Skip to main content

Browse the archive

https://github.com/JuliaLang/julia
19 April 2024, 06:28:37 UTC
sort by:
Revision Author Date Message Commit Date
16dc265 KristofferC 13 September 2023, 11:08:07 UTC do not bail when finding directories with matching name in dlload 13 September 2023, 11:09:00 UTC
572fa50 Shuhei Kadowaki 13 September 2023, 10:35:20 UTC remove compiler/ssair/driver.jl (#51289) I don't think that the extra layer of indirection is needed. 13 September 2023, 10:35:20 UTC
aa6b34e Shuhei Kadowaki 13 September 2023, 10:34:48 UTC refactor post-opt analysis and get rid of closures (#51290) 13 September 2023, 10:34:48 UTC
0fdcd55 Shuhei Kadowaki 13 September 2023, 10:33:50 UTC irinterp: fix `is_terminator_or_phi` check (#51291) Also adds some cosmetic changes. @nanosoldier `runbenchmarks("inference", vs=":master")` 13 September 2023, 10:33:50 UTC
141e07e Shuhei Kadowaki 13 September 2023, 03:28:12 UTC elaborate `incremental` argument of `Base.generating_output` (#51281) Follows up #51216. 13 September 2023, 03:28:12 UTC
4af6be8 Ian Butterworth 12 September 2023, 22:19:26 UTC shorten stale_age for cachefile lock (#51175) 12 September 2023, 22:19:26 UTC
431e0bd Diogo Netto 12 September 2023, 18:57:23 UTC don't print corrupted chunk id on crash (#51283) This value will be corrupted if we reach this branch, so it seems pointless to print it. 12 September 2023, 18:57:23 UTC
832e46d Jameson Nash 12 September 2023, 16:23:19 UTC Ryu: make sure adding zeros does not overwrite trailing dot (#51254) Fix #43129 12 September 2023, 16:23:19 UTC
5d82d80 Shuhei Kadowaki 12 September 2023, 02:14:30 UTC add tfuncs for `[and|or]_int` intrinsics (#51266) So that they can be constant folded when either of argument is known to be `Core([false|true])`. It may help inference accuracy. 12 September 2023, 02:14:30 UTC
ad27e67 Yichao Yu 11 September 2023, 21:15:25 UTC Fix Markdown test from #49454 (#51268) Actually testing the output instead of printing to stdout. 11 September 2023, 21:15:25 UTC
b9203ce Valentin Churavy 11 September 2023, 19:31:18 UTC Fix Profiler signal listener test (#51272) The `close(t)` seems to be a typo. 11 September 2023, 19:31:18 UTC
1097481 Tim Besard 11 September 2023, 14:39:29 UTC GMP: Gracefully handle more overflows. (#51243) Fixes https://github.com/JuliaLang/julia/issues/8286, which regressed due to a GMP upgrade in https://github.com/JuliaLang/julia/pull/45375. 11 September 2023, 14:39:29 UTC
ea49abe Shuhei Kadowaki 11 September 2023, 09:56:41 UTC SROA: remove dead lifting cache mechanism (#51014) It hadn't been fixed for years. Let's get rid of the unnecessary allocation cost. 11 September 2023, 09:56:41 UTC
7b9fdf8 Valentin Churavy 10 September 2023, 01:51:57 UTC Make _global_logstate a typed global (#51257) 10 September 2023, 01:51:57 UTC
bbbcc4f Gabriel Baraldi 09 September 2023, 20:43:01 UTC Add lock around uv_unref during init (#51236) This is not a legal operation outside the lock because it's not atomic Co-authored-by: Valentin Churavy <v.churavy@gmail.com> 09 September 2023, 20:43:01 UTC
8d4d641 Iztok Fister Jr 09 September 2023, 13:49:52 UTC Update man page with internet resources and licensing (#51155) Pull request proposes a minor modification to the man page. In this PR, two additional sections for the man page are added: - Internet resources and - licensing. 09 September 2023, 13:49:52 UTC
3950167 Gabriel Baraldi 09 September 2023, 10:21:54 UTC Check if malloc has succeeded before updating GC counters (#51247) 09 September 2023, 10:21:54 UTC
4c3aaa2 Shuhei Kadowaki 09 September 2023, 07:53:44 UTC reflection: define `Base.generating_output` utility function (#51216) And replaces `ccall(:jl_generating_out, ...)` with the new function. --------- Co-authored-by: Jameson Nash <vtjnash@gmail.com> 09 September 2023, 07:53:44 UTC
4363895 Keno Fischer 09 September 2023, 06:28:00 UTC Unify flags_for_effects handling (#51238) There was two places where we were turning Effects into flags, one in the optimizer, one in inference. Unify these two both use the same function, but we need to refactor a bit to make sure all the effects get passed to the correct place. --------- Co-authored-by: Shuhei Kadowaki <aviatesk@gmail.com> 09 September 2023, 06:28:00 UTC
4c024e6 Shuhei Kadowaki 09 September 2023, 01:47:02 UTC optimizer: code quality and performance improvements to post-opt analysis (#51211) This PR is composed of a set of fixes and improvements to `ipo_dataflow_analysis!`. I will rebase this once we merge #51185 and #51188, so still WIP, but let's see benchmark. @nanosoldier `runbenchmarks("inference", vs=":master")` --------- Co-authored-by: Keno Fischer <keno@juliahub.com> 09 September 2023, 01:47:02 UTC
80603a4 Shuhei Kadowaki 09 September 2023, 01:44:56 UTC mark return type of failed concrete eval of intrinsics as `Bottom` (#51251) Because of the consistency assumed by `is_pure_intrinsic_infer`, we can aggressively mark the return type as `Bottom` in a case when the concrete evaluation of an intrinsic fails. 09 September 2023, 01:44:56 UTC
