[sword-devel] Mingw and SWORD's internal regex (was: Re: 1.6.1 outstanding items? )

[Jonathan Marsden]

Matthew Talbert wrote:

> Hey Troy,
> 
> Could you explain what you did with need_local_regex and mingw?

Heh.  I think that was actually me, or at least partly me :)  Troy 
recently went through the patches I use for my SWORD packages (thanks!) 
and basically cherry-picked everything he liked the look of, and applied 
them to the SWORD codebase.  This helps me out by reducing the number of 
patches I have to maintain/track, and (assuming my patches are decent!) 
helps SWORD out by getting this work into the hands of all SWORD users, 
not just Ubuntu/Debian users.

My suspicion is that the change you are talking about is one that I 
made, some time back.  Back in May, I think.

 > The test never actually worked, which was a real pain, ...

Did you file a bug report about that? :)

I think (from memory, I'm at work) I fixed or added a test so that it 
worked as was apparently originally intended, distinguishing between 
Mingw and the rest of the (Unix-based) world.  I then used that test to 
conditionally include a SWORD-supplied regex.h header.  I think did that 
mainly so I could drop some old "convenience copy" regex code from my 
repacked DFSG-compliant SWORD tarball, and still have SWORD build on 
Debian and Ubuntu.

Obviously, on a Linux machine there is no sense in using a convenience 
copy of an old version of GNU regex, or whatever that was in the SWORD 
source tree... because the current, up to date and bugfixed version of a 
regex library is already there in Linux.  I think that also holds true 
on OS X.

> but I don't want to build with sword's regex either, as it appears
> I'll be forced to do now (assuming the test for mingw actually works
> now).

It should work now... I'm pretty sure I tested it that far under Mingw 
as well as under Ubuntu and Debian, back when I made the change :)

But if no-one wants to use the SWORD regex at all, then logically this 
should be removed from the source tree completely and not tested for :)

> What I would like to see is a test for -lregex rather than mingw. That
> would satisfy all use cases and make things easier for me. I have
> available a regex lib which is more up-to-date than sword's copy, plus
> it's a dependency for Xiphos anyway, so I'd rather use it. A simple
> test for the regex lib would be sufficient.

Sounds reasonable if there is actually still a use case where someone 
*does* want to use the SWORD-supplied regex code (BTW, if they do, can 
it be kept updated, please?).  Embedded platforms, maybe?  If there is 
in fact no such use case at all, then I vote for removing the unwanted 
code, and just requiring an externally-supplied regex library -- keep it 
simple :)

Jonathan