Configure: error: cannot compute sizeof (long double)

Ken Moffat ken at
Sun Jul 15 13:06:26 PDT 2007

On Sun, Jul 15, 2007 at 02:09:51PM -0400, Baho Utot wrote:
> It is hard to parse the config.log in glibc-build directory.

 Yes, but it is essential.  The last part of it is a dump of all the
variables, which you can probably ignore.  Before that, each test in
the configure script typically generates something to say what is
running, some code which you will see, and then command(s) to check
the result, plus the output including errors, and the one-liner that
appears on the screen.

>  It is the only 
> file there, there are no others.  I would like to know how the test for long 
> double works.:u

 Mostly, configure generates some code (you can see it in config.log)
- it will be near the error message.  In this case the code is
probably just a call to sizeof(long double).  Configure then ompiles
the code and tries to run it.  Sizeof uses glibc,  and compiling
uses gcc.  If it works, you would get a result of something like 12
or 16 (the number of bytes meeded to hold a long double - it varies
across different architectures).

> The only thing I see missing is configure looking for and
> Those are in /tools/lib.

 When a test works as expected, it often fails because something is
not present on a particular platform.  In your case, there is a real
problem.  You are getting some sort of error message ( ? is
that a typo for crt1.o ?) - please tell the list what it says.

 It looks as if glibc is broken, so from here, I _guess_ that
either you went wrong when adjusting the toolchain, or you missed
one of the symlinks, or perhaps you built in stages and something
is linked agaisnt the host's libraries.

 Start by checking each of the symlinks (does the link exists, is it
the right name, does it point to the right destination (beware of
typos), and does the file it points to exist).  After that, run
'readelf -l' against example programs to make sure they are linked
against /tools.

das eine Mal als Tragödie, das andere Mal als Farce

More information about the lfs-support mailing list