PR2 for zOS support #7973
PR2 for zOS support #7973
Conversation
assert.hpp - make MVS take the else block linenoise.cpp - include strings and sys/time tsd.cpp - include pthread and use pthread_equal
|
Maybe just disable jemalloc on zOS instead? |
|
With the patches in this PR, it goes through OK. |
|
@Mytherin Wise words indeed. |
|
Thanks - looks fine to me. Could you just fix the failing CI? |
|
CodeQuality - I'll push a commit for the format-check Windows - I see you were working on the ODBC tests, and the error is probably related to that (?) LinuxRelease - Fails on Deploy with for x in regex_include_statements:
included_file = x[1]
if skip_duckdb_includes and 'duckdb' in included_file:
continue
+ if ('pg_functions.cpp' or 'src_backend_parser_scansup.cpp' in fpath) and included_file == 'zos-tls.h':
+ continue
if 'extension_helper.cpp' in fpath and (included_file.endswith('_extension.hpp') or included_file == 'generated_extension_loader.hpp'):
continueWith the first and then this PR, DuckDB compiles & works on zOS, but some tests from the unittest fail. Assuming that this is not the end of zOS support work, is the above solution acceptable? |
|
For the amalgamation builds the wrong include tags are likely used. |
- include zos-tls as a system header - update test/CMakeLists.txt for format.py
|
Awesome, thank you. Bringing this back to 'ready for review'. |
|
@Mytherin could you please check if this is ready for pull now. |
|
|
|
Could you merge with master so that the CI can run again? |
|
Thanks! |
I done goof'd with
lightweightsemaphore.hin the first PR.Have verified that it builds fine with the unix block, so just using that.
My upstream PR to concurrentqueue does this correctly.
There's a situation with
jemalloc_internal_types.h.For example, here's the related command from compiler_commands.json:
{ "directory": "/v1g/zopen/dev/duckdbport/duckdb/extension/jemalloc/jemalloc", "command": "/usr/lpp/IBM/oelcpp/v2r0/bin/clang++ -DDUCKDB_BUILD_LIBRARY -DJEMALLOC_NO_PRIVATE_NAMESPACE -DJEMALLOC_NO_RENAME -I/v1g/zopen/dev/duckdbport/duckdb/src/include -I/v1g/zopen/dev/duckdbport/duckdb/third_party/fsst -I/v1g/zopen/dev/duckdbport/duckdb/third_party/fmt/include -I/v1g/zopen/dev/duckdbport/duckdb/third_party/hyperloglog -I/v1g/zopen/dev/duckdbport/duckdb/third_party/fastpforlib -I/v1g/zopen/dev/duckdbport/duckdb/third_party/fast_float -I/v1g/zopen/dev/duckdbport/duckdb/third_party/re2 -I/v1g/zopen/dev/duckdbport/duckdb/third_party/miniz -I/v1g/zopen/dev/duckdbport/duckdb/third_party/utf8proc/include -I/v1g/zopen/dev/duckdbport/duckdb/third_party/miniparquet -I/v1g/zopen/dev/duckdbport/duckdb/third_party/concurrentqueue -I/v1g/zopen/dev/duckdbport/duckdb/third_party/pcg -I/v1g/zopen/dev/duckdbport/duckdb/third_party/tdigest -I/v1g/zopen/dev/duckdbport/duckdb/third_party/mbedtls/include -I/v1g/zopen/dev/duckdbport/duckdb/third_party/jaro_winkler -I/v1g/zopen/dev/duckdbport/duckdb/extension/jemalloc/include -I/v1g/zopen/dev/duckdbport/duckdb/extension/jemalloc/jemalloc/include -DNSIG=42 -D_XOPEN_SOURCE=600 -D_ALL_SOURCE -D_OPEN_SYS_FILE_EXT=1 -D_AE_BIMODAL=1 -D_ENHANCED_ASCII_EXT=0xFFFFFFFF -DZOSLIB_OVERRIDE_CLIB=1 -fzos-le-char-mode=ascii -mnocsect -fno-short-enums -Wno-error -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -O3 -fgnu-keywords -I/v1g/zopen/prod/zoslib-zopen/include -O3 -DNDEBUG -O3 -DNDEBUG -fPIC -w -std=c++11 -o CMakeFiles/jemalloc.dir/src/ctl.cpp.o -c /v1g/zopen/dev/duckdbport/duckdb/extension/jemalloc/jemalloc/src/ctl.cpp", "file": "/v1g/zopen/dev/duckdbport/duckdb/extension/jemalloc/jemalloc/src/ctl.cpp" },ctl.cpp is being built with clang++, ok.
However, it refers to the internal types header file, which looks for
__STDC_VERSION__.At least on zOS, clang++ doesn't know about
__STDC_VERSION__, so it goes down this path of thinking it's <199901L.Have addressed this, but wanted to check if this requires any further investigation.