- Mutable struct fields may now be annotated as
const
to prevent changing them after construction, providing for greater clarity and optimization ability of these objects (#43305). - Type annotations can now be added to global variables to make accessing them type stable (#43671).
- Empty n-dimensional arrays can now be created using multiple semicolons inside square brackets,
e.g.
[;;;]
creates a 0×0×0Array
(#41618). try
-blocks can now optionally have anelse
-block which is executed right after the main body only if no errors were thrown (#42211).@inline
and@noinline
annotations can now be placed within a function body (#41312).@inline
and@noinline
annotations can now be applied to a function call site or block to enforce the involved function calls to be (or not to be) inlined (#41312).∀
,∃
, and∄
are now allowed as identifier characters (#42314).- Support for Unicode 14.0.0 (#43443).
Module(:name, false, false)
can be used to create amodule
that contains no names (it does not importBase
orCore
and does not contain a reference to itself) ([#40110, #42154]).
- Newly-created Task objects (
@spawn
,@async
, etc.) now adopt the world age for methods from their parent Task upon creation, instead of using the global latest world at start. This is done to enable inference to eventually optimize these calls. Places that wish for the old behavior may useBase.invokelatest
(#41449). - Unbalanced Unicode bidirectional formatting directives are now disallowed within strings and comments, to mitigate the "trojan source" vulnerability (#42918).
Base.ifelse
is now defined as a generic function rather than a builtin one, allowing packages to extend its definition (#37343).- Every assignment to a global variable now first goes through a call to
convert(Any, x)
(orconvert(T, x)
respectively if a typeT
has been declared for the global). This means great care should be taken to ensure the invariantconvert(Any, x) === x
always holds, as this change could otherwise lead to unexpected behavior (#43671). - Builtin functions are now a bit more like generic functions, and can be enumerated with
methods
(#43865).
- Bootstrapping time has been improved by about 25% (#41794).
- The LLVM-based compiler has been separated from the run-time library into a new library,
libjulia-codegen
. It is loaded by default, so normal usage should see no changes. In deployments that do not need the compiler (e.g. system images where all needed code is precompiled), this library (and its LLVM dependency) can simply be excluded (#41936). - Conditional type constraints can now be forwarded interprocedurally (i.e. propagated from caller to callee) (#42529).
- Julia-level SROA (Scalar Replacement of Aggregates) has been improved: allowing elimination of
getfield
calls with constant global fields (#42355), enabling elimination of mutable structs with uninitialized fields (#43208), improving performance (#43232), and handling more nestedgetfield
calls (#43239). - Abstract call sites can now be inlined or statically resolved as long as the call site has a single matching method (#43113).
- Inference now tracks various effects such as side-effectful-ness and nothrow-ness on a per-specialization basis.
Code heavily dependent on constant propagation should see significant compile-time performance improvements and
certain cases (e.g. calls to uninlinable functions that are nevertheless effect free) should see runtime performance
improvements. Effects may be overwritten manually with the
@Base.assume_effects
macro (#43852).
- The default behavior of observing
@inbounds
declarations is now an option viaauto
in--check-bounds=yes|no|auto
(#41551). - New option
--strip-metadata
to remove docstrings, source location information, and local variable names when building a system image (#42513). - New option
--strip-ir
to remove the compiler's IR (intermediate representation) of source code when building a system image. The resulting image will only work if--compile=all
is used, or if all needed code is precompiled (#42925). - When the program file is
-
the code to be executed is read from standard in (#43191).
Threads.@threads
now defaults to a new:dynamic
schedule option which is similar to the previous behavior except that iterations will be scheduled dynamically to available worker threads rather than pinned to each thread. This behavior is more composable with (possibly nested)@spawn
and@threads
loops (#43919, #44136).
- New function
eachsplit(str)
for iteratively performingsplit(str)
(#39245). - New function
allequal(itr)
for testing if all elements in an iterator are equal (#43354). hardlink(src, dst)
can be used to create hard links (#41639).setcpuaffinity(cmd, cpus)
can be used to set CPU affinity of sub-processes (#42469).diskstat(path=pwd())
can be used to return statistics about the disk (#42248).- New
@showtime
macro to show both the line being evaluated and the@time
report (#42431). - The
LazyString
and thelazy"str"
macro were added to support delayed construction of error messages in error paths (#33711).
@time
and@timev
now take an optional description to allow annotating the source of time reports, e.g.@time "Evaluating foo" foo()
(#42431).range
accepts eitherstop
orlength
as a sole keyword argument (#39241).precision
andsetprecision
now accept abase
keyword argument (#42428).- TCP socket objects now expose
closewrite
functionality and support half-open mode usage (#40783). extrema
now accepts aninit
keyword argument (#36265, #43604).Iterators.countfrom
now accepts any type that defines+
(#37747).