3fc4f6b DilumAluthgeBot 09 September 2023, 00:54:17 UTC 🤖 [master] Bump the Downloads stdlib from f97c72f to 8a614d5 (#51246) Stdlib: Downloads URL: https://github.com/JuliaLang/Downloads.jl.git Stdlib branch: master Julia branch: master Old commit: f97c72f New commit: 8a614d5 Julia version: 1.11.0-DEV Downloads version: 1.6.0(It's okay that it doesn't match) Bump invoked by: @DilumAluthge Powered by: [BumpStdlibs.jl](https://github.com/JuliaLang/BumpStdlibs.jl) Diff: https://github.com/JuliaLang/Downloads.jl/compare/f97c72fbd726e208a04c53791b35cc34c747569f...8a614d592810b15d17885838dec61da244a12e09 ``` $ git log --oneline f97c72f..8a614d5 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> 09 September 2023, 00:54:17 UTC
8e77b63 Jameson Nash 08 September 2023, 14:59:21 UTC refine effects of some reflection queries (#51235) 08 September 2023, 14:59:21 UTC
a7697fd DilumAluthgeBot 08 September 2023, 13:29:55 UTC 🤖 [master] Bump the NetworkOptions stdlib from 976e51a to aab83e5 (#51237) Stdlib: NetworkOptions URL: https://github.com/JuliaLang/NetworkOptions.jl.git Stdlib branch: master Julia branch: master Old commit: 976e51a New commit: aab83e5 Julia version: 1.11.0-DEV NetworkOptions version: 1.2.0(Does not match) Bump invoked by: @DilumAluthge Powered by: [BumpStdlibs.jl](https://github.com/JuliaLang/BumpStdlibs.jl) Diff: https://github.com/JuliaLang/NetworkOptions.jl/compare/976e51a48abb4e09356f1979c3cde8fbc0852e2e...aab83e5dd900c874826d430e25158dff43559d78 ``` $ git log --oneline 976e51a..aab83e5 aab83e5 Reset BUNDLED_KNOWN_HOSTS_FILE in case we serialized a value due to precompilation (#31) ``` Co-authored-by: Dilum Aluthge <dilum@aluthge.com> 08 September 2023, 13:29:55 UTC
5e85876 Gunnar Farnebäck 08 September 2023, 10:18:45 UTC Warn about mutating keys in Dict docstring. (#51171) Be clear in the `Dict` docstring that it's not in general safe to mutate keys. Cf. https://discourse.julialang.org/t/unable-to-access-value-in-dict-via-known-keys/103434. 08 September 2023, 10:18:45 UTC
de297fa Shuhei Kadowaki 08 September 2023, 06:11:21 UTC remove special case for the no-op first statement in `fully_eliminated` (#51227) It looks like the special case is no longer needed. If this gets broken by future changes, I would like fix it or mark the test cases as `broken`. 08 September 2023, 06:11:21 UTC
39a5316 Shuhei Kadowaki 08 September 2023, 01:01:15 UTC optimize: fix `effect_free` refinement in post-opt dataflow analysis (#51185) Currently we never refine this information though. @nanosoldier `runbenchmarks("inference", vs=":master")` 08 September 2023, 01:01:15 UTC
b3741c0 Gabriel Baraldi 07 September 2023, 21:55:47 UTC Check again if the tty is open inside the IO lock (#51222) This can cause segfaults when exiting julia. Co-authored-by: Valentin Churavy <vchuravy@users.noreply.github.com> 07 September 2023, 21:55:47 UTC
e446bba Gabriel Baraldi 07 September 2023, 21:47:50 UTC Remove references to the LegacyPassManager (#51197) We still keep references in a couple places because the objectfile emission still uses it. 07 September 2023, 21:47:50 UTC
af6e08d Tim Besard 07 September 2023, 20:02:05 UTC Include JuliaSyntax.jl from the build directory. (#51231) This ensures the source isn't modified during an out-of-tree build. Fixes https://github.com/JuliaLang/julia/issues/51230 --------- Co-authored-by: Jameson Nash <vtjnash@gmail.com> 07 September 2023, 20:02:05 UTC
eab8d6b Tim Besard 07 September 2023, 19:36:30 UTC Fix getfield codegen for tuple inputs and unknown symbol fields. (#51234) 07 September 2023, 19:36:30 UTC
8599e2f Valentin Churavy 07 September 2023, 17:55:25 UTC add PersistentDict based on a HAMT (#51164) The implementation is based on a [Hash Array Mapped Trie (HAMT)](https://en.wikipedia.org/wiki/Hash_array_mapped_trie) following [Bagwell (2000)](http://infoscience.epfl.ch/record/64398/files/idealhashtrees.pdf). A HAMT uses a fixed branching factor (commonly 32) together with each node being sparse. In order to search for an entry we take the hash of the key and chunk it up into blocks, with a branching factor of 32 each block is 5 bits. We use those 5 bits to calculate the index inside the node and use a bitmap within the node to keep track if an element is already set. This makes search a `log(32, n)` operation. Persistency is implemented by path-copying. When we insert/delete a value into the HAMT we copy each node along the path into a new HAMT, all other nodes are shared with the previous HAMT. A noteable implementation choice is that I didn't add a (resizeable) root table. Normally this root table is dense and uses the first `t` bits to calculate an index within. This makes large HAMT a bit cheaper since the root-table effectivly folds multiple lookup steps into one. It does hurt persistent use-cases since path-copying means that we also copy the root node/table. Importantly the HAMT itself is not immutable/persistent, the use of it as part of the `PersistentDict` is. Direct mutation of the underlying data breaks the persistentcy invariants. One could use the HAMT to implement a non-persistent dictionary (or other datastructures). As an interesting side-note we could use a related data-structure [Ctrie](http://lamp.epfl.ch/~prokopec/ctries-snapshot.pdf) to implement a concurrent lock-free dictionary. Ctrie also support `O(1)` snapshotting so we could replace the HAMT used here with a Ctrie. 07 September 2023, 17:55:25 UTC
