[PATCH v3] Cygwin: respect PC_SYM_FOLLOW and PC_SYM_NOFOLLOW_REP with inner links

Jeremy Drake cygwin@jdrake.com
Tue Jul 6 17:38:02 GMT 2021


On Tue, 6 Jul 2021, Corinna Vinschen wrote:

> This formatting is just ugly.  I suggest to move the PC_SYM_* test
> to the block after the 32 bit code and reuse the existing braces,
> just with adapted indentation, i. e.:

+1.  I was trying to avoid reformatting otherwise unchanged lines to
reduce patch size.

> > @@ -3704,7 +3708,8 @@ chdir (const char *in_dir)
> >
> >        /* Convert path.  PC_NONULLEMPTY ensures that we don't check for
> >          NULL/empty/invalid again. */
> > -      path_conv path (in_dir, PC_SYM_FOLLOW | PC_POSIX | PC_NONULLEMPTY);
> > +      path_conv path (in_dir, PC_SYM_FOLLOW | PC_POSIX | PC_NONULLEMPTY
> > +                             | PC_SYM_NOFOLLOW_REP);
> >        if (path.error)
> >         {
> >           set_errno (path.error);
>
> I'm still not convinced that we should do this.  I'm pretty certain this
> will result in problems in Cygwin processes when you least expect them.
>
> Consider that the output of getcwd and realpath/readlink on the same
> path may differ after this patch.  Using PC_SYM_NOFOLLOW_REP like this
> also changes the normal sym follow handling for the last path component
> in path_copnv::check, potentially.
>
> This looks like here be dragons.  A good solution would change the
> used native tools to allow paths > MAX_PATH finally, or to use other,
> equivalent tools already allowing that.

I am not convinced that this even completely solved the issues I was
seeing, or some of the reports of issues with unc paths suddenly showing
up instead of mapped drives in native tools that weren't expecting them.

But, I do think respecting the PC_SYM_NOFOLLOW_REP flag for inner links is
correct, and I am sending a new version.


More information about the Cygwin-patches mailing list