> configure: error: cannot compute sizeof (size_t)
There is a known generic problem in configure scripts. Internally they use
a small C program for type checking. Sometimes that test program fails for
some other reason, something like library not found; and the error message
does not report the real problem. That is probably what is happening
here. A careful examination of config.log may reveal the real problem.
Feel free to post config.log as link or attachment, and I will take a look.
Alternatively, try building netCDF with Cmake rather than configure. It
uses completely different code for testing system features.
On Mon, May 26, 2025 at 11:24 AM Alvaro Fernandez <
afernandez@xxxxxxxxxxxxxxx> wrote:
> Hi,
>
> I'm trying to build NETCDF on (a QEMU emulation of) a RISC-V platform
> we're developing.
>
> I've successfully built other libraries (like HDF5, zlib, etc.) on this
> QEMU with gcc 14.2.0.
>
> However, I am having difficulty completing the configuration phase of the
> NETCDF build process.
>
> This machine will not be accessing the web or pulling down anything, so I
> disabled DAP and XML, and am not giving it curl.
>
> I started with this configure command:
>
> ./configure --disable-dap --disable-libxml2
> CPPFLAGS="-I/home/root/WRF/include" LDFLAGS="-L/home/root/WRF/lib" 2>&1 |
> tee configure-net.out
>
> But I get errors like:
>
> checking size of size_t... configure: error: in
> '/home/root/repos/netcdf-c-4.9.3':
> configure: error: cannot compute sizeof (size_t)
> See 'config.log' for more details
>
> I reasoned that the configure script might be confused about the host or
> built targets, so I decided to be explicit.
>
> Below is the latest attempt. I added the cache file:
>
> root@qemuriscv64:~/repos/netcdf-c-4.9.3# cat config.cache
> ac_cv_sizeof_off_t=8
>
> And then I ran:
>
> ./configure --disable-dap --disable-libxml2
> CPPFLAGS="-I/home/root/WRF/include" LDFLAGS="-L/home/root/WRF/lib"
> --host=riscv64-linux-gnu --build=riscv64-linux-gnu
> --cache-file=config.cache 2>&1 | tee configure-net.out
>
> It fails with the same error.
>
> This error is detected in line 26715 of the configure script.
>
> I see from the comments that this test serves to guard against an error in
> the HP compiler. Is this still a useful test?
>
> I would suspect our toolchain usually, but it's built hdf5 and multiple
> other applications.
>
> Thanks in advance,
>
> Alvaro Fernandez, PhD
> Inspire Semi
>