27fa5de Daniel Karrasch 07 September 2023, 14:23:49 UTC Introduce cholesky and qr hooks for wrapped sparse matrices (#51220) This is a follow-up to #51161. It introduces one level in the promotion pipeline which allows to call out-of-place versions of `cholesky` and `qr` for arguments whose `similar` copy yields a `SparseMatrixCSC`. 07 September 2023, 14:23:49 UTC
1305f40 Kristoffer Carlsson 07 September 2023, 09:13:32 UTC call some Pkg functions with invokelatest for BugReporting (#51219) When Pkg is not in the sysimage, this causes world age errors. 07 September 2023, 09:13:32 UTC
aeae142 Cody Tapscott 06 September 2023, 19:02:19 UTC slot2ssa: Fix spurious φᶜ edges (#51199) The unreachable here seems to be caused by the fact that (as of #50943) we re-use a more narrow type from Inference that correctly ignores these edges, but when inserting the `φᶜ` node in `slot2reg` we were including extra edges that never get exercised at runtime. I'm not sure _why_ this causes us to hit an unreachable, since the narrow type from inference is technically still valid (the catch block will never observe these spurious assignments at runtime), but this seems to fix the issue and anyway improves the quality of the IRCode generated by `slot2ssa`. Resolves #51159 06 September 2023, 19:02:19 UTC
a5b2197 Shuhei Kadowaki 06 September 2023, 12:56:36 UTC optimizer: skip post-opt IPO analysis when effects are precise already (#51188) 06 September 2023, 12:56:36 UTC
c1153d0 Cody Tapscott 06 September 2023, 11:32:04 UTC Tweak tests 06 September 2023, 11:32:04 UTC
f3d50b7 Colin Caine 06 September 2023, 10:36:31 UTC Fix extended help hint to give full line to enter (#51193) Before this commit, the REPL would say: Extended help is available with `??` Now it will give the full line required to display extended help: Extended help is available with `??LazyString` I believe that this was the intended function of the code that's already here, because currently the `extended_help_on` constant is never used. I've resisted the urge to pass the `line` through to `trimdocs` without a global because I don't know why @timholy did it that way to start with and because the smaller change makes this PR easier to review. 06 September 2023, 10:36:31 UTC
3d88550 Gabriel Baraldi 06 September 2023, 10:35:01 UTC Wait for other threads to finish compiling before exiting (#51213) This avoids a crashes where we run the destructors because C++ is fun and runs destructors before thread exit. 06 September 2023, 10:35:01 UTC
f9d1767 Jameson Nash 06 September 2023, 08:39:44 UTC [FileWatching] clarify mkpidlock docstring (#51214) 06 September 2023, 08:39:44 UTC
f066500 Lilith Orion Hafner 06 September 2023, 06:03:51 UTC Fix typos in JuliaSyntax.mk (#51210) before ``` x@x julia % make -C deps fastcheck-JuliaSyntax make: *** No rule to make target `fastcheck-JuliaSyntax'. Stop. x@x julia % make -C deps fastcheck-JuliSyntax make: *** No rule to make target `compile-JuliSyntax', needed by `check-JuliSyntax'. Stop. ``` after ``` x@x julia % make -C deps fastcheck-JuliaSyntax make: Nothing to be done for `fastcheck-JuliaSyntax'. ``` Co-authored-by: Lilith Hafner <Lilith.Hafner@gmail.com> 06 September 2023, 06:03:51 UTC
14c5f3d Lilith Orion Hafner 06 September 2023, 05:52:36 UTC Remove unused io argument to `insert_hlines` (#51182) 06 September 2023, 05:52:36 UTC
c84324f Jakob Nybo Nissen 05 September 2023, 19:16:26 UTC Bugfix: Convert numbers to Int in extendedterminfo (#51195) Before, in the extendedterminfo function in terminfo.jl, if the numbers array was nonempty, the function would fail as a `UInt32` cannot be implicitly converted to the output `Int` type. Do this conversion explicitly. Closes #51190 05 September 2023, 19:16:26 UTC
ba16663 Lars Göttgens 05 September 2023, 18:45:15 UTC Revert "Standardize the entry-point for Julia execution (#50974)" (#51196) This reverts commit 8e14322b5aa344639dd86bf9eabb84afe831fcba. As proposed by @Seelengrab in https://github.com/JuliaLang/julia/pull/50974#issuecomment-1706641240. This is not meant to be a permanent revert of #50974, but a possibility to have working PkgEval and nightly testing for packages (https://github.com/julia-actions/julia-buildpkg `@latest` doesn't work with it) until a solution to the discussion there has been decided on. 05 September 2023, 18:45:15 UTC
78e0ca3 Shuhei Kadowaki 05 September 2023, 16:10:12 UTC optimizer: fix the `all_rettypes_consistent` case of post-opt analysis (#51187) The case does not seem to be functional, although it looks like we never hit it currently. We should have added test cases for this, so WIP until we come up with some. 05 September 2023, 16:10:12 UTC
