Re: lt_dlerror changes

[Gary V. Vaughan]

Hi Chuck,

[Warning: thread hijack - please start a new thread and paste in any context 
you need if you care to reply further.]

On 25 Jun 2010, at 06:23, Charles Wilson wrote:
> On 6/17/2010 4:54 PM, Peter O'Gorman wrote:
>> Well, this is what I ended up with, it does not change the currently
>> documented saving of error messages until lt_dlerror() is called, it
>> copies the error message to ensure that we don't return garbage when
>> lt_dlerror is called. I think I also got all the places where we were
>> setting file not found.
>> 
>> Still, I am not overjoyed with this.
> 
> I'm sorry I sidetracked this thread, with the discussion on *using* this
> patch to track down an error in an different part of libltdl on cygwin.
> 
> But...
> 
> I think this patch is a decent, incremental approach to fixing the error
> reporting in libltdl.  Unless somebody else is working on something
> better in secret, I think this version ought to be pushed.

Agreed.

As an aside, I *am* working on something "in secret" actually.  I finally gave 
up trying to fully grok the spaghetti code at the core of libltdl, and with all 
the lessons I learned in teasing out the other parts of it between 1.5.x and 
2.2.x, I am in the early stages of a complete rewrite.  I have spent so many 
hours staring at the original code that I'm now of the opinion that I can put 
out a cleaner, more stable, ground up rewrite in less time than it would take 
to bring the incumbent implementation any closer to sanity than what we already 
have.  A ton of smart talented programmers have been defeated by the error 
reporting of the current code base alone: Here we are 8 years down the line... 
and *still* we don't have it working quite right.  If I'd started this rewrite 
a year or two ago, we'd have something better by now already, so any effort I 
would otherwise be investing in the current libltdl tree goes to the rewrite 
instead.

However. with this rewrite, something useable is a long way off yet. I'm still 
on the fence regarding whether dlfcn.h is even a good starting point for the 
API... any nuggets of wisdom or insight would be appreciated more at this point 
than they will when I post a draft implementation in several months.

More here: http://lists.gnu.org/archive/html/libtool/2010-06/msg00057.html

Cheers,
-- 
Gary V. Vaughan (address@hidden)