boost: add boost/1.76.0 + bump b2 #5246

madebr · 2021-04-17T01:27:29Z

Specify library name and version: boost/1.76.0

Release notes: https://www.boost.org/users/history/version_1_76_0.html

No new libraries!

I've read the guidelines for contributing.
I've followed the PEP8 style guides for Python code in the recipes.
I've used the latest Conan client version.
I've tried at least one configuration locally with the
conan-center hook activated.

ghost · 2021-04-17T01:52:13Z

I detected other pull requests that are modifying boost/all recipe:

This message is automatically generated by https://github.com/ericLemanissier/conan-center-conflicting-prs so don't hesitate to report issues/improvements there.

madebr · 2021-04-17T10:33:06Z

@jgsogo @uilianries
Boost/1.71.0 failed in test_package + log contains

MSBUILD : error MSB4166: Child node "2" exited prematurely. Shutting down. Diagnostic information may be found in files in the temporary files directory named MSBuild_*.failure.txt.

So I believe this is a CI infrastructure problem.

recipes/boost/all/conanfile.py

prince-chrismc · 2021-04-18T00:15:56Z

https://github.com/conan-io/conan-center-index/blob/master/docs/changelog.md#22-february-2021---1042-cet 😕

I'd of sworn I've seen newer features being used!

recipes/boost/all/conanfile.py

madebr · 2021-04-19T14:10:06Z

1.74.0 failure remains a mystery to me 😄

prince-chrismc · 2021-04-19T14:46:13Z

Odd it passed before.... might be a bug that's been hiding!

Last last retry this time I feel it!

madebr · 2021-04-19T18:46:48Z

“Insanity is doing the same thing, over and over again, but expecting different results.”

madebr · 2021-04-19T19:04:51Z

Ok, it's official.
I'm admitting myself to a psychic center.

prince-chrismc · 2021-04-20T00:14:16Z

recipes/boost/all/conanfile.py

@@ -139,6 +139,9 @@ def export(self):
    @property
    def _min_compiler_version_default_cxx11(self):
        # Minimum compiler version having c++ standard >= 11
+        if self.settings.compiler == "apple-clang":
+            # For now, assume apple-clang will enable c++11 in the distant future


I think 12 is 14 by default

I meant to say "c++11 by default"

Yep! It's our usual problem... AFAIK, in apple-clang 12 they finally set it to c++14 by default (they skipped 11)

According to the (failed) build on apple-clang 12, it does not have (at least) c++11 support by default.
See https://www.phoronix.com/scan.php?page=news_item&px=EXT4-Casefolding-With-Encrypt

jgsogo · 2021-04-20T07:45:21Z

I think the Windows Build machines are working under a heavy load in a too constrained environment.

@jgsogo @danimtb
Perhaps c3i needs some logic to detect these problems and let it try to build again later (when the load has diminished).
Maybe when MSB4166 and/or MSB6006 are detected in the logs?

Maybe it is just a matter of the number of jobs we are running in parallel inside each machine. We might be hitting some memory limits when running these big builds in parallel. 🤔

I mean, instead of parsing the logs and try to understand the errors (something I would love to do), it easier to run less jobs in parallel in each windows machine. From the user perspective, it can be much better even if it takes a little bit more time, because the alternative is a failure plus close/open pull-request.

prince-chrismc · 2021-04-20T12:54:04Z

We might be hitting some memory limits when running these big builds in parallel.

This particular error has been on the rise, even for medium size projects.

I think reducing the degree of parallelism is acceptable, most of the time I switch tasks while waiting for results 🙈

madebr · 2021-04-20T17:47:19Z

Maybe noteworthy is that it isn't the boost build that was failing in the last builds, but the test package.
Boost's heavy use of header files makes compiling it very memory hungry.

Perhaps lowering the parallel level of building the tests will avoid the unexpected process kills?

conan-center-bot · 2021-04-20T17:56:45Z

All green in build 12 (2ba4241e4b00452fdb2f149df599fa8e80fa6d9b):

boost/1.72.0@:
All packages built successfully! (All logs)
boost/1.74.0@:
All packages built successfully! (All logs)
boost/1.73.0@:
All packages built successfully! (All logs)
boost/1.69.0@:
All packages built successfully! (All logs)
boost/1.70.0@:
All packages built successfully! (All logs)
boost/1.71.0@:
All packages built successfully! (All logs)
boost/1.75.0@:
All packages built successfully! (All logs)
boost/1.76.0@:
All packages built successfully! (All logs)

SSE4 · 2021-04-21T06:08:48Z

recipes/boost/all/conanfile.py

@@ -247,6 +263,28 @@ def config_options(self):
            # Shared builds of numa do not link on Visual Studio due to missing symbols
            self.options.numa = False

+        if tools.Version(self.version) >= "1.76.0":


probably, in future we should add C++ standard requirements to the yml (existing or new), as different libraries tend to change their requirements across boost releases

madebr · 2021-04-23T07:34:46Z

@prince-chrismc @SpaceIm
@uilianries

Can I get some reviews please?

