[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gpsd-dev] [PATCH] Remove hard-coded LIBATH='.'. It destroys all the
From: |
Fred Wright |
Subject: |
Re: [gpsd-dev] [PATCH] Remove hard-coded LIBATH='.'. It destroys all the hard work done by calling pkg-config earlier. |
Date: |
Wed, 17 Feb 2016 14:02:29 -0800 (PST) |
On Tue, 16 Feb 2016, John Hein wrote:
> Fred Wright wrote at 17:36 -0800 on Feb 15, 2016:
[...]
> > The main issue blocking this is figuring out why all those LIBPATH='.'
> > overrides were in there in the first palce.
[...]
>
> As you know, it was committed here:
>
> ===========
> commit 6e1d6bd6f51b3a66de029792433dcbfc9fe4a048
> Author: Eric S. Raymond <address@hidden>
> Date: Tue Mar 31 09:55:06 2015 -0400
>
> Eliminate chrpath as a build dependency
> ===========
>
> No further context found that I could see. And, probably like you,
> I did some archeology to understand why LIBPATH='.' was added.
>
> It's certainly not needed to get the project directories into the lib
> search patch since it seems that scons puts '.' in the front of
> LIBPATH by default.
Well, regardless of what scons may or may not do internally, the
SConstruct script explicitly adds the current directory to LIBPATH (line
533 in the current master):
env.Prepend(LIBPATH=[os.path.realpath(os.curdir)])
> One reason I can think of to force LIBPATH='.' only is to make sure
> you only pull in libs built by the project. This means you'd have to
> build all dependencies (e.g., dbus) as part of the project. You might
> even want to link with -nostdlib in some more extreme cases.
>
> Some projects do offer the option of using a "system" lib vs an
> internal version of the same lib. It's usually when the project wants
> to use a particular "stable" version (API- & feature-wise) or to
> account for platform differences (e.g., perhaps zlib features varying
> widely when building for, say, HP-UX vs linx).
>
> In any case, I don't see support in gpsd for any such "internal" lib
> building.
Exactly.
> It's not clear to me why the LIBPATH='.' was added in the first place.
> Perhaps I'm missing the underlying reason.
In case you haven't found it yet, there's more discussion in this thread:
http://lists.nongnu.org/archive/html/gpsd-dev/2016-01/msg00074.html
http://lists.nongnu.org/archive/html/gpsd-dev/2016-02/msg00011.html
Fred Wright