another go at fixing 1.6 builds #115
Conversation
…nda-forge-pinning 2021.06.09.12.51.26
Julia uses this in `deps/Versions.make`
Per https://github.com/JuliaLang/julia/pull/37618/files#diff-2592b7284deb7b3ee2572fbb4f110a2d868137ba3970d01a7ea3c7d07b808642, Julia 1.6 supports the use of the system `libnghttp2` while `make`-ing.
Also removes some likely to be outdated dependencies. This is based on conda-forge#111 and likely to still fail, but I could not find the logs for that PR anymore. Also, would it be possible to run the entire testsuite (should take ~1h with multiple threads), since that is also likely to catch any issues with third-party dependencies?
simeonschaub
requested review from
abhi18av,
acaprez,
bgruening,
dfornika,
mariusvniekerk and
SylvainCorlay
as code owners
|
Hi! This is the friendly automated conda-forge-linting service. I just wanted to let you know that I linted all conda-recipes in your PR ( |
|
Does anyone know why conda sets |
It seems they prefer tarballs: |
|
Julia cannot find libopenblas: The openblas-feedstock contains the following shared libraries: Per When using OpenBLAS_jll, Julia normally looks for a file named libopenblas64_.so: However, since We may need to add a |
|
Also a bird told me that Julia's other JLLs depend on "libopenblas64_.so" It sounds like we may need a variant or at least a modification of libopenblas-feedstock similar to https://github.com/JuliaPackaging/Yggdrasil/blob/master/O/OpenBLAS/common.jl#L60-L62 |
|
Looks like we have a better shot at using libopenblas-ilp64 which has the "64_" suffix. It is also versioned, however: Why don't they have |
|
For |
Julia needs libopenblas64_.so but this is provided by openblas rather than libopenblas.
Use openblas rather than libopenblas
|
Hi! This is the friendly automated conda-forge-linting service. I wanted to let you know that I linted all conda-recipes in your PR ( Here's what I've got... For recipe:
|
|
Hi! This is the friendly automated conda-forge-linting service. I just wanted to let you know that I linted all conda-recipes in your PR ( |
|
@conda-forge-admin, rerender |
…nda-forge-pinning 2021.08.05.06.02.11
All suggestions were applied Co-authored-by: Isuru Fernando <[email protected]>
| USE_SYSTEM_CURL=1 \ | ||
| USE_SYSTEM_FFTW=1 \ |
There was a problem hiding this comment.
does fftw need to be added to the dependencies?
There was a problem hiding this comment.
No, FFTW isn't a dependency of Julia itself since JuliaLang/julia#21956
|
I ran |
|
@isuruf Where do we stand on this pull request? If this is workable for conda-forge, I suggest we make this available and move on to refining the recipe for 1.6.2 where there may be some helpful backported patches. |
|
|
||
| test: | ||
| commands: | ||
| - julia -e 'Base.runtests("LibGit2/online Pkg/pkg")' | ||
| - julia -e 'Base.runtests(["core", "regex", "Downloads"])' |
There was a problem hiding this comment.
Are the previous tests part of the new tests?
There was a problem hiding this comment.
No, LibGit2/online Pkg/pkg are not part of the new tests.
The old tests focus only on subsets of the LibGit2 and Pkg stdlib packages. They are actually not included in the default tests except when explicit requested. They were a weird choice for the only tests to run.
https://github.com/JuliaLang/julia/blob/68c458519d369d922e1bfc2c8f9bb240c8718eaa/test/choosetests.jl#L153-L154
Above in a previous comment I ran the standard test suite. It takes some time to run, which has been the subject of complaints in the git history.
The available tests to run are below:
$ make print-TESTS
TESTS=all stdlib unicode strings compiler NetworkOptions Artifacts CompilerSupportLibraries_jll SHA Zlib_jll Statistics Random LibGit2_jll DelimitedFiles Pkg CRC32c Profile OpenBLAS_jll nghttp2_jll OpenLibm_jll SparseArrays LibSSH2_jll p7zip_jll LibCURL_jll InteractiveUtils MPFR_jll MbedTLS_jll LinearAlgebra Distributed Test TOML ArgTools Logging FileWatching UUIDs Libdl Base64 Tar Markdown Sockets GMP_jll Serialization LibGit2 LibCURL Dates LibUnwind_jll Mmap SuiteSparse REPL PCRE2_jll Unicode SuiteSparse_jll libblastrampoline_jll SharedArrays LibUV_jll LLVMLibUnwind_jll libLLVM_jll MozillaCACerts_jll Future Printf LazyArtifacts Downloads dSFMT_jll subarray ambiguous filesystem arrayops read path error float16 precompile stacktraces triplequote iobuffer bitarray client stress corelogging sysinfo exceptions ordering checked loading complex ranges deprecation_exec math binaryplatforms cartesian threads_exec misc ccall sorting syntax goto hashing channels numbers spawn simdloop rational intrinsics subtype reinterpretarray interpreter test_sourcepath secretbuffer int rounding combinatorics opaque_closure euler asyncmap mod2pi parse specificity test_exec boundscheck_exec some choosetests smallarrayshrink worlds backtrace fastmath broadcast meta docs reduce env staged osutils iostream namedtuple show boundscheck core dict download_exec tuple atomics copy download gmp cmdlineargs reducedim floatapprox file atexit reflection abstractarray intfuncs char threads llvmcall2 sets mpfr iterators stress_fd_exec bitset vecelement errorshow generic_map_tests missing operators enums keywordargs functional offsetarray floatfuncs testenv stack_overflow version regex llvmcall ryu unicode/utf8 strings/io strings/util strings/search strings/types strings/basic compiler/inference compiler/validation compiler/interpreter_exec compiler/codegen compiler/contextual compiler/inline compiler/ssair compiler/irpasses
|
Thanks |
Checklist
0(if the version changed)conda-smithy(Use the phrase@conda-forge-admin, please rerenderin a comment in this PR for automated rerendering)This is based on #111 and likely to still fail, but I could not find the logs for that PR anymore. Also, would it be possible to run the entire testsuite (should take ~1h with multiple threads), since that is also likely to catch any issues with third-party dependencies?