extra load | 08 June 2023, 15:28:34 UTC |
