This reverts commit adee5315265cc46aa6a3057071527abb16e1cd94.
Turns out one of the "other things" is a dealbreaker.
We only create kegs using a formula's canonical name. However, we do not
check that this is the case when mapping existing kegs back to formula
objects, and thus a keg with a name that happens to be an alias can fool
Homebrew into thinking the canonically-named keg exists.
So anything that enumerates kegs and then tries to do stuff with the
resulting formula objects will just break. This is obviously worse than
the debugger being broken, so reverting this for the time being.
The Readline class clashes with the Readline module from the Ruby
stdlib. This has mostly worked, but with the recent debugging support's
integration of IRB, it is no longer possible for them to coexist. So we
need to rename it.
The implications of this are:
- Anything that depends on readline will reinstall it as
"gnu-readline". Anything already installed will continue to function.
- "brew upgrade readline" will say "gnu-readline not installed", as
"readline" is now an alias.
- Probably other things.
So there are some downsides, but we will just have to deal with them.
FixesHomebrew/homebrew#15776.
The MacOS.version? family of methods (other than "leopard?") are poorly
defined and lead to confusing code. Replace them in formulae with more
explicit comparisons.
"MacOS.version" is a special version object that can be compared to
numerics, symbols, and strings using the standard Ruby comparison
methods.
The old methods were moved to compat when the version comparison code
was merged, and they must remain there "forever", but they should not be
used in new code.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
OS X ships with iconv. glib failed to detect it, and this snowballed
into a few formulas requiring a superfluous duplicate install of
libiconv. All of the formulas in this commit compile without the
libiconv dependency.
jn:
The libiconv shipped with 10.5, 10.6, and 10.7 is version 1.11, so there
shouldn't be any cases of "10.x needs a newer libiconv but 10.y does
not".
In any case, we'll wait a week or so just in case any weird issues crop
up, and if not, then we can remove the libiconv formula and blacklist
it.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
Fixes some build issues on Lion.
Apparently it's enough to remove just the '-O' flags and preserve our
other optimization flags. This may only be needed on Lion, but frankly
clisp tends to have issues on all platforms so it's probably acceptable
to just disable it everywhere.
ClosesHomebrew/homebrew#10450.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
This patch from MacPorts seems to be all that is necessary to enable
building with LLVM-GCC. I can't test it with any LLVM builds other than
2336, though.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
It seems that the main GNU download site has issues in some places
outside the U.S., so we'll use the provided "ftpmirror.gnu.org" to pick
a nearby mirror.
Signed-off-by: Jack Nagel <jacknagel@gmail.com>
