Control: severity -1 wishlist
On Sun, Oct 01, 2017 at 03:57:54PM +0200, tho...@habets.se wrote:
> 1a) Why not use `AC_TRY_LINK`?
That'd work as well.
> 1b) Why not do the simplest change possible, and just `s/RUN/COMPILE/`
> (or `LINK`) without further code change?
AC_TRY_RUN and AC_TRY_COMPILE take different arguments. It's not that
easy unfortunately.
> 2) Indeed, why not use `AC_TRY_RUN` with third argument[1]?
I think that using the third argument of AC_TRY_RUN is a bad thing in
general. It encodes a guess on what is supposed to happen. The guess can
be right or wrong, so using it makes the build less reproducible and we
all want reproducible builds, right?
> As for runtime property: Yeah, that's true. I've mostly fixed the
> drawback of the false positive (turning runtime error when compile
> time worked into a warning) with a commit just now[2].
Thank you.
So if you insist on using RUN checks, I can add the relevant cache
variables[1], even though I think this is a bad solution. Whenever we
can reasonably do without RUN checks, we should. At least the two
prototype checks clearly allow doing so. For CLOCK_MONOTONIC, both of
our contradictory views make sense and we'll likely agree to disagree. I
could be persuaded that falling back to AC_CHECK_DECL for cross builds
only is reasonable.
I see that you understand the issues at hand very well and ask you to
implement the option you prefer (if any). My patch was meant as a
convenience and demonstration, but there are more solutions as you
figured.
In any case, we should not let this bug languish but try to close it
soonish with whatever resolution. If all else fails, the resolution is
to let the builder set the cache variables.
Helmut
[1] Unlike many other upstreams, arping does provide cache variables for
these things!