SpaceIm · 2021-04-23T12:08:44Z

recipes/boost/all/conanfile.py

+                min_compiler_version = self._min_compiler_version_default_cxx11
+                if min_compiler_version is None:
+                    self.output.warn("Assuming the compiler supports c++11 by default")
+                elif tools.Version(self.settings.compiler.version) < min_compiler_version:
+                    disable_math()


Why not inject C++11 flag instead if math is enabled? I understand that math will be disabled if compiler default standard is less than C++11, even if it supports C++11, unless compiler.cppstd is explicitly set.

I don't think it's right thing to do. if compiler.cppstd is set explicitly, it should be passed according to the setting. if it's not defined at all, it means just follow the compiler's default, whatever it is. in that case, injecting standard different from the default will only cause confusion, or even problems (as boost changes ABI for C++11 which might be undesired).

this is what we do indirectly in almost all libs based on CMake, since usually they force CXX_STANDARD to the minimum standard required to properly build.

I hope that this won't cause an influx of issues reporting missing boost math libraries.

yes, we're doing it now, but it doesn't mean we're doing the right thing :)
I think the plan is to eventually add cppstd support for CCI and process it accordingly, instead of silently injecting -std=c++xx in some recipes.

We don't inject this flag ourself, it's already there usually in underlying build files of libraries, this will require too much patches. I think that the pressure to recipes maintainers is highly underestimated for this feature.

for boost, the flag is controlled by b2 cxxstd=<version> and it's not hard-coded. so for us, it should be a simple mapping compiler.cppstd <-> cxxstd.
for other builds systems, it will be another option (e.g. CMAKE_CXX_STANDARD in CMake or cpp_std in meson), thus similar mapping will happen in the recipe or build helper.
some projects tend to hard code specific C++ standard in their makefiles (e.g. hardcoding CXXFLAGS=-std=c++<version> or set(CMAKE_CXX_STANDARD <version>). should we do something about it?
there are two ways:

provide patches to remove hardcoded C++ standard flags

assume the hard-coded C++ standard is the only one supported by the library, thus, throw ConanInvalidConfigurations for all other compiler.cppstd values besides hard-coded one

I personally think the second approach is the most correct one.
that we're doing right now is a complete mess, and the only excuse is that CCI doesn't currently support compiler.cppstd, and if we try to do things correctly, we will be unable to build many libraries.
but so far it works fine, and we had almost no reports of it caused any actual problems.
doing things correctly, on other hand, will increase pressure on maintainers, as you pointed out, as well it will make CI more complicated and will require way more resources (and will make builds even slower, multiply boost build time by the number of active C++ standards).
probably, the benefit doesn't worth the effort right now, and there are other areas which will bring more benefit for much less effort. still, in future, some C++ standard may introduce a huge ABI break, which will make "correct" handling of C++ standard in CCI a vital and must have feature.

I personally think the second approach is the most correct one.

This is the policy we have, before CMake was the main build script it was a wild west and it's the on sensible way...

As a consumer if you build locally you expect the project to be configured as intended. It's not our place to change that value... nor test all possible combinations and patch in between cppstd

SpaceIm · 2021-04-23T12:13:00Z

recipes/boost/all/conanfile.py

@@ -16,7 +16,7 @@
 except ImportError:
    from io import StringIO

-required_conan_version = ">=1.28.0"
+required_conan_version = ">=1.33.0"


Suggested change

required_conan_version = ">=1.33.0"

required_conan_version = ">=1.32.0"

even lower ;)

CCI policy is to require latest conan version, so 1.33 is okay, no need to change

Thanks, I prefer to avoid triggering a rebuild for this trivial change 😄

It's not very important. But I disagree, required_conan_version is a useful information. It avoids cryptic errors for consumers. Moreover, think to beginners, conan is already complex, so let be honest, few people read the all documentation.

It's requiring conan 1.33.0 now. Isn't that fine?

But sure, if this is the only modification, let's skip it since it takes 4 of 5 hours to build (which is not too long actually for so many versions).

The biggest problem is that the build infrastructure is a bit unstable when building boost. I needed to trigger a few rebuilds before it could complete.

SpaceIm · 2021-04-23T12:18:34Z

recipes/boost/all/conanfile.py

+            def disable_math():
+                super_modules = self._all_super_modules("math")
+                for smod in super_modules:
+                    try:
+                        setattr(self.options, "without_{}".format(smod), True)
+                    except ConanException:
+                        pass


just to be sure to understand:

super modules is the complete list of modules which might have math in their own dependency graph?

what could fail in setattr(self.options, "without_{}".format(smod), True)? Is it not under control?

just to be sure to understand:

* super modules is the complete list of modules which might have math in their own dependency graph?