431bb81 Cody Tapscott 05 September 2023, 15:45:34 UTC Remove unused variable from `compute_trycatch` This variable is currently always zero. 05 September 2023, 15:47:34 UTC
354c367 Daniel Karrasch 05 September 2023, 15:39:45 UTC Allow SparseArrays to catch `lu(::WrappedSparseMatrix)` (#51161) Over the `AbstractMatrix` relaxation in v1.9, we missed a potential indirection for wrapped sparse matrices. Instead, by default, a `similar` copy is created (hence a sparse matrix) and then `lu!` with a pivot argument is called. Such a method, however, does not exist in SparseArrays.jl, which means that the `generic_lufact!` method gets called, which is probably really bad performance-wise, due to heavy reading and writing into the sparse matrix. This PR introduces one more level at which SparseArrays.jl (and perhaps other external packages) may interfere and redirect to their own implementations, in-place or out-of-place. 05 September 2023, 15:39:45 UTC
ee50411 Cody Tapscott 05 September 2023, 15:31:34 UTC slot2ssa: Fix spurious φᶜ edges The unreachable here seems to be caused by the fact that (as of #50299) we re-use a more narrow type from Inference that correctly ignores these edges, but when inserting the `φᶜ` node in `slot2reg` we were including extra edges that never get exercised at runtime. I'm not sure _why_ this causes us to hit an unreachable, since the narrow type from inference is technically still valid (the catch block will never observe these spurious assignments at runtime), but this seems to fix the issue and anyway improves the quality of the IRCode generated by `slot2ssa`. Resolves #51159 05 September 2023, 15:36:52 UTC
933ea1d Benjamin Lorenz 05 September 2023, 14:42:39 UTC build: don't default to m(arch|cpu|tune)=native when cross-compiling (#51172) 05 September 2023, 14:42:39 UTC
00ca93c Lilith Orion Hafner 05 September 2023, 14:05:05 UTC Put back `@inline` on unsafe_getindex (#51160) Fixup for #50467 05 September 2023, 14:05:05 UTC
fbf73f4 DilumAluthgeBot 05 September 2023, 11:21:54 UTC 🤖 [master] Bump the Pkg stdlib from 047734e4c to f570abd39 (#51186) Stdlib: Pkg URL: https://github.com/JuliaLang/Pkg.jl.git Stdlib branch: master Julia branch: master Old commit: 047734e4c New commit: f570abd39 Julia version: 1.11.0-DEV Pkg version: 1.11.0 Bump invoked by: @KristofferC Powered by: [BumpStdlibs.jl](https://github.com/JuliaLang/BumpStdlibs.jl) Diff: https://github.com/JuliaLang/Pkg.jl/compare/047734e4c7e2b5e99622b4803ec4ad0d49f1d1d3...f570abd39d0a6ab483e658cf65c69af59a152314 ``` $ git log --oneline 047734e4c..f570abd39 f570abd39 tweak how Pkg is loaded for precompiling when testing (#3606) d3bd38b90 sort compat entries in `pkg> compat` (#3605) 5e07cfed0 Ensure that `string(::VersionSpec)` is compatible with `semver_spec()` (#3580) 6bed7c41a Clarify handling of LOAD_PATH in docstring and REPL help (#3589) 5261b3be7 tweak test dep docs (#3574) 72b430d50 status: expand 2 symbol upgrade note to highlight *may* be upgradable (#3576) b32db473d precompile: show ext parent in output report (#3603) ``` Co-authored-by: Dilum Aluthge <dilum@aluthge.com> 05 September 2023, 11:21:54 UTC
fc2f997 Shuhei Kadowaki 05 September 2023, 08:57:09 UTC compiler: add `show` methods for `InferenceState` and `CachedMethodTable` (#51170) Both `InferenceState` and `CachedMethodTable` contain caches to store inference results, making their default `show` methods excessively verbose. This commit introduces specialized `show` methods designed to display only the most essential information. 05 September 2023, 08:57:09 UTC
1829cee Keno Fischer 05 September 2023, 05:43:00 UTC optimizer: Do not run if we know the function is ConstABI eligible (#51143) If we can determine that a function is sufficiently pure and returns a constant, we have special ABI that lets us throw away the code for it and just return the constant. However, we were still going through all the steps of actually computing the code, including running the optimizer on it, compressing it in preparation for caching, etc. We can just stop doing that and bypass the optimizer entirely. The actual change to do this is pretty tiny, but there's some unexpected fallout which this needs to cleanup: 1. Various tests expect code_* queries of things inferred to ConstABI to still return reasonable code. Fix this by manually emitting a ReturnNode at the end of inference if the caller is a reflection function. 2. This kinda wreaks havoc on the EscapeAnalysis tests, which work by using a side-effect of the optimizer, but a lot of the functions are ConstABI eligible, so the optimizer doesn't run any more. Fortunately, I'm in the middle of looking at overhauling EscapeAnalysis, so I'll have some chance to figure out how to change the test system to actually do this properly, rather than exfiltrating side effects. That said, since EscapeAnalysis is not in the default pipeline, I don't think we should hold the perf improvement until that is done. Benchmarked to be about 10% faster locally, but we'll see what nanosoldier thinks. --------- Co-authored-by: Shuhei Kadowaki <40514306+aviatesk@users.noreply.github.com> 05 September 2023, 05:43:00 UTC
