hawq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Lei Chang <lei_ch...@apache.org>
Subject Re: enforce -Werror (if gcc) in hawq?
Date Thu, 08 Sep 2016 06:21:46 GMT
I think consistent behavior is much better to avoid confusions.

different options at different places looks strange. And only people on
this thread might be able follow this discussion. And maybe someday, they
will forget this too.

With new contributors/committers added, it is much more easier for them to
follow simple/default options.

Cheers
Lei




On Thu, Sep 8, 2016 at 2:12 PM, Hong Wu <xunzhangthu@gmail.com> wrote:

> I still have some reservation and do not recommend that. For the original
> purpose of adding -Werror option, I think following alternative solutions
> are better:
> 1. Try to avoid warnings duringreviewing process.
> 2. Add -Werror in travis CI script. If it fails to compile in travis, it is
> not allowed to check the pull request in.
>
> Above solutions could both be added and not conflict with reasons mentioned
> by Paul. Any comments?
>
> Best
> Hong
>
> 2016-09-08 13:41 GMT+08:00 Paul Guo <paulguo@gmail.com>:
>
> > For HAWQ, it is an open source project so I think from this respective,
> we
> > should
> > try to make all contributors have  the same development/test experience,
> > else
> > main contributors (e.g. committer) will have to waste time to fix those
> > errors introduced
> > by some contributors who know nothing about the context. Making the
> > build/test
> > less complex is our target. This is my answer to Hong and concern 1) of
> > Zhanwei
> >
> > Back to the -Wall and -Werror details. To my best of knowledge,
> > 1) gcc has been adjusting (probably mainly adding) the warning cases.
> > 2) -Wall actually enables just part of warnings and the set of warning
> > could
> >     vary in various gcc versions.
> >
> > But I still think it should be enabled by default. The reasons are:
> >
> > 1) We actually just mainly tested on centos6.x by now. We actually did
> not
> > test on
> >     mac and centos7 that much. So "officially" we do not support that
> many
> > os and
> >     the fix will be not that tedious.
> >
> > 2) In the future if we want to have more os support, "gcc -Werror" is
> just
> > one small
> >     step of the whole work. I do not think this introduces much more
> work.
> >
> > 3) Although different gcc versions have different -Wall definitions, I
> > think the fixes
> >     for various gcc versions should be quite mutually tolerated.
> >
> > 4) Even above solutions do not work finally, we could specifies common
> > warnings
> >    via -Werror=, e.g. -Werror=-Wunused-variable,...
> >
> >
> >
> > 2016-09-06 10:31 GMT+08:00 Hong Wu <xunzhangthu@gmail.com>:
> >
> > > I strongly agree with ZhanWei's opinion, I think it is much more
> flexible
> > > to honor environment variable to do that. Take CMake system for
> example,
> > it
> > > is usual to honor "-DCMAKE_XXX" when doing configuration.
> > >
> > > As ZhanWei's recommendation, I think we should try it ourselves to make
> > > sure good coding style and code quality. For customers and contributors
> > > from community, it is better not doing that by default.
> > >
> > > Thanks
> > > Hong
> > >
> > > 2016-09-06 9:52 GMT+08:00 Zhanwei Wang <wangzw@apache.org>:
> > >
> > > > Hi Hong
> > > >
> > > > I removed -Werror flag when we push HAWQ to open source. In Pivotal
> we
> > > > build HAWQ on specific OS with specific GCC and dependent libraries.
> > > > -Werror flag worked fine since we fix all warnings. But for a open
> > source
> > > > project. It is impossible to make users and contributors to build
> HAWQ
> > on
> > > > specific OS and compiler just like what we did before. We cannot say
> we
> > > > just support HAWQ on Centos 6 with GCC 4.4.2.
> > > >
> > > > So if we enforce -Werror flag, I guess the build will fail on many
> > > > environments.
> > > >
> > > > I propose three solutions and let’s discuss which is better.
> > > >
> > > > 1) Do not enforce -Werror flag but add it to our test (concourse and
> > > > Travis) like this.
> > > > ./configure —prefix=/path/to/install CFLAGS=“-Werror”
> > > >
> > > > By this way we can enforce -Werror flag on our tested environment.
> > > >
