graphviz @2.40.1_2 +pangocairo-x11-universal: no member named 'min' in namespace 'std::__1'

[dubiousjim]

I can't build graphviz with a new install of OS X 10.15.6, Xcode 11.6. Have sudo port deactivate active; sudo port install -Nv graphviz build.jobs=1. My /opt/local/etc/macports/variants.conf has +no_x11 -x11 +quartz -universal.

Will attach log of failed install.

[jmroot (Joshua Root)]

Looks very similar to #60927. Are you missing the Command Line Tools? Please follow wiki:ProblemHotlist#reinstall-clt and then clean graphviz and retry the installation.

[ryandesign (Ryan Schmidt)]

Has duplicate #61071.

[ddrum2000]

Replying to jmroot:

Looks very similar to #60927. Are you missing the Command Line Tools? Please follow wiki:ProblemHotlist#reinstall-clt and then clean graphviz and retry the installation.

This seemed to work. Thanks.

[ryandesign (Ryan Schmidt)]

Has duplicate #61093.

[breiter (Brian Reiter)]

Forced reinstall of Command Line Tools solves this for me as well.

[ryandesign (Ryan Schmidt)]

Has duplicate #61949.

[jboydon (John Boydon)]

On BigSur 11.1 the command line tools (and Xcode) reinstallation doesn't solve the problem. The first error is slightly different:

In file included from /opt/local/libexec/llvm-11/bin/../include/c++/v1/string_view:175:
/opt/local/libexec/llvm-11/bin/../include/c++/v1/__string:337:12: error: no member named 'copy_n' in namespace 'std::__1'; did you mean 
'copy'?
    _VSTD::copy_n(__s2, __n, __s1);
    ~~~~~~~^
/opt/local/libexec/llvm-11/bin/../include/c++/v1/iterator:1374:93: note: 'copy' declared here
template <class _Ip, class _Op> _Op _LIBCPP_INLINE_VISIBILITY _LIBCPP_CONSTEXPR_AFTER_CXX17 copy(_Ip, _Ip, _Op);
                                                                                            ^
/opt/local/libexec/llvm-11/bin/../include/c++/v1/__bit_reference:424:43: error: no member named 'min' in namespace 'std::__1'

I think is due to the same reason.

[ryandesign (Ryan Schmidt)]

John, this ticket is about a problem that is solved by reinstalling the CLT. Since reinstalling the CLT didn't solve your issue, it's a different issue that should probably be in a different ticket. Note not only the different error message but also the fact that your error is referencing a file inside /opt/local, not a file provided by the CLT.

[jboydon (John Boydon)]

Thanks Ryan, I will create a new ticket.

[mascguy (Christopher Nielsen)]

Interestingly enough, this issue is also occurring for graphviz-devel, on 10.6, when built via MacPorts Clang 11. With the caveat that graphviz-devel is at version 5.0, though it's still the same type of failure.

Meanwhile, our 10.6 buildbots failed earlier in the process, with a different issue, which I haven't been able to reproduce locally. So perhaps they simply didn't get far enough to see this.

Anyhow, I've queued rebuilds for 10.6, and will review once the buildbots are caught up.

Thought all...?

[mascguy (Christopher Nielsen)]

Replying to mascguy:

Interestingly enough, this issue is also occurring for graphviz-devel, on 10.6, when built via MacPorts Clang 11. With the caveat that graphviz-devel is at version 5.0, though it's still the same type of failure.

Meanwhile, our 10.6 buildbots failed earlier in the process, with a different issue, which I haven't been able to reproduce locally. So perhaps they simply didn't get far enough to see this.

Anyhow, I've queued rebuilds for 10.6, and will review once the buildbots are caught up.

Building with MacPorts Clang 9 works fine, though.

Should we blacklist Clang versions greater than 9, for 10.6...?

[kencu (Ken)]

it doesn't build with clang-11 on any system, I think -- certainly not on Monterey.

Most likely there are some missing headers in the source, like <algorithm> and maybe others. Possibly the c++ standard needs to change to something different.

Sounds like something to bounce upstream to me.

[mascguy (Christopher Nielsen)]

Replying to kencu:

it doesn't build with clang-11 on any system, I think -- certainly not on Monterey.

Most likely there are some missing headers in the source, like <algorithm> and maybe others. Possibly the c++ standard needs to change to something different.

Sounds like something to bounce upstream to me.

I'll engage with upstream then.

In the interim, what's the best approach to fix via the portfile?

[Christopher Nielsen <mascguy@…>]

In d890cbaaf6d2326a7e82b00d0ff0c3c3ebd6f0b7/macports-ports (master):

graphviz*: temp fix: use macports-clang-9.0 for 10.6
See: #60962

[barracuda156]

Replying to Christopher Nielsen <mascguy@…>:

In d890cbaaf6d2326a7e82b00d0ff0c3c3ebd6f0b7/macports-ports (master):

graphviz*: temp fix: use macports-clang-9.0 for 10.6
See: #60962

This fix breaks Rosetta build. It should be instead:

platform darwin {
    if {${os.major} == 10 && ${build_arch} ne "ppc"} {
     compiler.whitelist          macports-clang-9.0
    }
}

Or something alike.

[Christopher Nielsen <mascguy@…>]

In a7c20d8744b000707028f79229d857e43015ae0b/macports-ports (master):

graphviz*: temp fix: accommodate rosetta build
See: #60962