8e14322 Keno Fischer 03 September 2023, 15:55:38 UTC Standardize the entry-point for Julia execution (#50974) This is a bit of a straw-man proposal (though I think mergable if people agree) to standardize the execution entrypoint for Julia scripts. I think there's at least four different ways that people might run a script: - As `julia main.jl` - As a PkgCompiler sysimage, then calling the main entry point - As a PkgCompiler "app", with the magic `julia_main` function - As a StaticCompiler product with an explicit entrypoint specified on the API. The main problem I have with all of these variants is that they're all different and it's kind of a pain to move between them. Here I propose that we standardize on `Main.main(ARGS)` as the entrypoint for all scripts. Downstream from that proposal, this PR then makes the following changes: 1. If a system image has an existing `Main.main`, that is the entry point for `julia -Jsysimage.so`. 2. If not, and the sysimage has a REPL, we call REPL.main (we could handle this by defaulting `Main.main` to a weak import of `REPL.main`, but for the purpose of this PR, it's an explicit fallback. That said, I do want to emhpasize the direction of moving the REPL to be "just another app". 3. If the REPL code is called and passed a script file, the REPL executes any newly defined Main.main after loading the script file. As a result, `julia` behaves the same as if we had generated a new system image after loading `main.jl` and then running julia with that system image. The further downstream implication of this is that I'd like to get rid of the distinction between PkgCompiler apps and system images. An app is simply a system image with a `Main.main` function defined (note that currently PkgCompiler uses `julia_main` instead). --------- Co-authored-by: Martijn Visser <mgvisser@gmail.com> 03 September 2023, 15:55:38 UTC
0cc0dbd Diogo Netto 03 September 2023, 12:26:07 UTC Metric for number of live bytes in the pool allocator (#51151) We want to study what is the degree of fragmentation we have in the pool allocator specifically. `pool_live_bytes` / `(live pages * GC_PAGE_SZ)` should provide an estimate of that. 03 September 2023, 12:26:07 UTC
70000ac Keno Fischer 03 September 2023, 02:15:42 UTC sysimg: Allow loading a system image that is already present in memory (#51121) I've written this code probably three times at this point, but for some reason it never made it into a PR. This allows loading a system image that has already been loaded into memory. This happen when wanting to distribute a static or mostly-static binary of julia code where the system image (and optionally other libraries like libjulia, etc.) are linked directly into the main executable. It is also useful for deployment to environments that do not have (or have incomplete) support for dynamic linking (e.g. wasm or embedded). 03 September 2023, 02:15:42 UTC
da86735 Jameson Nash 03 September 2023, 02:12:28 UTC fix debug typo in "add missing invoke edge for nospecialize targets (#51036)" (#51153) Causes `matches` to get replaced with `MethodMatch` instead, which then later will fail to match with the expected value. Fixes #51146 Co-authored-by: Dilum Aluthge <dilum@aluthge.com> 03 September 2023, 02:12:28 UTC
d949bb4 Jameson Nash 03 September 2023, 02:11:55 UTC broadcast: use recursion rather than ntuple to map over a tuple (#51154) Inference seems to have trouble with the anonymous function version, so go back to the recursive version. Fixes #51129 Probably also fixes #50859 03 September 2023, 02:11:55 UTC
6c1168a Keno Fischer 02 September 2023, 10:10:28 UTC Document and clean up CodegenParams (#51140) - Add documentation to CodegenParams fields per request in #51123 - Fix compare_cgparams which hadn't been updated for recent additions - Remove unused and untested generic_context The latter was a codegen option that I added, but eventually ended up not using anywhere, so remove it for the time being. That said, I may end up in a situation where I need it again in the very near future, so I may end up eating my words here, but if I need to put it back, I'll include a test at least ;). 02 September 2023, 10:10:28 UTC
ec2f1d3 Neven Sajko 01 September 2023, 19:52:47 UTC fix rawbigints (#51122) Using `Ptr` like that was incorrect. Among other issues, a `Ptr` doesn't own the data it points to, so hold a `String` instead. Fixes #51111 01 September 2023, 19:52:47 UTC
4954af9 Keno Fischer 01 September 2023, 19:52:14 UTC Make OffsetArray maximum! not as type unstable (#50447) The change in #50429 moves around some dispatch boundaries and pushes the allocations in the offsetarrays `maximum!` test over the limit. The implementation of that code is massively type unstable. Somewhat, ironically, the whole original point of that test was to test that the implementation was not type-unstable (#28941), so actually opt our OffsetArrays implementation into the interface that's supposed to guarantee that. If this PR is fine here, I'll submit the same upstream to avoid diverging the implementations too much. Co-authored-by: Jameson Nash <vtjnash@gmail.com> 01 September 2023, 19:52:14 UTC