- Keys with value
nothing
are now removed from the environment inaddenv
(#43271). Iterators.reverse
(and hencelast
) now supportseachline
iterators (#42225).- The
length
function on certain ranges of certain element types no longer checks for integer overflow in most cases. The new functionchecked_length
is now available, which will try to use checked arithmetic to error if the result may be wrapping. Or use a package such as SaferIntegers.jl when constructing the range (#40382). - Intersect returns a result with the eltype of the type-promoted eltypes of the two inputs (#41769).
- Iterating an
Iterators.Reverse
now falls back on reversing the eachindex iterator, if possible (#43110).
- New macro
@time_imports
for reporting any time spent importing packages and their dependencies (#41612).
- The BLAS submodule now supports the level-2 BLAS subroutine
spr!
(#42830). cholesky[!]
now supportsLinearAlgebra.PivotingStrategy
(singleton type) values as its optionalpivot
argument: the default ischolesky(A, NoPivot())
(vs.cholesky(A, RowMaximum())
); the formerVal{true/false}
-based calls are deprecated (#41640).- The standard library
LinearAlgebra.jl
is now completely independent ofSparseArrays.jl
, both in terms of the source code as well as unit testing (#43127). As a consequence, sparse arrays are no longer (silently) returned by methods fromLinearAlgebra
applied toBase
orLinearAlgebra
objects. Specifically, this results in the following breaking changes:- Concatenations involving special "sparse" matrices (
*diagonal
) now return dense matrices; As a consequence, theD1
andD2
fields ofSVD
objects, constructed upongetproperty
calls are now dense matrices. - 3-arg
similar(::SpecialSparseMatrix, ::Type, ::Dims)
returns a dense zero matrix. As a consequence, products of bi-, tri- and symmetric tridiagonal matrices with each other result in dense output. Moreover, constructing 3-arg similar matrices of special "sparse" matrices of (nonstatic) matrices now fails for the lack ofzero(::Type{Matrix{T}})
.
- Concatenations involving special "sparse" matrices (
- Now uses
textwidth
for formatting%s
and%c
widths (#41085).
- CPU profiling now records sample metadata including thread and task.
Profile.print()
has a newgroupby
kwarg that allows grouping by thread, task, or nested thread/task, task/thread, andthreads
andtasks
kwargs to allow filtering. Further, percent utilization is now reported as a total or per-thread, based on whether the thread is idle or not at each sample.Profile.fetch()
includes the new metadata by default. For backwards compatibility with external profiling data consumers, it can be excluded by passinginclude_meta=false
(#41742). - The new
Profile.Allocs
module allows memory allocations to be profiled. The stack trace, type, and size of each allocation is recorded, and asample_rate
argument allows a tunable amount of allocations to be skipped, reducing performance overhead (#42768). - A fixed duration cpu profile can now be triggered by the user during running tasks without
Profile
being loaded first and the report will show during execution. On MacOS & FreeBSD pressctrl-t
or raise aSIGINFO
. For other platforms raise aSIGUSR1
i.e.% kill -USR1 $julia_pid
. Not currently available on windows (#43179).
RadioMenu
now supports optionalkeybindings
to directly select options (#41576).?(x, y
followed by TAB displays all methods that can be called with argumentsx, y, ...
. (The space at the beginning prevents entering help-mode.)MyModule.?(x, y
limits the search toMyModule
. TAB requires that at least one argument have a type more specific thanAny
; use SHIFT-TAB instead of TAB to allow any compatible methods (#38791).- New
err
global variable inMain
set when an expression throws an exception, akin toans
. Typingerr
reprints the exception information (#40642).
- The code for SparseArrays has been moved from the Julia repo to the external repo at https://github.com/JuliaSparse/SparseArrays.jl. This is only a code movement and does not impact any usage (#43813).
- New sparse concatenation functions
sparse_hcat
,sparse_vcat
, andsparse_hvcat
returnSparseMatrixCSC
output independent from the types of the input arguments. They make concatenation behavior available, in which the presence of some special "sparse" matrix argument resulted in sparse output by multiple dispatch. This is no longer possible after makingLinearAlgebra.jl
independent fromSparseArrays.jl
(#43127).
- The standard log levels
BelowMinLevel
,Debug
,Info
,Warn
,Error
, andAboveMaxLevel
are now exported from the Logging stdlib (#40980).
- Added function
isequal_normalized
to check for Unicode equivalence without explicitly constructing normalized strings (#42493). - The
Unicode.normalize
function now accepts achartransform
keyword that can be used to supply custom character mappings, and aUnicode.julia_chartransform
function is provided to reproduce the mapping used in identifier normalization by the Julia parser (#42561).
@test_throws "some message" triggers_error()
can now be used to check whether the displayed error text contains "some message" regardless of the specific exception type. Regular expressions, lists of strings, and matching functions are also supported (#41888).@testset foo()
can now be used to create a test set from a given function. The name of the test set is the name of the called function. The called function can contain@test
and other@testset
definitions, including to other function calls, while recording all intermediate test results (#42518).TestLogger
andLogRecord
are now exported from the Test stdlib (#44080).
GC.enable_logging(true)
can be used to log each garbage collection, with the time it took and the amount of memory that was collected (#43511).