WIP: Improve accuracy of ^ for ill-conditioned matrices #12584

…l's" (JuliaLang#17906) This reverts commit 97f1841. ref JuliaLang#15521

So that we don't need to run `jl_exit` in strange (signal handler) context due to missing exception handler.

Use it to make sure that `jl_rethrow` and `jl_exit` are running on the right thread and right stack when an exception/exit is caused by a signal. Fix JuliaLang#17706

…Lang#17901)

…uliaLang#17909) and Symmetric, it is redundant to also have an uplo argument and occasionally, it also gave the wrong result. This can therefore be considered a bugfix.

Fixes JuliaLang#17780

) * Add cross-release compatibility note to CONTRIBUTING.md * tag -> specify

ref JuliaLang#17911 [ci skip]

* improve heuristic for when to refresh line * fix trailing whitespace... * use indent when not on first line * remove unused argument and change local var to not share name with function

rst apparently requires that the closing double backticks for code literals appear at the end of a "word", rather than preceded by a space as markdown allows. This means that ``` ``julia> `` ``` doesn't parse correctly. The fix is to just remove the trailing space and explicitly mention the need for a space after the `>`.

Transform `x->f(x)` to `f`.

See JuliaLang#16766. `ndigits(big(0),1)` doesn't raise an error, but we were sometimes (randomly, rarely) testing that it does. This change makes the test reliable. The issue of how this case should actually behave is still open.

* Make llvm-gcroot a ModulePass * Use llvm-gcroot to calculate the minimum number of exception frames necessary.

…7896)

fix output of getaddrinfo in networking and streams documentation for v0.5.

The PR fixes jl_static_show for bitstypes

(cherry picked from commit 84ad7c7) [ci skip]

also fix some code formatting

* Replace a malformed link in `write` docstring with a literal `write` rather than a real cross-reference, since that would just cross-reference itself. Also move `write` docstring inline. * Remove cross-reference for `<<` from inside a literal since text inside a literal can't contain a link. A cross-reference to `<<` already appears at the end of the docstring.

Vararg is only exact if in covariant position also make getfield_tfunc monotonic for the case where the type has one field, to avoid the same bug fix JuliaLang#16530

required for the inliner to work, since all of the logic for computing edges is supposed to be handle by the inference step fix JuliaLang#17759

defining MEMFENCE will tell the gc to validate all object tags (before they get freed) to attempt to detect out-of-bounds writes that corrupted a object tag or the freelist

It turns out we were returning a Vector{Int} for operations that should return a Range

…mitian (JuliaLang#17909)" This reverts commit d992f3d.

and Symmetric, it is redundant to also have an uplo argument and occasionally, it also gave the wrong result. This can therefore be considered a bugfix.

…g#17910] (JuliaLang#17955)

…uliaLang#17959)

Also update out-of-date comment about codegen steps

* Fix atahn to atanh * More tests

parse("[1] [2]") should throw a ParseError

This fixes a performance regression in the lucompletepiv benchmark triggered by generalizing _array_for to work with indices. This call, first introduced in b363cc7, likely always triggered dynamic method dispatch. The generalization to indices seemed to make that worse.

…non-Pair return type

Fixes JuliaLang#17819. Also document not being allowed to call back into Julia.

Fix JuliaLang#17908

This header prevents me from building Julia on musl libc. ``` /home/alba/julia/src/sys.c:16:24: fatal error: sys/sysctl.h: No such file or directory compilation terminated. Makefile:94: recipe for target sys.o failed make[1]: *** [sys.o] Error 1 ```

* fix a typo in Base.split * fixup! fix a typo in Base.split

that have non-Int indices, introduced by JuliaLang#13612 and which has caused `make -C test/perf` to fail for the last 9 months searchsortedfirst does not have methods for general Integer indices

…d no other associative type implement the method.

Fix JuliaLang#18020

fix JuliaLang#18015

There seem to be a few variations on the tools and arguments available for creating split debuginfo files. I used the following: objcopy --only-keep-debug libjulia.so libjulia.so.debug strip -g libjulia.so objcopy --add-gnu-debuglink=libjulia.so.debug libjulia.so

This should help JuliaLang#17932. If a function returns an unanticipated type that wasn't part of the `Union` seen by inference, do dynamic dispatch instead of erroring.