a173010 DilumAluthgeBot 01 September 2023, 18:50:41 UTC 🤖 [master] Bump the SparseArrays stdlib from 54f4b39 to 4e6776a (#51142) Stdlib: SparseArrays URL: https://github.com/JuliaSparse/SparseArrays.jl.git Stdlib branch: main Julia branch: master Old commit: 54f4b39 New commit: 4e6776a Julia version: 1.11.0-DEV SparseArrays version: 1.11.0 Bump invoked by: @KristofferC Powered by: [BumpStdlibs.jl](https://github.com/JuliaLang/BumpStdlibs.jl) Diff: https://github.com/JuliaSparse/SparseArrays.jl/compare/54f4b39d1cd34a908ff4fdfcd052a38fe23f6555...4e6776a825f2a26c3c580f9b77ba230a6598d7dd ``` $ git log --oneline 54f4b39..4e6776a 4e6776a faster cat performance (#432) ``` Co-authored-by: Dilum Aluthge <dilum@aluthge.com> 01 September 2023, 18:50:41 UTC
3527213 Jameson Nash 01 September 2023, 18:50:26 UTC simplify call to promote_eltype with repeated elements (#51135) Helps to short-circuit calls to large splat calls, since those have all the same type elements. Fixes #51011 01 September 2023, 18:50:26 UTC
91b8c9b Cody Tapscott 01 September 2023, 18:46:25 UTC Add `JL_DLLIMPORT` to `small_typeof` declaration (#50892) Resolves #50714 01 September 2023, 18:46:25 UTC
631b5c3 Jameson Nash 01 September 2023, 14:14:08 UTC inference: fix bad effects for recursion (#51092) Effects are not converged, so they need to always be correct, and not a bestguess, even during recursion. This could be refined, but we don't really need to, and it might be unnecessarily costly to do so. 01 September 2023, 14:14:08 UTC
befe6f8 Jameson Nash 01 September 2023, 14:13:39 UTC reflection: make error message clearer when using methods wrong (#51134) Not really a functional change, but just improving the error message here to be consistent with other code reflection calls here. Closes #49982 01 September 2023, 14:13:39 UTC
0f4a54c Keno Fischer 01 September 2023, 04:29:10 UTC Add a codegen option to disable use of jlplts (#51123) Alternative to https://github.com/JuliaLang/julia/pull/51108 with the same objectives. 01 September 2023, 04:29:10 UTC
7481fbe Keno Fischer 01 September 2023, 04:28:38 UTC Add options to jl_dump_native (#51124) I am attempting to more or less unify the codepaths between sysimage-like binary emissions and StaticCompiler-like binary emissions (the latter currently uses a GPUCompiler-derived pipeline), with a view towards enabling things that are somewhat in the middle (may have parts, but not all of the runtime, etc.). To that end, add a parameters struct to jl_dump_native that will likely grow over time, but for now, the primary thing it controls is whether or not to emit the sysimage "metadata" section. 01 September 2023, 04:28:38 UTC
77439e5 Shuhei Kadowaki 31 August 2023, 23:37:19 UTC make `code_llvm` work with OC with tuple input types (#51128) 31 August 2023, 23:37:19 UTC
69392ee Shuhei Kadowaki 31 August 2023, 23:35:43 UTC make `code_llvm` work with custom lookup function (#51127) 31 August 2023, 23:35:43 UTC
36fc9ec Nathan Daly 31 August 2023, 23:07:41 UTC Add devdocs for how to set up & run the LLVM tests locally (#51094) I had forgotten the command to install the llvm dev tools so that I could run the tests, and so I documented it this time, once I figured it out. 31 August 2023, 23:07:41 UTC
e0f5247 Lilith Orion Hafner 31 August 2023, 21:40:36 UTC Refactor rounding (including bugfixes) and document API (#50812) 31 August 2023, 21:40:36 UTC
a3e2316 Jeff Bezanson 31 August 2023, 18:57:50 UTC inference: type bound error due to free typevar in sparam env (#51013) Fix #50709 This issue *appears* fixed on master due to #50432, which simply removed the error. This fixes the underlying cause, which is that we sometimes return typevars in the environment from intersection that have free vars in their bounds. I think it's reasonable to just widen these aggressively, since we usually cannot do much with these kinds of unknown static parameters anyway. 31 August 2023, 18:57:50 UTC
15f7db8 Jameson Nash 28 August 2023, 20:55:48 UTC inference: fix bad effects for recursion Effects are not converged, so they need to always be correct, and not a bestguess, even during recursion. 31 August 2023, 15:46:01 UTC
00c6ea7 Jameson Nash 30 August 2023, 21:40:17 UTC inference: teach tfunc that typeof tuples must have fixed lengths 31 August 2023, 15:44:59 UTC
8307dbf Jameson Nash 30 August 2023, 16:23:08 UTC improve effects of typejoin and simplify implementation of Tuple _compute_eltype - Repair definition of Core._foldable_meta - Handle Vararg better in eltype - Mark has_free_typevars ccall as total - Mark tailjoin foldable explicitly also, since it calls typejoin recursively - Handle non-constructable Tuple{:a,2} and Tuple{T} types - Prepare special code for inlining (improves generic effects too) - Only optimize typejoin in concrete eval, not needlessly generating specializations otherwise on specific types The compiler seems better able to deal with this version of _compute_eltype, and it does not seem that necessary to form the IdSet. 31 August 2023, 15:44:59 UTC