> > > >
> > > > 2) Only enforce -Werror flag on development build, remove it on
> release
> > > > build.
> > > >
> > > >
> > > > 3) Enforce -Werror flag and we setup more test environments(different
> > > > versions of CentOS Ubuntu SUSE with default compiler and latest GCC
> and
> > > > MacOS with default compiler and latest clang)
> > > >
> > > >
> > > > Any comments?
> > > >
> > > >
> > > > Best Regards
> > > >
> > > > Zhanwei Wang
> > > > wangzw@apache.org
> > > >
> > > >
> > > >
> > > > > 在 2016年9月6日,上午9:22,Ed Espino <espino@apache.org>
写道:
> > > > >
> > > > > +1
> > > > >
> > > > > Have we considered setting up separate public Concourse pipelines
> to
> > > try
> > > > > the various build scenarios.
> > > > >
> > > > > -=e
> > > > >
> > > > > On Tue, Sep 6, 2016 at 12:58 AM, Hong Wu <xunzhangthu@gmail.com>
> > > wrote:
> > > > >
> > > > >> Ming's comment makes sense, but I think it is another thread.
I
> have
> > > > >> already tried those[1] but there are some further works need
to
> > do[2].
> > > > >>
> > > > >> [1]
> > > > >> - clang analysis scan report: I uploaded the result of not that
> > fresh
> > > > HAWQ
> > > > >> in my personal link, please check the report out here
> > > > >> <http://xunzhangthu.org/tmp/hawq_check> if you are interested.
> > > > >> - coverity scan: the latest reported is here
> > > > >> <https://scan.coverity.com/projects/apache-incubator-hawq>.
If
> you
> > > > want to
> > > > >> see the defects in detail, you need to submit a permission
> request.
> > > > >>
> > > > >> [2]
> > > > >> - Make clang analysis scan reported generating periodically and
> > > > publishing
> > > > >> to the public automatically. I suggest we donate a domain such
as
> > > > hawq.io
> > > > >> and
> > > > >> a host for it, also for automatically publishing the report,
we
> need
> > > to
> > > > >> write a web service to reply the requests and transmitting html
> > data.
> > > > >>
> > > > >> - Travis CI script have already integrated the coverity scan
> service
> > > > using
> > > > >> github webhook. we need to create a coverity_scan branch for
hawq
> > and
> > > > then
> > > > >> modify the .travis.yml file. I have done that under Redhat
> > > environment.
> > > > >> While the only environment Travis server supports for Linux is
> > > > >> Ubuntu/Debian. Although hawq could be built under Ubuntu, it
needs
> > > extra
> > > > >> effort to extend the .travis.yml script to support that. For
osx
> > > > >> environment, I am not sure what the problem is, the issue is
that
> > the
> > > > >> report could not be sent to the coverity scan server
> automatically.
> > > > >>
> > > > >> ps: I think Chunlin <https://github.com/wcl14> is starting
> working
> > on
> > > > the
> > > > >> defects generated by coverity.
> > > > >>
> > > > >>
> > > > >> Back to main thread mentioned by Paul, I think we should just
try
> to
> > > > open
> > > > >> the flag and discuss errors after opening -Werror.
> > > > >>
> > > > >> Best
> > > > >> Hong
> > > > >>
> > > > >>
> > > > >> 2016-09-05 21:51 GMT+08:00 Ming Li <mli@pivotal.io>:
> > > > >>
> > > > >>> Good suggestion.
> > > > >>>
> > > > >>> However, IMHO, we may need to firstly enable coverity scan
check
> or
> > > > clang
> > > > >>> analysis scan. Also we should make the output of these check
on a
> > > > public
> > > > >>> server so that all contributor can access them.
> > > > >>>
> > > > >>> On Mon, Sep 5, 2016 at 6:05 PM, Paul Guo <paulguo@gmail.com>
> > wrote:
> > > > >>>
> > > > >>>> -Werror
> > > > >>>> Make all warnings into errors.
> > > > >>>> I've seen many cases (not just hawq) before that ignoring
gcc
> > > warning
> > > > >>> leads
> > > > >>>> to bugs. I'm wondering we should add the option for the
gcc
> case.
> > > > Given
> > > > >>>> there may be a lot of warnings when building the common
postgres
> > > code
> > > > >> in
> > > > >>>> hawq, we could at least enforce it in our own code at
first
> > > > >>>> (src/backend/cdb, src/backend/resourcemanager, src/test/feature,
> > > other
> > > > >>>> directories?)? Any suggestion?
> > > > >>>>
> > > > >>>
> > > > >>
> > > > >
> > > > >
> > > > >
> > > > > --
> > > > > *Ed Espino*
> > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message