Age | Commit message (Collapse) | Author |
|
|
|
|
|
llvm15 got dropped, but the julia flavoured llvm based on 15.x got added as replacement.
Thanks to MarsSeed for the hint.
|
|
which hopefully allows for easier tracking/comparison.
Huge changes:
* drop the split package (maybe only for the time being)
* switch from master (1.11-DEV) to release-1.9 (for the time being)
* drop the separate file for the build options
Issues: running check is kinda finicky. Suitesparse related issues aren't resolved.
|
|
|
|
libunwind, pcre2 and mpfr (not to use some more of the system libs,
as occurred arch system versions got out of sync against julia choice)
- make test instead of testall.
`make test` does currently the following bit less than the
preceding command in check() correspondent to `make testall`
LibGit2/online
Pkg
the difference is given by the argument `default` or `all` to runtests.jl
See diff <(make -n test) <(make -n testall)
then add argument --help-list at the invocation of runtests.jl
to list and compare the tests run by `default` or `all` flavor
consider running some `make` command with -j1 argument in low memory
machines (the minimum seems to be in the order of 12Gb+ of available
memory and 20Gb of Swap)
- while testing this change the failure of test for
pcre2 and mpfr is just due to a version-string comparison,
'pcre2=10.36.0' current system version would be 10.39
'mpfr=4.1.0' a string version check refuses current version 4.1.0-p13
I let julia build internally such libraries too
the official changes description of pcre are in
https://github.com/PhilipHazel/pcre2/blob/master/ChangeLog
Also the attempt to prepare the deps source files with
`make full-source-dist` (now commented out) fails because
since at least November 4th 2020 the used url in pcre.mk,
ftp.pcre.org is "not anymore" online (as said in https://pcre.org)
I opened julia development issue https://github.com/JuliaLang/julia/issues/43895
- Remove an obsolete cp Make.user, because being specified in the
sources, an absolute symbolic link is created in src dir
- Removed a comment mentioning USE_SYSTEM_ARPACK, not anymore used
- Update dependency also with version range according to
https://github.com/JuliaLang/julia/blob/master/doc/src/devdocs/build/build.md#required-build-tools-and-external-libraries
if you have enough RAM there is chance that the build succeeds
|
|
|
|
less system depends: OpenBLAS/cblas/lapack,GMP,SuiteParse,mbedTLS,libssh2,curl,libgit2
so succeed building all tests (except Distributed which is still skipped)
skip hash detail for source Make.user (it's not downloaded externally from the repo)
OpebBLAS system provided failed test on a computational check, returning - instead of +
so I put aside the effort to use system OpenBLAS and patch it,
making the packaging smaller
using julia provided OpenBLAS instead of the system OpenBLAS prevents the following error:
Expression: #= /home/mar/aur/julia-git/src/julia/usr/share/julia/stdlib/v1.7/LinearAlgebra/test/lu.jl:287 =# @inferred(logdet(ComplexF32[1.0f0 0.5f0; 0.5f0 -1.0f0])) === 0.22314355f0 + 3.1415927f0im
remove patches not enrured necessary while encountering abundant test errors,
make-install-no-build.patch omitted a second build with possibly
a finalized environment, I rather prefer to let a second run for the moment
otherwise this patch intended for efficiency should be broguth to Julia source,
compilers are also designed not to produce the output when the input is older,
if makefile order the build and it does build I expect its's for a reason,
but I didn't search for it
|
|
add also a comment with a command to extract depends information
grep =1 Make.user|cut -c 11-|cut -d: -f1|tr _A-Z \\ta-z
Note that this is inconsistent change in produced binary
and is in preparation of precise and clear maintainance
of dependencies, however depends isn't currently in sync
with what specified in Make.user
|
|
this is an intermediate commit before a complete fix
julia Makefile needs to built contrib/stringreplace.c in usr/tools/
previously skipped with make-install-no-build.patch
patchelf would be provided with julia/deps/ to run
patchelf --set-rpath '$ORIGIN:$ORIGIN/..' pkg/julia-git/usr/lib/julia/libjulia-internal.so
but is currently specified system provided
in Make.user: USE_SYSTEM_PATCHELF:=1
Note:
check() can be commented out in PKGBUILD to obtain a package,
currently the following tests shown to report errors:
LinearAlgebra/lu
read
stacktraces
GMP_jll
InteractiveUtils
LibCURL
LibGit2_jll
LibSSH2_jll
MbedTLS_jll
MozillaCACerts_jll
NetworkOptions
OpenBLAS_jll
REPL
SuiteSparse_jll
precompile
then keeping the OpenBLAS patch but with USE_SYSTEM_BLAS:=1 only
and all the others USE_SYSTEM_ to 0, test with error remains:
LinearAlgebra/lu
OpenBLAS_jll
SuiteSparse_jll
(the three related to OpenBLAS)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Incorporate changes from the repo PKGBUILD, which changes deps.
System llvm cannot be used currently, as they use a patched version. Until those patches are upstreamed nothing can be done.
|
|
Apparently the same situation as with the 0.6.x release
|
|
USE_LLVM_SHLIB: Yeah, dunno.
USE_BLAS64: Should be handled perfectly by the detection system
|
|
Isn't included in the base package anymore.
If someone wants to use it, they need the fftw.jl package
|
|
Replaced it internally and got rid of this external dependency for upcoming 0.7 release
|
|
|
|
Imo a nice way to declare openblas as an alternative for those via optdepends
|
|
multiline for oversight
|
|
Now corrected
|
|
Currently the build were failing, because some new deps got included, and the default was used: Don't use the system ones.
Building the new deps required cmake as a makedep.
The new deps are:
* curl: a dep of pacman, which is a base package, so satisfied
* libssh2: Already satisfied due to curl and libgit2
* mbedtls: A new dep
Another build failure that occured was because of missing deps.
Apparantly the depends lines from the package functions aren't checked, so the vast makedepends is needed (And I really thought it changed, but well.. Dunno why I thought that)
|
|
makedeps and deps will be installed before building.
Building just one package from a SplitPKGBUILD was removed from makepkg.
So it should suffice, if julia main package get the deps, makedeps stripped down to real makedeps.
This removes the weird makedeps of the doc package
|
|
Stuff done there is handled by hooks
|
|
|
|
I screwed really fine the PKGBUILD up, when I compared it with the Community one.
This is now fixed.
Additionally building on i686 should be working again (failed for me)
Thanks to yuyichao for pointing out a possible solution[0]
______
[0] https://github.com/JuliaLang/julia/issues/13913
|
|
|
|
Various reasons.
* It's the prefered version to customize the build process.
* It removes duplicate lines.
* And it will work again. Could just be a case of the newly added :, but before that all were set to USE_SYSTEM = 0
|
|
|
|
|
|
|
|
To remove some false positives if comparing with the PKGBUILD of the community/julia
|
|
|
|
Adjust deps in the commented makedeps line and point to aur/juliadoc-git package.
But if it's missing it will be downloaded by the julia build process.
Remove the echo statements, include the html pages in the doc package and remove the double man-page
|
|
Removing the old counter made the problem even worse. Now all revs will be counted (not depending on the tag) so this problem shouldn't occur.
Thanks to haawda for the solution
|
|
This thing is misleading with the branching system used by the julia devs.
And use the correct $branch:VERSION information if another branch than master is used (like release-0.4 and co)
|
|
|
|
|
|
Both from the AUR
|
|
History: https://github.com/Narrat/PKGBuilds
|