X-Authentication-Warning: delorie.com: mail set sender to djgpp-bounces using -f
X-Recipient: djgpp AT delorie DOT com
Message-ID: <5556D918.4050302@iki.fi>
Date: Sat, 16 May 2015 08:43:52 +0300
From: "Andris Pavenis (andris DOT pavenis AT iki DOT fi)" <djgpp AT delorie DOT com>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.6.0
MIME-Version: 1.0
To: djgpp AT delorie DOT com
Subject: Re: DXE related problem in DJGPP v2.05 beta
References: <554F1325 DOT 2040805 AT iki DOT fi>
In-Reply-To: <554F1325.2040805@iki.fi>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 7bit
Reply-To: djgpp AT delorie DOT com

On 05/10/2015 11:13 AM, Andris Pavenis (andris DOT pavenis AT iki DOT fi) wrote:
> I had to update gettext and libiconv to try to build current early development snapshot of 
> GCC-6.0.0.
> Build failed due to GCC build process defining explicitly LD_LIBRARY_PATH.
>
> Having only one environment variable is not sufficient in this case. For example in Linux system 
> wide
> shared library lookup directories are managed by ldconfig anbd ldconfig.conf. Directories 
> specified in
> LD_LIBRARY_PATH are searched before system ones. We need something similar for DJGPP
>
> Attached patch add additional DXE search path that cannot be overridden by user from environment
> directly. Unfortunately that causes compatibility problem with already existing packages which use
> DXE. I had as result no other choice rather than defining LD_LIBRARY_PATH to the same value in 
> earlier
> way. Some small optimization was added to dlopen() to avoid doing unnecessary lookup twice (not very
> foolproof, but harm from looking up twice is not very noticeable)
>
> I will commit it unless there are objections
>
> Change will require rebuilding packages which uses DXE to fully use this change. There should be no
> regressions without rebuilding, but without it freely overriding LD_LIBRARY_PATH will cause DJGPP 
> own
> DXE loading to fail (as it is now)
>
> Andris

Committed to CVS.

Mentioned small optimization was somewhere lost on the way (it was implemented, but not included in 
patch).
We do not perhaps need it

>
>
> PS. Noticed also that file name handling in dlopen() assumes non-LFN filename without directory part
> (length 12). That also should be fixed. I did not want to mix more than one thing into change at 
> the same
> time, so I left this thing for now.
>
>
I should fix this also in a separate commit

Andris