Odd differences with cURL and GCC 9 and 10: curl_easy_setopt

Dear cURL Gurus,

I'm having some odd issues. Now, in truth the odd issues are manifesting in netCDF-C:

https://github.com/Unidata/netcdf-c/issues/1725

but one of the devs pointed me to one place to explain some of it: Sometimes when I build with GCC 10.1, netCDF-C reports:

  checking for curl_easy_setopt in -lcurl... no

and sometimes it reports:

  checking for curl_easy_setopt in -lcurl... yes

The thing is, in both cases, I have *never* changed how I build cURL (and haven't for a while) and I don't change how I build netCDF! I essentially do:

        @echo "Configuring curl"
        @(cd curl; \
          export PATH="$(prefix)/bin:$(PATH)" ;\
          export CPPFLAGS="$(INC_SUPP)";\
          export LIBS="-lm";\
          ./buildconf;\
          ./configure --prefix=$(prefix) \
                      --includedir=$(prefix)/include/ \
                      --libdir=$(prefix)/lib \
                      --with-zlib=$(prefix) \
                      --disable-ldap \
                      --enable-manual \
                      --disable-shared \
                      --enable-static \
                      --without-libidn \
                      --without-libidn2 \
                      CFLAGS="$(CFLAGS)" CC=$(CC) CXX=$(CXX) FC=$(FC) )
        @touch $@

to build cURL. (I run buildconf because my "Base Libraries" (aka Baselibs) have curl as a submodule in a Git repo.)

I have nothing between the two that seems to be different other than whether I load my GCC 9.3 module or my GCC 10.1 module. If I diff curl-config after each successfully builds, the main difference is:

❯ diff curl.930.log curl.1010.log
31d30
< parallel-tests: installing './test-driver'
726c725
< Install prefix: /Users/mathomp4/installed/MPI/gcc-gfortran-9.3.0/openmpi-4.0.3/Baselibs/6.0.12-Snap13/Darwin
---
> Install prefix: /Users/mathomp4/installed/MPI/gcc-gfortran-10.1.0/openmpi-4.0.3/Baselibs/6.0.12-Snap13/Darwin

(Note CPPFLAGS and LDFLAGS are also different because of different installation prefixes for sub-libraries)

I'm kind of at a loss. The log output from my curl build is about as identical as you can get. Thus, I can't see why support for curl_easy_setopt would be different.

If I try building simple.c:

https://curl.haxx.se/libcurl/c/simple.html

with my GCC 9.3 and GCC 10.1 build of cURL they both work.

Is it possible the test for curl_easy_setopt in netCDF-C's configure isn't good? I don't read autotools well enough to know.

Any ideas? Am I perhaps building cURL incorrectly in some way that netCDF is annoyed with it?

Thanks,
Matt

--
Matt Thompson, SSAI, Ld Scientific Programmer/Analyst
NASA GSFC, Global Modeling and Assimilation Office
Code 610.1, 8800 Greenbelt Rd, Greenbelt, MD 20771
Phone: 301-614-6712 Fax: 301-614-6246
http://science.gsfc.nasa.gov/sed/bio/matthew.thompson

-----------------------------------------------------------
Unsubscribe: https://cool.haxx.se/list/listinfo/curl-users
Etiquette: https://curl.haxx.se/mail/etiquette.html
Received on 2020-05-12