479743e Elliot Saba 31 August 2023, 15:32:56 UTC Consistently report sysimg size limits (#51119) We used to be reporting the maximum system image size in hexadecimal, while reporting the image that was just built in decimal. This is a recipe for confusion when the size limit contains no blatantly hexadecimal characters. This PR addresses the issue by consistently printing out both values in hexadecimal (making this more obvious by prefixing the numbers with `0x`). In order to use `PRIxPTR` in both format strings, I have cast the current image size to `uintptr_t` from `intmax_t`, which should be safe in all situations. Co-authored-by: Dilum Aluthge <dilum@aluthge.com> 31 August 2023, 15:32:56 UTC
e118207 Keno Fischer 31 August 2023, 07:08:32 UTC Add a codegen option to disable use of jlplts Alternative to #51108. 31 August 2023, 07:10:20 UTC
a355403 Keno Fischer 31 August 2023, 06:41:11 UTC Annotate fieldnames for default cgparams [NFC] 31 August 2023, 07:10:14 UTC
95af5a0 Keno Fischer 31 August 2023, 03:12:40 UTC Refine effects based on optimizer-derived information (#50805) The optimizer may be able to derive information that is not available to inference. For example, it may SROA a mutable value to derive additional constant information. Additionally, some effects, like :consistent are path-dependent and should ideally be scanned once all optimizations are done. Now, there is a bit of a complication that we have generally so far taken the position that the optimizer may do non-IPO-safe optimizations, although in practice we never actually implemented any. This was a sensible choice, because we weren't really doing anything with the post-optimized IR other than feeding it into codegen anyway. However, with irinterp and this change, there's now two consumers of IPO-safely optimized IR. I do still think we may at some point want to run passes that allow IPO-unsafe optimizations, but we can always add them at the end of the pipeline. With these changes, the effect analysis is a lot more precise. For example, we can now derive :consistent for these functions: ``` function f1(b) if Base.inferencebarrier(b) error() end return b end function f3(x) @fastmath sqrt(x) return x end ``` and we can derive `:nothrow` for this function: ``` function f2() if Ref(false)[] error() end return true end ``` 31 August 2023, 03:12:40 UTC
81b5a9e Shuhei Kadowaki 31 August 2023, 02:19:40 UTC allow `code_[llvm|native]` to take custom `params::CodegenParams` (#51109) 31 August 2023, 02:19:40 UTC
c659011 Cody Tapscott 31 August 2023, 01:58:57 UTC Workaround upstream FreeBSD issue #272992 (#51114) ELF doesn't handle WEAK symbols dynamically the way it handles them statically. Looking up overloaded WEAK symbols via a library-specific handle will often give you the empty stub (in `libc.so.7` in this case) instead of the strong implementation elsewhere (`ld-elf.so.1` here). Even after the [upstream fix](https://cgit.freebsd.org/src/commit/?id=21a52f99440c9bec7679f3b0c5c9d888901c3694), `dlsym`, `dladdr` and a ton of other symbols still have stubs with no trampoline in FreeBSD's libc: https://cgit.freebsd.org/src/tree/lib/libc/gen/dlfcn.c?id=21a52f99440c9bec7679f3b0c5c9d888901c3694 Thankfully `dl_iterate_phdr` appears to be the only function that we directly `ccall` from Julia's Libdl so we can leave this fix incomplete for now. Resolves #50846. 31 August 2023, 01:58:57 UTC
743ed16 Gabriel Baraldi 31 August 2023, 00:40:44 UTC Make threadcall a bit more threadsafe (#51040) Threadcall currently assumes that the IO loop gets run in the same thread that spawned it. While I wish to supercede threadcall by something that uses tasks + adopt_thread. For now lets make it a better. ```julia julia> function foo() Threads.@threads for i = 1:8 ptr = @threadcall(:jl_malloc, Ptr{Cint}, (Csize_t,), sizeof(Cint)) @test ptr != C_NULL unsafe_store!(ptr, 3) @test unsafe_load(ptr) == 3 ptr = @threadcall(:jl_realloc, Ptr{Cint}, (Ptr{Cint}, Csize_t,), ptr, 2 * sizeof(Cint)) @test ptr != C_NULL unsafe_store!(ptr, 4, 2) @test unsafe_load(ptr, 1) == 3 @test unsafe_load(ptr, 2) == 4 @threadcall(:jl_free, Cvoid, (Ptr{Cint},), ptr) end end foo (generic function with 1 method) julia> foo() ``` This crashes julia currently. 31 August 2023, 00:40:44 UTC
e3e1f1e Gabriel Baraldi 31 August 2023, 00:31:02 UTC Swap Base.Event to use a ThreadSynchronizer (#51093) and make some of its other fields const. 31 August 2023, 00:31:02 UTC
6f026e3 Lilith Orion Hafner 30 August 2023, 22:06:58 UTC Refactor scalar range getindex (#50467) 30 August 2023, 22:06:58 UTC
fb76136 Lilith Orion Hafner 30 August 2023, 19:38:21 UTC Fix edge cases where inexact conversions to UInt don't throw (#51095) 30 August 2023, 19:38:21 UTC
