Re: "libmpfr.so.4" not found

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Wed, 3 Oct 2018 14:24:01 +0200
    From:        Joerg Sonnenberger <joerg%bec.de@localhost>
    Message-ID:  <20181003122401.GA4078%britannica.bec.de@localhost>

  | I have no idea what you are doing.

I do, now (sheer stupidity) - see below, but ...

  | As I said, I can build it perfectly fine using src from a null read-only mount.

Yes, I remember that, but I suspect not using a shell that doesn't
support $LINENO (not well enough for configure anyway).
(I was going to offer to make and send one, just for this purpose!)

I got fixated on that, and didn't look at the more obvious things.

uwe%stderr.spb.ru@localhost said:
  | Are you sure you did't have an objdir accident before? 

Yes, I'm sure ... I've seen that kind of thing before (as in, it has
happened and I wasted lots of time cleaning up) ... that's why I
switched to a read-only source build, and I always use that
(in this source tree) so that cannot happen any more.   (I have
scripts that run the build, to make sure I don't mess up things
like that.)

It was even simpler, there simply was no obj directory - that is,
my most common builds use -o to build.sh to save some time
(every little bit helps to make sure I am not tempted to commit
changes without doing a build first).   Usually I change stuff,
build, fix, build, test, fix, build, test, commit, make more changes
(and start the build/fix/test/commit cycle all over again) and keeping
-o through all of that saves a bunch of time.

But every now and then, when something new is added (like
xz for example) it fails, and the read only src prevents files being
created in the wrong place, and normally it is obvious what
happened, and I just start again, without -o and (slightly slower)
it all works.   This time, because I as expecting possible failures
in the xz build, (it was what I was looking for) and because of the
$LINENO issue (and because I truly dislike autoconf to start with)
the cause of the problem wasn't as obvious as it should have
been, and so I went chasing the wrong things.

It would help though, if the Makefile system would test
whether an objdir should exist, and if so, use it, rather than
testing if one does exist, and using it if so, and just building
in the source directory if not.   Whether it would still need a
"make objdirs" pass first or not for other reasons, or whether
that could simply be part of building each directory doesn't
matter, but if the build failed with a "no objdir found" or
"can't cd to $OBJ/..." type error, this would have been obvious
right from the start - having to deduce the cause of the
problem from the "can't create xxx: read only filesystem"
error is not really the best way.

In any case, apologies for the waste of time.

kre