fix: backport #3048 to 1.4: don't watch non-existent include directory (

#3174)

fix(local-toolchains): don't watch non-existent include directory (#3048)

Apparently, Macs can mis-report their include directory.

Since includes are only needed if C extensions are built, skip watching
the directory
if it doesn't exist.

Work around for
#3043

---------

Co-authored-by: Richard Levasseur <richardlev@gmail.com>
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Co-authored-by: Ignas Anikevicius <240938+aignas@users.noreply.github.com>

fix(local-toolchains): don't watch non-existent include directory (#3048

)

Apparently, Macs can mis-report their include directory.

Since includes are only needed if C extensions are built, skip watching
the directory
if it doesn't exist.

Work around for
#3043

---------

Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Co-authored-by: Ignas Anikevicius <240938+aignas@users.noreply.github.com>
(cherry picked from commit be55942)

fix(core): do not assume rules_python runtime (#3134)

This change reverts the behaviour where we assume that particular
attributes will be always present - if bazel is doing autoloading for
WORKSPACE builds (7.6.1), then we will crash with attribute error.

I could not think how to add a unit test, which would test this fix
because it seems to only happen with a released version of rules_python
where we are not using `local_repository` override.

Fixes #3119

---------

Co-authored-by: Richard Levasseur <richardlev@gmail.com>
(cherry picked from commit acf7507)

Cherry-pick notes: adjusted changelog to mention 1.5.2

fix(pypi): only generate namespace package shims if implicit namespac…

…es are disabled (#3059)

The refactoring to move the pkgutil shim generation to build phase
inverted
the logic for when it should be activated. When
`enable_implicit_namespace_pkgs=True`, it means to not generate the
pkgutil shims
("respect the Python definition of the namespace package").

To fix, just invert the logic that activates it.

A test will be added in a subsequent PR because the necessary helper
isn't
in the 1.5 branch.

Fixes #3038

---------

Co-authored-by: Ignas Anikevicius <240938+aignas@users.noreply.github.com>
(cherry picked from commit 47c681b)

fix: work around version parsing by only parsing if site-packages is …

…enabled (#3031)

There's a bug in the version string parser that doesn't handle local
identifiers correctly.

Thankfully, it's only activated in the experimental code path when site
packages for
libraries is eanbled. Moving the logic within that block works around
it.

Work around for
#3030

(cherry picked from commit aab2650)

fix: work around version parsing by only parsing if site-packages is …

…enabled (#3031)

There's a bug in the version string parser that doesn't handle local
identifiers correctly.

Thankfully, it's only activated in the experimental code path when site
packages for
libraries is eanbled. Moving the logic within that block works around
it.

Work around for
#3030

(cherry picked from commit aab2650)

fix(pypi): namespace_pkgs should pass correct arguments (#3026)

It seems that the only function that did not have unit tests have bugs
and the integration tests did not catch it because we weren't creating
namespacepkg `__init__.py` files.

This change fixes the bug, adds a unit test for the remaining untested
function.

Fixes #3023

Co-authored-by: Richard Levasseur <richardlev@gmail.com>
(cherry picked from commit 4978027)

fix(toolchains): use posix-compatible exec -a alternative (#3010)

The `exec -a` command doesn't work in dash, the default shell for
Ubuntu/debian.

To work around, use `sh -c`, which is posix and dash compatible. This
allows changing
the argv0 while invoking a different command. Also adds a test to verify
the the
runtime_env toolchain works with bootstrap script.

Fixes #3009

(cherry picked from commit c4543cd)

fix: use platform_info.target_settings in toolchain aliases (#3001)

During the refactor we forgot one more place where the `flag_values` on
the platform information was used. They were no longer populated and
broke.

The solution is to use `selects.config_setting_group` to maintain
behaviour and in order to smoke test I have added a target to verify
that the aliases work.

Related to #2875
Fixes #2993

Co-authored-by: Richard Levasseur <richardlev@gmail.com>
(cherry picked from commit 107a878)

fix: support pre-release versions and add new toolchain versions (#2969)

Add latest toolchain builds and attempt adding a beta build.

This shows/tests that we can handle pre-release versions just fine
and we are able to test the toolchain matching. Whilst at it it
implements
the static advertising of the remaining interpreter information.

Fixes #2837

---------

Co-authored-by: Richard Levasseur <richardlev@gmail.com>