Storing the cache in an exported environment variable can
result in build failures if the cache exceeds the MAX_ARG_STRLEN
limit (128k on linux) which can be triggered by building in a
rather deeply nested directory.
Signed-off-by: Matthias Reichl <hias@horus.com>
This reverts commit 3b5986704e.
The config shell is set as an interpreter in libtool scripts
and as Linux has a hard limit of 127 bytes on the first line
in scripts (including the hashbang) this causes the build to
fail with a "bad interpreter" error if the path to the config
shell exceeds this limit.
These errors were first noticed on jenkins builds but can be
easily reproduced by building LibreELEC in a longer build dir.
eg RPi2 clean build failing on libtool invocation in kmod build:
./doltlibtool: /home/hias/rpi/libreelec-testing-with-a-long-directory-path-as-on-jenkins/build.LibreELEC-RPi2.arm-9.0-devel/kmod-24/.x86_64-linux-gnu/libtool: /home/hias/rpi/libreelec-testing-with-a-long-directory-path-as-on-jenkins/build.LibreELEC-RPi2.arm-9.0-devel/toolchain/bin/d: bad interpreter: No such file or directory
Signed-off-by: Matthias Reichl <hias@horus.com>
Builds use a mix of styles for the hash table in the ELF header. Only
one style of table is necessary, and the GNU one is more performant
for lookups. Eliminating the SysV style hash table trims ~450kb. This
is done via setting the default linker in gcc instead of LDFLAGS as
some packages ignore LDFLAGS.
Note that if MIPS is ever added as a supported architecture, it
does not support hash-style=gnu at this time.
Signed-off-by: Ian Leonard <antonlacon@gmail.com>
