This adds a CMake cache entry to std_cmake_args specifying that the
function clock_gettime is not available on 10.11 in order to avoid
runtime errors such as
dyld: lazy symbol binding failed: Symbol not found: _clock_gettime
when the build system is confused by Xcode 8's weak symbols.
Other weak symbols on 10.11, which may merit the same treatment in the
future, can be found with
grep 'weak$os10.11' MacOSX.sdk/usr/lib/system/libsystem_c.tbd
Previously, this method would:
- Include the current formula
- Include the current target of the installed alias, which by definition
is not "old"
This commit fixes both of these issues.
Only formulae that are the current target of the alias they were installed
with are now considered to have old installed formulae.
Not quite a mass replacement as I've used OS X and Mac OS X where
describing specific older versions and added compatibility methods
for things in the DSL.
Software that tries to print different things (e.g. ANSI color) based on
different termcaps often relies on the TERM environment variable, and
could fail without it. This results in confusing test issues where
certain tests can be successfully run by users testing locally with TERM
set, but choke up on our CI.
Always setting TERM to dumb leads to better consistency between local
tests and CI tests, and saves much probing in certain cases.
These definitions are scattered throughout the codebase which makes it
hard to refactor them later (my goal is to move them outside of
HOMEBREW_LIBRARY). Unify their definitions for clearer code and easier
movement later.
If formula is pinned it shouldn't be eligible for cleanup.
Otherwise we can encounter the following behaviour:
1. Install <formula>
2. Pin <formula>
3. <formula> gets outdated
4. Install <other_formula> that depends on <formula>, which requires
up-to-date installation of <formula>. Thus, <formula> gets updated.
5. Now, if we cleanup <formula> we don't remove pin, so it points to
the keg which doesn't exist any longer. The right behaviour is not
to cleanup pinned kegs, which is done in this commit.
Also see #831 for further info.
While using `"true"` instead of `true` wouldn't break anything, it doesn't
seem to be expected, and using `"false"` instead of `false` _would_ break
things.
* add `latest_head_version` to return latest HEAD version installed
* add `latest_head_prefix` to return Pathname with latest HEAD version
* add `head_version_outdated?` to check if HEAD version is up-to-date
`any?` is not the opposite of `empty?`. Besides the case that
`[false, nil].any?` will return false, `any?`(O(n)) has much worse
performance than `empty?`(O(1)).
Substitue each Version.new and HeadVersion.new with Version.create
to unify Version and HeadVersion instantiation among core code.
Note that this does not relate to Mac::OS::Version class.
Users may have ~/.curlrc file to include options like proxies.
However, since we overwrite HOME environment variable during the build
and test, curl won't be able to find it.
This commit solves this issue by using CURL_HOME environment variable,
which will be pointed to the original HOME path.
From `curl(1)`:
> 1) curl tries to find the "home dir": It first checks for the
CURL_HOME and then the HOME environment variables. Failing that,
it uses getpwuid() on Unix-like systems (which returns the home
dir given the current user in your system). On Windows, it then
checks for the APPDATA variable, or as a last resort the '%USER-
PROFILE%\Application Data'.
This gives the JSON output the same type of information as
`recommended_dependencies` or `optional_dependencies`, but for those
marked `:build` in the formula.
