tests/hdf5: set C++ standard to c++11, which is required for macos CI

[dcbaker]

No description provided.

[bruchar1]

Yes, I realized that too. Unfortunately, that doesn't fix the broken CI caused by the broken homebrew package...

[dcbaker]

Yeah, we can't fix homebrew. sigh

[eli-schwartz]

It's easy to fix homebrew. Homebrew just needs a PR to revert their hdf5 package to use autotools instead of cmake -- like everyone else builds it.

hdf5 has a totally broken h5cc when built with cmake -- and possibly other issues as well. It's been a long time, but in the past there used to be serious concerns that the cmake build was altogether totally broken on unix, and as a result, no one built with it or reported issues with it so I'd be hesitant to claim it's well tested even.

[jpakkane]

Undefined symbols for architecture x86_64:
  "_H5close", referenced from:
      _main in main.c.o
  "_H5get_libversion", referenced from:
      _main in main.c.o
  "_H5open", referenced from:
      _main in main.c.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
[3/4] c++ -Iexecpp.p -I. '-I../test cases/frameworks/25 hdf5' -I/usr/local/include -fdiagnostics-color=always -D_GLIBCXX_ASSERTIONS=1 -Wall -Winvalid-pch -std=c++11 -O0 -g -MD -MQ execpp.p/main.cpp.o -MF execpp.p/main.cpp.o.d -o execpp.p/main.cpp.o -c '../test cases/frameworks/25 hdf5/main.cpp'
ninja: build stopped: subcommand failed.

[eli-schwartz]

Yes, this is Homebrew/homebrew-core#159691

[dcbaker]

It looks like using h5cc is deprecated, so we should deprecate that support and skip the test

[eli-schwartz]

No, because they refuse to support pkg-config in the build system that all distributors other than "homebrew in the past week" use.

Which means that projects have zero choice, if they want to build against the hdf5 packages in any distro, they must use h5cc -show, there is no other option -- so that's what they do.

It also means, of course, that meson cannot check for hdf5 with pkg-config if there are no operating systems where hdf5 comes with .pc files (or cmake files). :)

(There are a couple people that build hdf5 with autotools and then manually install some pkg-config files, e.g. by configuring cmake in a temporary directory and then copying the .pc files out of cmake's working directory.)

[dcbaker]

Nix uses cmake... But it's broken in different ways. I guess I'll send a patch for the nixos build and get that working. What a cluster. I guess what we really should do here is check that we got something reasonable from h5cc and return a graceful message if h5cc is the cmake version. I can do that

[eli-schwartz]

I guess what we really should do here is check that we got something reasonable from h5cc and return a graceful message if h5cc is the cmake version. I can do that

From 2bc592b0e426afab88bd5a7abf3067ea214ff198 Mon Sep 17 00:00:00 2001
From: Eli Schwartz <eschwartz93@gmail.com>
Date: Wed, 17 Jan 2024 13:23:56 -0500
Subject: [PATCH] dependencies: hdf5: mark configtool dependency not-found for
 cmake build

When hdf5 is built with cmake instead of autotools, it makes a number of
weird changes. What we care about in particular is that h5cc exists but
doesn't work -- it happily ignores -show and tries to compile stuff,
then leaves us with a dependency that has no libraries, and fails when
running `ninja`.

See: #12748
Signed-off-by: Eli Schwartz <eschwartz93@gmail.com>
---
 mesonbuild/dependencies/hdf5.py | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/mesonbuild/dependencies/hdf5.py b/mesonbuild/dependencies/hdf5.py
index a00b5988b..622653025 100644
--- a/mesonbuild/dependencies/hdf5.py
+++ b/mesonbuild/dependencies/hdf5.py
@@ -122,13 +122,20 @@ class HDF5ConfigToolDependency(ConfigToolDependency):
         # and then without -c to get the link arguments.
         args = self.get_config_value(['-show', '-c'], 'args')[1:]
         args += self.get_config_value(['-show', '-noshlib' if self.static else '-shlib'], 'args')[1:]
+        found = False
         for arg in args:
             if arg.startswith(('-I', '-f', '-D')) or arg == '-pthread':
                 self.compile_args.append(arg)
             elif arg.startswith(('-L', '-l', '-Wl')):
                 self.link_args.append(arg)
+                found = True
             elif Path(arg).is_file():
                 self.link_args.append(arg)
+                found = True
+
+        # cmake h5cc is broken
+        if not found:
+            raise DependencyException('HDF5 was built with cmake instead of autotools, and h5cc is broken.')
 
     def _sanitize_version(self, ver: str) -> str:
         v = re.search(r'\s*HDF5 Version: (\d+\.\d+\.\d+)', ver)
-- 
2.43.0

[thesamesam]

We hit this in Gentoo before as well (https://bugs.gentoo.org/813924, https://bugs.gentoo.org/820158).

See also HDFGroup/hdf5#1814.

[dcbaker]

@eli-schwartz Is it really correct to raise in that case? shouldn't it just be set to self.found = false; return and let the normal code go on to say "couldn't find anything, sorry"?

[eli-schwartz]

Raising a DependencyException would also mean that it logs the exception text in the log file, and if it cannot resolve via a different method then detect() will use that exception text as part of the "error: dependency hdf5 not found" message.

[eli-schwartz]

I think this was fixed in #12802