8a5a872 Gabriel Baraldi 30 August 2023, 18:34:01 UTC Remove spurious iostream in codegen (#51104) 30 August 2023, 18:34:01 UTC
476572f Jameson Nash 30 August 2023, 18:32:23 UTC makefile option to generate better code (#51105) Disables an "optimization" for -O2 and higher. We already set this for JIT code (NO TOUCHIE!), and we should probably set this for all other code too. It is common otherwise for stacktraces to have the wrong line number for errors. 30 August 2023, 18:32:23 UTC
197180d Yi Lin 30 August 2023, 14:03:28 UTC Add GC write barrier in jl_reserve_excstack (#51096) `jl_push_excstack` and `jl_reserve_excstack` take an address to the field `excstack` in `jl_task_t`, and may update the field. In such a case, a GC write barrier is needed. This PR adds the missing write barrier. Note: I am not sure if this affects the correctness of Julia GC or not. This could be special cased in the GC code, as the task is always rooted. However, generally if the task is in the old generation, it does not have to be scanned even if it is a root. A write barrier is needed to make sure the GC is aware of the update. 30 August 2023, 14:03:28 UTC
f9792b4 Jishnu Bhattacharya 30 August 2023, 13:56:13 UTC docs: simplify array creation in `sortslices` docstring (#51101) Using concatenation instead of `permutedims(reshape(...))` makes the construction easier to read and understand. 30 August 2023, 13:56:13 UTC
374563e Aravindh Krishnamoorthy 30 August 2023, 07:57:06 UTC Speed up `logdet` for diagonal and triangular matrices (#50950) 30 August 2023, 07:57:06 UTC
a271349 Jishnu Bhattacharya 30 August 2023, 07:41:51 UTC Simplify the docstring for Diagonal (#51097) 30 August 2023, 07:41:51 UTC
4e1c965 Jishnu Bhattacharya 30 August 2023, 07:38:15 UTC Remove size(::StructuredMatrix, d) specializations (#51083) 30 August 2023, 07:38:15 UTC
a84882e Shuhei Kadowaki 29 August 2023, 08:11:57 UTC effects: taint overlay-ed method's `:nonoverlayed` effect bit (#51078) 29 August 2023, 08:11:57 UTC
8ace9c5 Shuhei Kadowaki 29 August 2023, 00:49:40 UTC reflection: fix up handling of `Core.Compile.findall` (#51081) `Core.Compiler.findall(sig, ::MethodTableView)` may return `nothing` but never returns `missing`. Also fixes up the docstring of `Core.Compiler.findsup`. 29 August 2023, 00:49:40 UTC
0b29986 Cody Tapscott 28 August 2023, 21:15:45 UTC Always rewrite GotoIfNot with unreachable branches (#51062) This resolves a regression introduced in https://github.com/JuliaLang/julia/pull/50943#issuecomment-1694198019 That PR requires the Goto/GotoIfNot statements of the IR to correspond 1-1 (in terms of reachability) to the information we get from inference. To make that happen, we have to unconditionally re-write control flow to match the branches that inference ended up actually exploring. The problem is that we were choosing not to do this if the GotoIfNot condition seemed to be maybe-non-Boolean. Thankfully, it turns out that check is unnecessary because Inference when unwrapping conditionals does not consider "true or non-Bool" etc. If it did, we'd instead have to re-write these branches as a `typeassert; goto` to encode the reachability 28 August 2023, 21:15:45 UTC
8dc69aa Zentrik 28 August 2023, 20:36:56 UTC Fix typo in `postdominates` docstring (#51089) 28 August 2023, 20:36:56 UTC
1d56cdf Shuhei Kadowaki 28 August 2023, 19:19:27 UTC minor tweaks on `Base.[@constprop|@assume_effects]` docs (#51079) 28 August 2023, 19:19:27 UTC
4d3de1a Cody Tapscott 26 August 2023, 17:46:36 UTC Always rewrite GotoIfNot with unreachable branches This resolves a regression introduced in https://github.com/JuliaLang/julia/pull/50943#issuecomment-1694198019 The problem was that we now expect the explicit CFG of the IR to correspond 1-1 in terms of reachability to the information we get from inference. That means that we have to unconditionally re-write control flow to match the branches that inference ended up actually exploring. This change also modifies Inference to update the ssaflags on GotoIfNot and GotoNode statements, so that we can be sure these `@assert`s will trip if Inference is ever made smart enough to fold Union{Const(true), Float64}-style conditions. 28 August 2023, 17:26:09 UTC
b4052a5 Jishnu Bhattacharya 28 August 2023, 15:33:19 UTC Forward `axes` to the parent for a `Diagonal` (#50514) Given that [quite a few](https://juliahub.com/ui/Search?q=axes%5Ba-zA-Z%5C%28%5C%29%3A%5D%2BDiagonal&type=code&r=true) packages define `axes(d::Diagonal{T, CustomVector{T}})` as `ax = parent(d); (ax,ax)`, perhaps it makes sense to have this defined in `LinearAlgebra`. After this, ```julia julia> using StaticArrays, StructArrays, LinearAlgebra julia> D = Diagonal(StructArray{Complex{Int}}((SA[1,2], SA[1,2]))) 2×2 Diagonal{Complex{Int64}, StructVector{Complex{Int64}, @NamedTuple{re::SVector{2, Int64}, im::SVector{2, Int64}}, Int64}} with indices SOneTo(2)×SOneTo(2): 1+1im ⋅ ⋅ 2+2im julia> axes(D) (SOneTo(2), SOneTo(2)) ``` The static sizes are preserved. --------- Co-authored-by: Daniel Karrasch <daniel.karrasch@posteo.de> 28 August 2023, 15:33:19 UTC
back to top