Re: [PATCH] posix: if glob has a trailing slash match directories only.

[Dmitry Goncharov <dgoncharov at users dot sf dot net>]

On Wed, Nov 29, 2017 at 03:01:46PM -0800, Paul Eggert wrote:
> On 11/29/2017 01:50 PM, Dmitry Goncharov wrote:
> > One option is we can decide the patch is compatible and apply it.
> 
> It's compatible only if we ignore performance. But I don't think that 
> would be right: I expect that some callers expect GLOB_ONLYDIR to be 
> fast (because that's what the Glibc documentation says), and that they 
> will be disappointed by this performance regression.

Paul, i am missing where you see the performance hit.
The current contract makes the caller check after the library by calling stat
on each entry. A stricter contract like
"if GLOB_ONLDIR is set then glob returns only directories"
is more efficient on the filesystems which set the type.
On the filesystems which don't set the type stat'd be called once for each
entry by the library rather than by the caller which keeps the performance
intact.

> 
> > If the existing semantics has to be preserved then we can introduce
> > another flag and use the new flag instead of GLOB_ONLYDIR in
> >
> >    if (pattern[0] && pattern[strlen (pattern) - 1] == '/')
> >      flags |= GLOB_ONLYDIR;
> 
> That would be a better approach. Another possibility might be to change 
> the last assignment to:
> 
>      flags |= GLOB_ONLYDIR | GLOB_MARK;
> 
> and then at the end, filter out all matches that aren't marked with 
> trailing '/'. This would avoid creating a new GLOB_XXX option and would 
> probably be easier to implement.

Thansk for the pointer. Will have a look.


regards, Dmitry