Update to add type annotation otherwise we get ``` ERROR: MethodError: no method matching display(::MyType) ```

* added example of array type conversion ... because it seems to bite less experienced users like myself. (Stolen from Terry Seaward's email, https://groups.google.com/d/msg/julia-users/wzaJw_FfNlA/qcuZ4FggEL4J) * added example of array type conversion

the getfield_tfunc was missing a test for whether the fields were all equivalent before concluding that the result type was exact fixes the fix JuliaLang#17953 fix JuliaLang#18037

Fixes the doc problem mentioned in JuliaLang#17926, but not the underlying problem.

The function might get finalized, invalidating the IP. However, in some cases this invalid IP may get saved and restored, accessing the invalid IP while doing so. Example code path accessing an invalid IP: -> jl_cfunction_object (nested_compile=true, but doesn't change IP) -> gen_cfun_wrapper -> jl_compile_linfo (saves and restores invalid IP)

fixes regression caused by JuliaLang#18017

Caused by the return type of `ord()` being unknown.

Note that this test requires a memory sanitizer (ASAN, valgrind) to detect failure.

…)` etc.

An SSH URL with a path uses a colon between the host and the path: git@github.com:JuliaLang/Example.jl (correct) git@github.com/JuliaLang/Example.jl (incorrect)

This mostly reverts JuliaLang#17906 and puts JuliaLang#15521 back in place, but from a slightly more permanent download location now. Unfortunately using the latest opensuse copy of the gcc dll's is causing issues when a cygwin-built Julia tries to load an opensuse-built libzmq.dll. Using these slightly old gcc 5 versions seems to fix it. I suspect the difference is which libstdc++ ABI is being used. GCC 5 has both old and new available, but many distros had it using the old ABI by default for compatibility. GCC 6 (which opensuse is using now) is more likely to be using the new ABI by default.

Test was only working for environments which had GitHub SSH keys setup.

Added a new explicit SSH protocol test which generated other changes.

* Add missing tests for diagonal and uniformscaling * Change A_mul_B to * * Moved things out of nested loop that don't need to be there

most users of static-eval didn't need most of it's complexity and didn't actually want its deprecation warning behavior this also will make it easier to delete this function as the users of it get fixed not to need it

This link has moved.

* Set LIB_INSTALL_DIR for cmake-based projects This fixes `MULTIARCH_INSTALL=1` installations that place libraries not into `<prefix>/lib` but into `<prefix>/lib/<host-triplet>`, like Debian and Ubuntu. Without this change, `libgit2`, `libssh2` and `mbedtls` would place libraries into `<prefix>/lib`, silently failing until the Julia bootstrap process attempted to dlopen one of them. * Apply `LIB_INSTALL_DIR` to `CMAKE_COMMON`

[ci skip]

* Allow `cpu_target` to specify a generic arch, matching the behavior on x86 * Detect the CPU arch version with `uname` * Require `armv6` Close JuliaLang#13270 (`armv5` is not supported) Fix JuliaLang#18042 * Remove warning about generic arch since it's not really useful Fix JuliaLang#17549 * Require at least the same ARM arch version and profile the C code is compiled with

[ci skip]

jl_get_llvmf was left in so that this can be backported to v0.5 without altering the C api (in case it was used in a package) also correct / add error handling and proper locks

When saving native code we re-run optimization on the whole module as part of object file emission, so it is not needed for JIT compiled code. We are already passing -O0 for package precompilation; this extends that benefit to the system image.

Link to our own libssh2.

Closures that are part of out-of-scope method adds were not lifted out of top-level loops.

…uest (JuliaLang#18061) * make dlopen more conservative about opening files the user didn't request * update manual * recognize \server\foo and /foo/bar absolute paths on windows, similar to isabspath * simplify isabspath from init.c and use it in dlopen instead of duplicating

…caling (JuliaLang#17950)

Fixes precision problem for rational powers of BigFloats (JuliaLang#18114)

* Make similar faster and safer By converting the eltype argument into a ::Type{T} parameter, we avoid runtime method lookup. Perhaps more importantly, the introduction of NeedsShaping makes `similar` safer, by preventing any possibility of an infinite recursion. * Test that similar throws a MethodError for unsupported dims types

The `::` syntax appearing before the `doctest::` block was causing the doctest not be rendered correctly.

fix JuliaLang#18073: `any([])` and `all([])` (JuliaLang#18105) (cherry picked from commit d259be5)

