These now return an Array of all the target destinations.
Previously, if a single argument was passed a single non-
Array was returned.
This behavior has been changed so that an Array is always returned
even for a single argument.
Updated the test.
Hopefully this won't break any custom code out there.
Rationale: some software (e.g. GNU Coreutils, GnuTLS 3.x), have started
distributing _only_ xz-compressed tarballs. There is no system XZ
utility provided by OS X, but it is necessary so that we can continue to
provide formulae for this software.
If XZUtils isn't installed, we abort and prompt the user to `brew
install xz`.
The `xz` command itself doesn't do any untarring, so we write the
decompressed archive to stdout and pipe it to tar.
Commit dec4b73aa0 ("Allow `brew update --rebase`") changed the format of
the `git pull` command slightly, introducing an extra space (to allow
for the --rebase option), so let's update the test expectations for this.
We still need to write tests for '--rebase', but in the meantime they
once again pass for the normal case.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
Commit 2695821e98 ("Only use the bottle if its version is up-to-date")
essentially broke the ImageMagick bottle, because the version parsing
logic returns "1" as the bottle version.
Fixing this requires only a slight modification to the bottle URL regex;
includes a test.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
I inadvertently gzipped it last night. That in itself doesn't cause any
tests to fail because the download strategy code is robust, but it's
fixed here anyway to be consistent. Basically, I'm dumb.
I also took the opportunity to remove the other resource fork file, so
the tarball is clean.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
The test_a_basic_install was failing because it expected
libexec.children.length to be 1, when in fact there were two files in
the untarred libexec: NOOP, and the Mac OS X resource fork file ._NOOP.
That file is now removed, and the test passes.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
Consequence: you can no longer install when something is already installed, you must upgrade it. This doesn't apply if the formula in question was unlinked. You can still --force installs though.
Rationale: the old way of installing over the top would leave symlinks to multiple versions in /usr/local if the old version had a file the newer version didn't. The new upgrade command handles everything properly.
This fixes reporting of which formulae changed in git versions where `pull`
output is not compatible.
Signed-off-by: Adam Vandenberg <flangy@gmail.com>
SourceForge links can end with /download which appears to be part of
what is used to drive mirror selection. Modifying the version detection
extension of pathname.rb to detect this case for both sourceforge.net
and sf.net download selections.
This is sufficient for version detection but may not be sufficient for
other uses of /download links. I have not found any problems with links
that use /download, though.
Signed-off-by: Adam Vandenberg <flangy@gmail.com>
The code was sucking. To the extent that maintenance was hard. It's a lot
easier to work with code that is sensibly split at sensible boundaries. So
now it is more like that.
But the refactor is minimal. Because we don't want you to have more merge
hell than absolutely necessary.
If you merge you will need to pay attention to brew.h.rb (as it is deleted)
and bin/brew (as command logic is gone). It will be painful, but you will just
have to help git out by moving any changes around manually.
Note compatibility.rb. It ensures that any function renames or removals don't
break anything. We're pretty serious about backwards compatibility. And that's
because we encourage you to hack around with the innards. And we couldn't do
that if we would then just make stuff disappear behind your back.
The update_git_pull_output_with_formulae_changes fixture (which consists
of mocked-up `git pull` output) listed a few files as both updated and
deleted. This caused the test_update_homebrew_with_formulae_changes test to
fail. My guess is that this fixture was mocked-up before formula deletion
handling was introduced, and became broken when it was. This revision just
makes this fixture internally consistent, and allows the associated test to
pass.
Signed-off-by: Adam Vandenberg <flangy@gmail.com>
Sphinx currently has a download name 'http://sphinxsearch.com/downloads/sphinx-1.10-beta.tar.gz', which homebrew currently can't correctly identify - with a small change to the regex for 'foobar-4.5.0-beta1' (make that trailing number optional) - it works like a charm.
Signed-off-by: David Höppner <0xffea@gmail.com>
