Hi Vagrant, On Thu, Nov 30, 2023 at 10:43:18AM -0800, Vagrant Cascadian wrote: > I tried this patch against the current sources, but now it fails at > dh_auto_configure. > > Build log attached for cross-building i386 (also tried arm64 and armhf > with the same basic issue).
I think this is all well. The failure you see arises from point 2. Originally, this was an AC_RUN_IFELSE, which was a hard failure. The patch changes that to be AC_CHECK_FILE and exporting ac_cv__dev_input makes configure continue beyond that point. You may be wondering why this is necessary. The build system wants to determine whether the host system has /dev/input. Clearly, checking the existence of that file on the build system is a stupid idea. This situation is precisely what AC_CHECK_FILE is meant for and it rightly fails during cross builds, because it has no access to a host system either. Then you may wonder whether such a test is reasonable in the first place. A container may reasonably come without /dev/input. If lirc were to then disable libinput support, that'd be bad. I agree. Yet, this is what configure tries to achieve and my patch changes this aspect in a minimal way (retaining this property). You may be going further and suggesting to upstream that this check is a bad idea to start with. I would not disagree. The AC_CHECK_FILE may be an easier sell to upstream. So if you got this far, I suggest that you just close the bug after applying most of the patch that made sense and calling it a day. Thanks for looking into it Helmut
