Make [email protected] the main python

[fxcoudert]

Here we go…

[fxcoudert]

Testing cannot proceed because:

[email protected]:
  * Versioned formulae in homebrew/core should use `keg_only :versioned_formula`

[fxcoudert]

With this PR the output of python3-config --cflags --ldflags --embed (for [email protected]) on macOS 10.14 is:

-I/usr/local/Cellar/[email protected]/3.9.0_1/Frameworks/Python.framework/Versions/3.9/include/python3.9 -I/usr/local/Cellar/[email protected]/3.9.0_1/Frameworks/Python.framework/Versions/3.9/include/python3.9 -Wno-unused-result -Wsign-compare -Wunreachable-code -fno-common -dynamic -DNDEBUG -g -fwrapv -O3 -Wall -isysroot /Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk -I/Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/usr/include -I/Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/System/Library/Frameworks/Tk.framework/Versions/8.5/Headers -L/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.9/lib/python3.9/config-3.9-darwin -lpython3.9 -lintl -ldl -framework CoreFoundation

Without the PR, it is currently (also for [email protected]), on my local machine (macOS 10.15):

-I/usr/local/Cellar/[email protected]/3.9.0/Frameworks/Python.framework/Versions/3.9/include/python3.9 -I/usr/local/Cellar/[email protected]/3.9.0/Frameworks/Python.framework/Versions/3.9/include/python3.9 -Wno-unused-result -Wsign-compare -Wunreachable-code -fno-common -dynamic -DNDEBUG -g -fwrapv -O3 -Wall -isysroot /Library/Developer/CommandLineTools/SDKs/MacOSX10.15.sdk -I/Library/Developer/CommandLineTools/SDKs/MacOSX10.15.sdk/usr/include -I/Library/Developer/CommandLineTools/SDKs/MacOSX10.15.sdk/System/Library/Frameworks/Tk.framework/Versions/8.5/Headers -L/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.9/lib/python3.9/config-3.9-darwin -lpython3.9 -ldl -framework CoreFoundation

I do not know where the extra -lintl flag comes from, but it is wrong. It triggers test failures for pybind11 and boost-python3. It may be a 10.14 vs. 10.15 issue, but I don't think that's likely (and we will know when the 10.15 testing is over).

[fxcoudert]

Testing this locally on 10.15, I cannot reproduce the issue. During python 3.9.0 compilation, I get:

checking libintl.h usability... no
checking libintl.h presence... no
checking for libintl.h... no

and

checking for textdomain in -lintl... no

The resulting python installation does not have -lintl in the output of python3-config --cflags --ldflags --embed

---

As far as I can see, the presence of -lintl in LIBS (and therefore in python3-config) comes from the check for textdomain. If for some reason gettext was present on the CI nodes (perhaps from the testing of dependents of [email protected] before building [email protected]?) then it will be picked up (I can reproduce this locally when gettext is installed).

[fxcoudert]

On 10.13, looking good:

brew test --retry --verbose freeling
brew install googletest
brew test --retry --verbose pcl
brew linkage --test urh

• Fixing urh in urh: rebuild to avoid uhd linkage #63484
• Fixing freeling in freeling: revision for boost #63485
• googletest conflicts with nss: fixing in googletest and nss conflict #63486
• pcl does not find Eigen3: it's preexisting issue, something apparently broke it since eigen: emergency patch for failures when used with OpenMP #62572

---

On 10.14, we have the same plus:

brew test --retry --verbose boost-python3
brew test --retry --verbose buku
brew test --retry --verbose glyr
brew test --retry --verbose pybind11

• boost-python3 and pybind11 are related to libint/gettext, see my investigations above
• glyr is a download error
• buku has a [ERROR] No matching index or title immutable or empty DB, not sure why

---

On 10.15, we have in addition:

brew test --retry --verbose ngs
brew install --only-dependencies oclgrind
brew install oclgrind

where oclgrind is due to No space left on device (it couldn't install the llvm@9 bottle)

[MikeMcQuaid]

Makes sense to me 👍🏻

[fxcoudert]

There is not switch in Python's configure to allow it to disable libintl if found.
We could configure with ac_cv_lib_intl_textdomain=no, or split this into two pull requests (but that might introduce some audits failures, and the question of the switching of the symlinks is not clear).

Edit: I've chosen the ac_cv_lib_intl_textdomain=no approach. 2nd try…