Collectd with Musl build failure since NUT version 2.8.0

[ffontaine]

collectd fails to link with nut on musl since version 2.8.0 and 4f760ba on:

configure:107607: checking for UPSCONN_t
configure:107607: /usr/lfs/hdd_v1/rc-buildroot-test/scripts/instance-1/output-1/host/bin/i586-linux-gcc -c -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64  -Os -g0  -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64  conftest.c >&5
In file included from conftest.c:155:
/usr/lfs/hdd_v1/rc-buildroot-test/scripts/instance-1/output-1/host/i586-buildroot-linux-musl/sysroot/usr/include/upsclient.h:102:87: error: unknown type name 'time_t'
  102 | ssize_t upscli_sendline_timeout(UPSCONN_t *ups, const char *buf, size_t buflen, const time_t timeout);
      |                                                                                       ^~~~~~

I don't know what is the proper way to fix this failure, timehead.h can't be included as this is an internal header that depends on config.h.

Full build log: http://autobuild.buildroot.org/results/22b758097e8fb72c68e41329cbc7abc748d81ca6/build-end.log

[jimklimov]

Thanks, this does not sound right indeed. A few weeks back I had some PRs to revise public headers (#1616, #1615) - does the problem still happen with current master-branch NUT?

What header(s) declare time_t on that platform? (at worst a fallback could be declared, but this one is a pretty basic common data type)

[jimklimov]

So, it seems that currently it would still at best include time.h indirectly, and PRs above did not touch this header - fixing that to be sure. Thanks for the report!

(Hm, guess our CI needs some test cases to build stuff against public headers and libs without knowledge of NUT build workspace)

[jimklimov]

As for the build log, I suppose most of the error noise about ups->conn as an int is because of C defaulting:

src/nut.c:40:2: error: #error "Unable to determine the UPS connection type."
   40 | #error "Unable to determine the UPS connection type."
      |  ^~~~~
src/nut.c:46:3: error: unknown type name 'collectd_upsconn_t'
   46 |   collectd_upsconn_t *conn;
      |   ^~~~~~~~~~~~~~~~~~

further due to this:

checking upsclient.h usability... yes
checking upsclient.h presence... yes
checking for upsclient.h... yes
checking for upscli_connect in -lupsclient... yes
checking for upscli_init in -lupsclient... yes
checking for upscli_tryconnect in -lupsclient... yes
checking for UPSCONN_t... no
checking for UPSCONN... no

Not sure why configure decides the header is usable if it then fails to parse/use it, I suppose - if it does not find the UPSCONN_t as usable. But the code build does not seem to fail because of the header syntax (consumer C file includes time.h elsewhere in the chain before upsclient.h?) - e.g. it knows to expect struct <anonymous> * here:

In file included from src/nut.c:33:
/usr/lfs/hdd_v1/rc-buildroot-test/scripts/instance-1/output-1/host/i586-buildroot-linux-musl/sysroot/usr/include/upsclient.h:82:13: note: expected 'UPSCONN_t *' {aka 'struct <anonymous> *'} but argument is of type 'int *'
   82 | const char *upscli_strerror(UPSCONN_t *ups);
      |             ^~~~~~~~~~~~~~

In fact, for that build log I do not see mention of time_t at all.

However some other changes to less loose data types, like uint16_t vs. earlier int for the port numbers may need to be fixed in the consumer codebase. At least that much of the change was "sort of" documented:
https://github.com/networkupstools/nut/blob/master/UPGRADING#L51-L59

[jimklimov]

Also FWIW there are some caveats noted in https://musl.libc.org/time64.html and the file is apparently here http://git.musl-libc.org/cgit/musl/tree/include/time.h

[jimklimov]

FWIW, recording a quick reproducer:

nut$ ./ci_build.sh && make DESTDIR=/tmp/.inst install

# cloned from https://github.com/collectd/collectd
collectd# ./build.sh ### generate configure script

collectd$ ./configure --disable-all-plugins --enable-nut \
    CFLAGS="-I/tmp/.inst/usr/local/ups/include/ " \
    CPPFLAGS="-I/tmp/.inst/usr/local/ups/include/ " \
    LDFLAGS="-L/tmp/.inst/usr/local/ups/lib " \
    PKG_CONFIG_PATH="/tmp/.inst/usr/local/ups/lib/pkgconfig"

[jimklimov]

Note: also posted a PR to collectd to fix the int-type discrepancies (actually to detect which types suit the NUT API the consumer is building against, and transparently use those in the consumer codebase).