…ted (JuliaLang#17607) out in JuliaLang#17083

This simplifies the array bounds check code that is emitted for multi-dimensional array accesses that use "regular" indexing, i.e. accesses of the form `A[i1,i2,...,iN]` to some `N`-dimensional array `A`. For example, with this change, the access `A[i,j,k]` to an array with the three dimensions `m`, `n` and `o` now leads to bounds checks that correspond to the following pseudo code: ``` if (i >= m) out_of_bounds_error(); else if (j >= n) out_of_bounds_error(); else if (k >= o) out_of_bounds_error(); ``` So far, the following more complicated bounds checks would have been emitted: ``` if (i >= m) out_of_bounds_error(); else if (j >= n) out_of_bounds_error(); else if (((k * n + j) * m + i) >= m * n * o) out_of_bounds_error(); ``` Julia also allows one-dimensional and "partial" linear indexing (see JuliaLang#14770), i.e. the number of indices used to access an array does not have to match the actual number of dimensions of the accessed array. For this case we still have use this old scheme. One motivation for this change was the following: expressions like `((k * n + j) * m + i)` are non-affine and Polly would not be able to analyze them. This change therefore also facilitates Polly's bounds check elimination logic, which would hoist such checks out of loops or may remove them entirely where possible.

probably addresses JuliaLang#18106

causing the cache to get re-uploaded

and build tests

should fix clean build of master indent line continuations

since it requires additional dev headers fix JuliaLang#18101

fix JuliaLang#17884

in CMAKE_COMMON, so libssh2 installs without an x86_64-linux-gnu multiarch prefix

since we don't need it Go back to just copying the mbedtls dll even in cygwin-cross, otherwise it puts .dll.a files that we don't need alongside the dlls (should probably fix this upstream)

since otherwise it puts pkg-config files in LIB_INSTALL_DIR which we don't need - if mbedtls separates runtime from library destinations then we could separate BIN_INSTALL_DIR from LIB_INSTALL_DIR or convince libgit2 to use the standard GNUInstallDirs names

and remove some unnecessary blank lines [ci skip]

* Add test for 15913 and 15896 * Change name of test

…`ENV`. (JuliaLang#18144) * Eliminate trailing null from iterator over `ENV`. Fixes JuliaLang#18141. * Add failing test for JuliaLang#18141.

fixes JuliaLang#16231

As mentioned in JuliaLang#18041 (comment) docstrings attached to aliases of functions should be avoided where possible.

Fixes JuliaLang#7022. In particular, these simplify testing for warnings.

Passing `--polly=no` will cause `@polly` declarations to be ignored. This facilitates debugging or evaluating performance differences between using/not using Polly without having to manually remove `@polly` declarations from functions.

of cmake projects with gcc - CC_ARG and CXX_ARG are getting double processed somewhere, so cmake is trying to call the compiler as '/usr/bin/gcc gcc' add cut -s

Complex int is GNU extension that Intel compiler does not support currently. JuliaLang#17145

* Fix codegen of `SSAValue` whose assignment appears later than the user in syntactic order due to backward branches. * Disallow more expression types in value position * Use improved debug info metadata in coverage and allocation tracking too

this works to avoid having `Expr(:invoke)` creating unintentional copies of LambdaInfo objects when they show up in the system image fix JuliaLang#18184

ensure that we are using consistent types and introduce ProfileFormat to contain all of the user options

the min-count argument to Profile.print can be used to filter out lines from the display that are insignificant

the noise-floor argument to Profile.print can be used to filter out lines from the display that are insignificant, based on the heuristic `n <= noisefloor * sqrt(N)`

…_list many types are easy to compute during deserialization, and don't need the full complexity of the flagref list to handle them some of these are easy to detect ahead-of-time, saving a small amount of deserialization effort

fix JuliaLang#15048

Fixes a crash while loading ASTInterpreter

* Preserve meta node and line number info during dead code elimination * Insert `:push_loc` and `:pop_loc` in pairs during lowering Fix JuliaLang#16578

Fixes JuliaLang#18213.

fix JuliaLang#18069

this gives the caller more control over when an how errors get printed

…pile caches this ensures that we only recompile a dependency if either: it can be loaded into the current session, or the user explicitly requests it (via reload) thereby reducing the occurrences and improving the accuracy of the "uuid didn't match" error message :)

fix JuliaLang#18115

fix JuliaLang#18222

* improve string docs to define String etc * tweak

Fixes JuliaLang#11249 to follow the CommonMark spec for lists which does not require a blank line prior to the start of a list.

…JuliaLang#17873) This makes it possible to solve problems with semidefinite B via explicit shift.

The permissions specified in MapViewOfFile specify the maximum permissions the file can ever have, which means that even permissions bypass by the debugger is not allowed to write it. Instead, set the maximum permissions to RWX and VirtualProtect down to the permissions we need. This is the same behavior as we use on Mac/Linux.

