Hi. Some authors indeed forget about cross compilation when they publish their m4 to autoconf-archive.
Good example is byte order check aka endianness. Before endian.h one way to check it was to run a program. But the cross-compile alternative was to grep the binary for a magic byte sequence without running. Sometimes running is the only option though. There are two possible behaviors in case of cross compile I think: 1) to assume some result or 2) to fail configure unless some explicit result is provided on configure command line. 10.03.2017 8:50, R0b0t1 пишет: > Hello, > > Others as well as myself have encountered issues with cross compiling > various programs making use of configuration scripts created with > autoconf. The main issue is that the cross compilation environment > causes configuration scripts to produce feature detection binaries > matching the architecture of the compilation target, not the > compilation host. E.g. the script will try to run an ARM binary on an > x86 processor. > > Here are two example bugs, filed with the respective project in the > hopes of an interim solution: > > https://savannah.gnu.org/bugs/?50385 > https://gitlab.com/jas/libidn2/issues/6 > > Per my reasoning in those bug reports, my suggestion is to provide > another way to pass information to the configuration script. While it > seems possible to make configuration scripts compile and run binaries > of the proper architecture any results they obtain are irrelevant as > they reflect the capabilities of the host, not the target. > > R0b0t1. >
