Re: [dpdk-dev] [PATCH v2] build: list symbols exports in a single file

From: Thomas Monjalon <thomas@monjalon.net>
To: David Marchand <david.marchand@redhat.com>
Cc: dev <dev@dpdk.org>, Bruce Richardson <bruce.richardson@intel.com>,
	Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>,
	Narcisa Ana Maria Vasile <navasile@linux.microsoft.com>,
	Dmitry Malloy <dmitrym@microsoft.com>,
	Pallavi Kadam <pallavi.kadam@intel.com>,
	Ray Kinsella <mdr@ashroe.eu>, Neil Horman <nhorman@tuxdriver.com>,
	Matan Azrad <matan@nvidia.com>,
	Shahaf Shuler <shahafs@nvidia.com>,
	Viacheslav Ovsiienko <viacheslavo@nvidia.com>
Subject: Re: [dpdk-dev] [PATCH v2] build: list symbols exports in a single file
Date: Wed, 07 Apr 2021 00:39:59 +0200	[thread overview]
Message-ID: <2532967.nAU1F58lTY@thomas> (raw)
In-Reply-To: <CAJFAV8xW=b6Ri1A5QUATgWyNA9SfjdnNNHW+Y-DnU2wWjwAmzw@mail.gmail.com>

06/04/2021 22:09, David Marchand:
> On Tue, Apr 6, 2021 at 8:53 PM Thomas Monjalon <thomas@monjalon.net> wrote:
> >
> > About the title, it is not a single file but one file per lib/driver.
> > I think it would better convey the idea by saying "remove .def files".
> 
> Ok.
> 
> >
> > 06/04/2021 19:59, David Marchand:
> > > Rather than have two files that keeps getting out of sync, let's
> > > annotate the version.map to generate the Windows export file.
> > >
> > > Note: Existing version.map annotation achieved with:
> > > $ for dir in lib/librte_eal drivers/common/mlx5; do
> > >     ./buildtools/map-list-symbol.sh $dir/*.map |
> > >     while read file version sym; do
> > >       ! git grep -qw $sym $dir/*.def || continue;
> > >       sed -i -e "s/$sym;/$sym; # WINDOWS_NO_EXPORT/" $dir/*.map;
> > >     done;
> > >   done
> >
> > This patch keeps the exact same behaviour, right?
> 
> I checked by sorting/diff'ing previously committed .def files for EAL
> and common/mlx5 with the newly generated ones.
> No difference, once aligning the mlx5 version.map.
> 
> There is one change in behavior.
> I thought as exports.def being a subset of the map files, and this
> patch enforces it.
> It seems there were some tries at going the opposite way, looking at
> the comment in check-symbol-maps.sh.
> But looking at the current common/mlx5 state, I don't see this as a
> requirement (Linux and Windows share the same interface).
> 
> 
> > There are a lot of symbols in EAL and mlx5 which should
> > be exported for Windows too, but it is open for future fixes, right?
> 
> Not sure I get this question.
> 
> We have exports missing in EAL for some series posted on the ml
> waiting for merges.
> I suspect some recent merged changes have issues too (but did not check yet).
> This current patch does not address such issues and we need fixes
> afterwards if this is what you mean.

Yes this is what I mean.
I wanted to make clear to people in Cc list that some fixes are required
as follow-up: some symbols in EAL and mlx5 are missing for Windows.