Re: [gpsd-dev] ✘CentOS fixed?

[Fred Wright]

On Fri, 14 Jun 2019, Gary E. Miller wrote:
> Maybe I just fixed all the CentOS warnings.  Please test.
>
> The CentOS I installed was 32-bit.  scons needs epel, which does not
> suupport 32-bit.  So I'll try again with a 64-bit CentOS.

For what I have here, it's down to warnings on Ubuntu, and both warnings 
and errors on OpenBSD.  I'll send the unstripped output privately, this 
time for just three VMs (Ubuntu, OpenBSD 64-bit, and OpenBSD 32-bit).  The 
others are either clean or have just the known initializer warnings.


BTW, there's another category of warnings I've been mostly ignoring, 
though I don't think they show up in these cases.  Some versions of clang 
warn that compiling C code as C++ is deprecated.  I'm not sure exactly 
what it means by that, since it's certainly legal to write "clean C", 
based on the intersection of legal C and legal C++, and then compiling 
either way.  I haven't tried to figure out exactly what it's complaining 
about.

In any case, the source of this is where libgps is compiled as C++ for the 
Qt library.  Since it's only for Qt, and since Qt support is automatically 
turned off when the Qt prerequisites are missing, it may not show up even 
with an applicable clang version.  I always see it here when building 
natively, but I probably haven't bothered to install the Qt stuff on the 
VMs.

I think the right fix for this is to have the Qt library call libgps as C 
code (via the usual extern "C" mechanism), rather than building a separate 
version just for Qt (though libgps would need a few tweaks to replace the 
Qt compile-time conditionals).  But that's not trivial, and I wouldn't 
touch the Qt code to that degree without at least one decent test program 
for Qt.  I did obtain the one that's referenced in the GPSD documentation, 
but it seems to have suffered from bit rot, so it would take some work to 
revive it even before messing with GPSD's Qt code.  So for now, I just 
ignore those warnings.

Right now, libgps is compiled up to *four* times, based on all four 
combinations of C v.s C++ and PIC vs. non-PIC.  It would be nice to at 
least get that down to two, if not one.

Fred Wright