Yes, the current implementation is quick and dirty and I don't guarantee it working under all combinations though.

* what could fail in `setattr(self.options, "without_{}".format(smod), True)`? Is it not under control?

The dependencies section of the dependencies-x.yy.yml file contains modules that are not valid options of boost.build (configure_options section). So I just ignore these.

* boost: bump b2 build dependency * boost: add boost/1.76.0 * boost: add 1.76.0 to config.yml * boost: move check to validate() + Boost.Math requires c++11 * boost: lower minimum required conan version * boost: remove dead comment * boost: assume apple-clang will never implement c++11

madebr added 2 commits April 17, 2021 03:21

boost: bump b2 build dependency

bbf27cc

boost: add boost/1.76.0

8ff7252

This comment has been minimized.

Sign in to view

conan-center-bot added the Failed label Apr 17, 2021

boost: add 1.76.0 to config.yml

b29e63a

ghost mentioned this pull request Apr 17, 2021

[service] Potentially conflicting recipes #3571

Closed

This comment has been minimized.

Sign in to view

madebr closed this Apr 17, 2021

madebr reopened this Apr 17, 2021

This comment has been minimized.

Sign in to view

boost: move check to validate() + Boost.Math requires c++11

6771590

madebr commented Apr 17, 2021

View reviewed changes

recipes/boost/all/conanfile.py Outdated Show resolved Hide resolved

This comment has been minimized.

Sign in to view

prince-chrismc previously approved these changes Apr 18, 2021

View reviewed changes

madebr commented Apr 18, 2021

View reviewed changes

recipes/boost/all/conanfile.py Outdated Show resolved Hide resolved

boost: lower minimum required conan version

8d92874

madebr dismissed prince-chrismc’s stale review via 8d92874 April 18, 2021 09:14

boost: remove dead comment

6a204f2

This comment has been minimized.

Sign in to view

conan-center-bot added the Unexpected Error label Apr 18, 2021

boost: assume apple-clang will never implement c++11

2ba4241

This comment has been minimized.

Sign in to view

conan-center-bot added the Unexpected Error label Apr 19, 2021

madebr closed this Apr 19, 2021

madebr reopened this Apr 19, 2021

This comment has been minimized.

Sign in to view

conan-center-bot removed Unexpected Error Failed labels Apr 19, 2021

prince-chrismc reviewed Apr 20, 2021

View reviewed changes

prince-chrismc approved these changes Apr 20, 2021

View reviewed changes

github-actions bot mentioned this pull request Apr 20, 2021

Pull Requests Ready to be Reviewed prince-chrismc/conan-center-index-pending-review#1

Open

SSE4 requested a review from uilianries April 21, 2021 06:07

SSE4 reviewed Apr 21, 2021

View reviewed changes

SSE4 approved these changes Apr 21, 2021

View reviewed changes

SpaceIm reviewed Apr 23, 2021

View reviewed changes

uilianries approved these changes Apr 26, 2021

View reviewed changes

conan-center-bot merged commit a5600b9 into conan-io:master Apr 27, 2021

madebr deleted the boost_1_76_0 branch April 27, 2021 10:58

uilianries mentioned this pull request Jun 15, 2022

[package] <proposal>/<7.3.1>: Unable to run conan install for newer boost versions #11207

Closed

	required_conan_version = ">=1.33.0"
	required_conan_version = ">=1.32.0"

boost: add boost/1.76.0 + bump b2 #5246

boost: add boost/1.76.0 + bump b2 #5246

Conversation

madebr commented Apr 17, 2021 • edited Loading

This comment has been minimized.

ghost commented Apr 17, 2021 • edited by ghost Loading

This comment has been minimized.

madebr commented Apr 17, 2021

This comment has been minimized.

This comment has been minimized.

prince-chrismc commented Apr 18, 2021

This comment has been minimized.

This comment has been minimized.

madebr commented Apr 19, 2021

prince-chrismc commented Apr 19, 2021

madebr commented Apr 19, 2021

This comment has been minimized.

madebr commented Apr 19, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jgsogo commented Apr 20, 2021 • edited Loading

prince-chrismc commented Apr 20, 2021

madebr commented Apr 20, 2021

conan-center-bot commented Apr 20, 2021

Choose a reason for hiding this comment

madebr commented Apr 23, 2021

SpaceIm Apr 23, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SpaceIm Apr 23, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SpaceIm Apr 23, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

madebr commented Apr 17, 2021 •

edited

Loading

ghost commented Apr 17, 2021 •

edited by ghost

Loading

jgsogo commented Apr 20, 2021 •

edited

Loading

SpaceIm Apr 23, 2021 •

edited

Loading

SpaceIm Apr 23, 2021 •

edited

Loading

SpaceIm Apr 23, 2021 •

edited

Loading