…ractTriangular. Remove tests of those convert methods. (JuliaLang#17723)

ref staticfloat/homebrew-julia@f57dcc2#diff-633fd9f8448015391714f35e04d39b5bL154 [av skip]

…higher order functions and multiple dispatch to displace eval. Fixes some apparent type instabilities.

…act broadcast syntax, accordingly revise and expand the associated tests, and add deprecations for the vectorized syntax.

Also clean up `julia_const_to_llvm` to avoid boxing due to `getfield`. Fix JuliaLang#18236

Closes JuliaLang#18231.

…omparison (JuliaLang#18234) See JuliaLang#18213.

* lufact for sparse matrix pivot option error base/sparse/umfpack.jl includes the following method definition for lufact `lufact(A::SparseMatrixCSC, pivot::Type{Val{false}}) = lufact(A)` This should likely be `lufact(A::SparseMatrixCSC, pivot::Type{Val{true}}) = lufact(A)` because in lufact pivoting is on by default. The error is shown in the following example ``` A = speye(4) A[1:2,1:2] = [-.01 -200; 200 .001] F = lufact(A,Val{false}) F[:p] ``` which returns ``` julia> F[:q] 4-element Array{Int64,1}: 3 4 1 2 ``` However it should return ``` julia> F[:q] 4-element Array{Int64,1}: 1 2 3 4 ``` because pivoting was turned off. * Added test for JuliaLang#18246 and JuliaLang#18244 * 4 space indent oops my bad * remove unnecessary lufact method * update test for removed lufact method definition

* Revert "add faster done for strings (JuliaLang#18259)" This reverts commit 30ee10b. * add test that failed with JuliaLang#18259

…liaLang#17848) This commit will allow a user to call: convert(Vector, my_abstractvector_of_eltype_Foo) convert(Matrix, my_abstractmatrix_of_eltype_Foo) which is useful because - for the case that the variable is a subarray of the same dimensionality - for consistency with PR JuliaLang#17066

* Fix JuliaLang#18176 (broadcast over mixtures of arrays and numeric scalars) * Fix JuliaLang#17984 (broadcast behavior over empty arrays)

so `JULIA_CPU_CORES=1 ./julia test/runtests.jl core compile` passes after JuliaLang#18150

…anRange) (JuliaLang#18277)

…uliaLang#18262) These were not tested systematically, and are more generally useful to test conversions from AbstractArray.

…Lang#18265)

…ng#18258) * Small build system modifications to make ppc64le compileable * Make things more explicit for big-endian PPC64

just a random thing I noticed

…iaLang#18296)

otherwise JuliaLang#18258 causes "Target architecture mismatch" errors in generic linux binaries

* Avoid aliasing in in UniformScaling*AbstractMatrix ...and remove unnecessary UniformScaling*SparseMatrixCSC methods * Broaden the tests for non-commutative multiplication * Add Quaternion test case for q*[q] and clean up the Quaternion test type

Consistently ignore line numbers and metadata nodes in all cases.

Move unexported `jl_sym_t*` GVs from `julia.h` to `julia_internal.h`

Also collapse redundant debug info when coverage is off. Fix incorrect `Expr(:inbounds, false)` elimination.

…liaLang#18254) * Fix JuliaLang#18189 (Add deprecation of etc. to NEWS) * Remove mentioning of @static, add as a note * Added reference to Handling Operating System Variation section in manual * Changed reference numbers * Change PR number

Also fix a few other compiler warnings.

…g#11481. Dividing by 8 leads to success on the second try in the case of the failure of --lisp on Windows 7.

we really badly need a heuristic for detecting simple methods like this.

add ! suffix to isgraphemebreak! to clarify that the state is changed

…mpact broadcast syntax.

…pact broadcast syntax.

…ompact broadcast syntax.

…pact broadcast syntax.

…compact broadcast syntax.

…oadcast syntax.

…act broadcast syntax.

…t broadcast syntax.

… broadcast syntax.

… compact broadcast syntax.

…of compact broadcast syntax.

…pact broadcast syntax.

…ons and deprecation of `@vectorize_1arg` and `@vectorize_2arg` themselves. Also add cross-referenes between max and maximum, and also between min and minimum.

…aries and enabling broadcast loop fusion.

…18342) * don't throw DomainError for negative integer powers of ±1 * fix replutil test that was expecting 1^(-1) to fail

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

WIP: Improve accuracy of ^ for ill-conditioned matrices #12584

WIP: Improve accuracy of ^ for ill-conditioned matrices #12584

Commits on Sep 5, 2016