Julia 1.13 Ships as Stable Release With a 30% Faster Precompiler, REPL Syntax Highlighting, and a Juliaup GUI
Julia 1.13 shipped September 10 with roughly 30% faster package precompilation, built-in REPL syntax highlighting, and a new graphical interface for the Juliaup version manager.
Overview
The Julia programming language has shipped version 1.13 as a stable release. The GitHub release page tags it as “Julia version 1.13.0, the thirteenth minor release in the 1.x series of releases,” with the tag published on September 10. LWN.net independently confirmed the release the same day, reporting that “Highlights include faster precompilation of packages, improvements to Julia’s REPL, and Juliaup, a graphical interface for the Julia version manager.”
The release follows a beta and release-candidate cycle that began in January, as previously reported, when the project shipped its first 1.13.0 release candidate on April 29.
What We Know
Faster precompilation and startup. According to the Julia 1.13 Highlights post from “The Julia contributors,” the release “takes roughly 30% less time to precompile packages than 1.12, and roughly 10-20% less time than 1.10 (LTS) depending on the machine.” The same post says “Julia 1.13 startup is also ~20% faster than 1.12,” backed by a benchmark showing Julia 1.13 running “1.22 ± 0.02 times faster” than 1.12 in a startup-time comparison.
REPL overhaul. The Julia 1.13 Highlights post says “The Julia REPL now has syntax highlighting (without having to load an external package like OhMyREPL.jl).” History search, triggered by Ctrl-R, “has been redesigned and now works similarly to the command-line fuzzy finder fzf.” Bracketed paste — which lets a terminal application distinguish pasted text from typed text — “has been enabled on Linux and macOS for a long time but is now also finally available on Windows,” per the same post.
Language and hashing changes. The official NEWS.md release notes list a “New @__FUNCTION__ macro to refer to the innermost enclosing function” and “Support for Unicode 17.” The blog post adds that @__FUNCTION__ “references the innermost containing function even if that function is anonymous” and “is public API, unlike the internal variable #self#.” Separately, the NEWS.md notes warn that “The hash algorithm and its values have changed for certain types, most notably AbstractString,” and that “Any hash specializations for equal types to those that changed, such as some third-party string packages, may need to be deleted.” The highlights post says the new byte-hashing algorithm is “RapidhashNano,” replacing an implementation “based on MurmurHash3,” and that it “has moved from C to pure Julia for better readability and maintainability.” A benchmark in the post shows hashing a long downloaded text file dropping from “8.555 μs” on 1.12 to “1.742 μs” on 1.13.
Garbage collection and concurrency. The highlights post describes a change where “objects in the sysimage and in package images are loaded as permanently marked and the mark phase never enters them,” so that “the cost of a full collection now scales with the size of the heap that your program actually created, not with the amount of code that has been loaded.” A timing example in the post shows a full collection via GC.gc() in a bare session dropping from “0.035493 seconds” on 1.12 to “0.000528 seconds” on 1.13. On the scheduler side, the post says “Ctrl-C reaches user code again, including scripts blocked in sleep or IO, and Distributed.interrupt works,” and that a change to how @spawn wakes threads means “Spawn-heavy code speeds up anywhere from not at all on macOS, to 1.1-1.6x on a 16-core Linux machine, to 10-300x on Windows and heavily oversubscribed machines.”
Package manager changes. The highlights post says Pkg “will now by default ask for a zstd-compressed archive instead of a gzipped one” when downloading from a package server, citing an example where downloading packages and artifacts for Plots, Makie, and ModelingToolkit fell from “307.99 MB” to “239.31 MB,” with total decompression time dropping from “8.77 s” to “5.50 s.” The post also says the registry each package came from “is recorded in the manifest and is automatically installed upon manifest instantiation,” that pkg> add now “prefers the currently loaded version of any package that is already loaded” when compatible, and that Pkg.test “now leaves the bounds-checking mode alone” rather than forcing --check-bounds=yes by default.
A graphical Juliaup. The version-manager tool Juliaup “now has a graphical interface alongside its command line,” according to the highlights post. It “ships with Juliaup 1.22 and later on every platform Juliaup supports, so after a juliaup self update it can be opened with: juliaup gui.” The post says the interface’s Installed tab lets a user launch, configure, set as default, or remove installed channels, while the Available tab “lists everything in the channel database, including release, lts, rc, nightly and pr{number} channels for testing pull requests.”
What We Don’t Know
The project has not shipped a full task-cancellation mechanism in this release. The highlights post says “Work on a proper task cancellation mechanism is in progress and is planned for Julia 1.14,” leaving the scope and timing of that feature undefined for now. The post also does not quantify how many third-party packages will need updates for the new hashing algorithm, beyond noting that packages with custom hash specializations for affected types “may need to be deleted.”
Analysis
The 1.13 release notes describe the preparation of this release as “partially funded by NASA under award 80NSSC22K1740,” underscoring that Julia’s development draws on institutional research funding alongside community contributions. Taken together, the performance changes in this release — faster precompilation, a lighter garbage collector, and a rewritten hashing algorithm — target the kind of steady-state runtime costs that matter most to Julia’s core audience in scientific and numerical computing, while the REPL overhaul and new Juliaup GUI are aimed more squarely at day-to-day developer experience.