wildcards (was Re: a problem I'm having with rsync-4.5.4)

[Dave Dykstra]

Well the modified wildmat certainly looks simpler but I think it's going to
be awfully hard to test for all the kinds of cases rsync is likely to run
into.  I don't have time to look at it very closely, but I would say it's
definitely too risky for 2.5.6.  The fact that it is based on previously
tested code is encouraging though.

- Dave Dykstra


On Wed, May 08, 2002 at 10:01:12AM -0700, Wayne Davison wrote:
> On Wed, 8 May 2002, Dave Dykstra wrote:
> > And in fact I think the non-wildcard-matching code actually succeeds,
> > doesn't it?
> 
> Yes, sorry for the unclear sentence.
> 
> > I doubt it's worth trying to fix the fnmatch() code, because fnmatch
> > is a standard function and it would be a lot of work to maintain our
> > own modified version.
> 
> The thing is, we don't really want what fnmatch() does.  I hadn't
> realized until I read the old bug email you cited that rsync can treat
> "*" characters as "**" (even though we mention that fact in the man
> page).  I think this in itself is a good reason to replace the fnmatch()
> call with something that can distinguish between the two idioms.  The
> wildmat.c code I used to create such a solution is much simpler than the
> GNU fnmatch.c code, and the test suite I used to test it indicates that
> it is working quite well.  This change also means that we could get rid
> of the fnmatch() compatibility code that is in the lib dir, and I've
> already gotten rid of the code that tested if fnmatch() was broken on
> the current machine (meaning less worries for oddball architectures).
> 
> As for the anchoring discrepancy, it's a simple matter to add a loop to
> the match test to check if the pattern matches later in the string.
> (This could even be done for the current fnmatch() call).  This is as
> efficient as adding "**/" to the front of the string (without affecting
> its matching of base-directory names).
> 
> Appended is a patch that tweaks my wildmat patch.  I've optimized the
> wildmat_tail() routine to avoid an extra (useless) match loop if the
> pattern starts with "**".
> 
> ..wayne..
> 
> ---8<------8<------8<------8<---cut here--->8------>8------>8------>8---
> Index: lib/wildmat.c
> --- save/lib/wildmat.c	Wed May  8 09:36:28 2002
> +++ ./lib/wildmat.c	Wed May  8 09:20:09 2002
> @@ -103,6 +103,8 @@
>  int
>  wildmat_tail(const char *text, const char *p)
>  {
> +    if (strncmp(p, "**", 2) == 0)
> +	return wildmat(text, p);
>      while (1) {
>  	if (wildmat(text, p))
>  	    return TRUE;
> ---8<------8<------8<------8<---cut here--->8------>8------>8------>8---