On Wed, 2020-10-21 at 23:31 +0200, Ludovic Courtès wrote:
> Hi,
> 
> Eric Bavier <bavier@posteo.net> skribis:
> 
> > On Wed, 2020-10-21 at 17:35 +0200, Ludovic Courtès wrote:
> > > Hi Eric,
> > > 
> > > Eric Bavier <bavier@posteo.net> skribis:
> > > 
> > > >                (mkdir-p (dirname result))
> > > >                (apply invoke #$compiler "-std=gnu99" "-static" "-Os" "-g0" "-Wall"
> > > >                       "run.c" "-o" result
> > > > +                     (string-append "-DWRAPPER_PROGRAM=\"" result "\"")
> > > 
> > > Can we just write (canonical-path result) here?  That way we wouldn’t
> > > need ‘find-input-files’ and related changes.
> > > 
> > 
> > Guile's canonicalize-path will raise and error if the path does not
> > already exist.  We could create a dummy file at result, then call
> > canonicalize-path? but that seems clumsier than forming a canonical
> > name in the first place?
> 
> The patch you sent has ‘find-files’ use ‘stat’ instead of the trailing
> slash.  It introduces a difference: by using ‘stat’ all the way, it
> follows all symlinks, not just the higher-level ones.  I don’t know if
> this could have undesired implications, like wrapping the same file
> twice because there’s a symlink pointing to it.  (Or am I too paranoid?)
> 
> To be on the safe side, we could write:
> 
>   (string-append (canonical-path (dirname) result) "/" result)
> 
> WDYT?

I don't like the "stat" change either.  Paranoia is not misplaced, I
think.

Following are current two patches.  I can add copyright to run-in-
namespace.c if you think it's needed.

`~Eric