Apt considers 1.5.1.rel to be a lower version than 1.5.rel so the
golang 1.5 package was never updated to 1.5.1.
Fix that by introducing an epoch number (1) and prefix it to 1.5.1,
so that the resulting version number 1:1.5.1 is considered higher.
Thanks to Kristof Kovacs for reporting.
Add --enable-unicode=ucs4 and note that langinfo.h exists. One
result is that sys.stdout.encoding now is UTF-8 instead of None,
which allows e.g. bpython to launch without messing with the
PYTHONIOENCODING environment variable (see #35).
Work around rpl_malloc being used, see
http://wiki.buici.com/xwiki/bin/view/Programing+C+and+C%2B%2B/Autoconf+and+RPL_MALLOC
for more information:
"The AC_FUNC_MALLOC macro makes sure that the malloc function when passed a zero
argument returns a valid memory block instead of a NULL pointer. This behaviour
conforms to the GNU C library. Normally, this is a reasonable test that autoconf
makes at build-time. In the case of cross-compilation, however, autoconf cannot
execute a program to verify proper behavior. It makes the conservative assumption
that the target library will produce non-conforming code.
Failure of this test causes autconf to replace malloc() calls with rpl_malloc()
calls. At link time, if there is no rpl_malloc() function, the linker will fail
with an error describing the missing symbol. The autoconf documentation recommends
adding this harmless code to the application to implement the function."
In Termux the rpl_malloc() usage is useless at best, and may also prevent building
some packages as well as giving runtime crashes for libgc-using packages or others
expecting to intercept malloc.
Previously some packages specified worked around this themselves, but the configure
arguments are now moved into build-package.sh.
My hope was to use zile given that Emacs is segfaulting. The good news
is that zile builds. The bad news is that zile also segfaults.
According to gdb, the crash is somewhere in GC_realloc.
Fredrik Fornwall
Iain Nicol