Exclude non-executables from command completion

Exclude non-executables from command completion

[Eike von Seggern]

Dear list,

I just changed my default shell from bash to zsh and I haven't regretted
it so far. There's only one thing that comes to my mind every now and
then:

Is it possible, that command completion only completes on commands that
are actually executable? As it happens I have to work on machines that
have non-executables in directories in $PATH.

I already tried something like

zstyle ':completion:*:complete:-command-::commands' ignored-patterns '*(f:u-x:)'

but it didn't work.

Cheers,
    eike

Re: Exclude non-executables from command completion

[Peter Stephenson]

On Thu, 17 Sep 2009 18:08:26 +0200
Eike von Seggern <eikevons@yahoo.de> wrote:
> Is it possible, that command completion only completes on commands that
> are actually executable? As it happens I have to work on machines that
> have non-executables in directories in $PATH.

I'm not sure why we don't check that commands are executables, was there
ever a reason?

P.S. we are woefully dependent on PATH_MAX.  That's one of those things it
would be really good to fix that's not going to change without a ten-fold
increase in people working on the shell.

Index: Src/hashtable.c
===================================================================
RCS file: /cvsroot/zsh/zsh/Src/hashtable.c,v
retrieving revision 1.31
diff -u -r1.31 hashtable.c
--- Src/hashtable.c	11 Feb 2009 20:42:16 -0000	1.31
+++ Src/hashtable.c	18 Sep 2009 10:22:05 -0000
@@ -630,20 +630,50 @@
 {
     Cmdnam cn;
     DIR *dir;
-    char *fn;
+    char *fn, *unmetadir, *pathbuf, *pathptr;
+    int dirlen;
 #if defined(_WIN32) || defined(__CYGWIN__)
     char *exe;
 #endif /* _WIN32 || _CYGWIN__ */
 
-    if (isrelative(*dirp) || !(dir = opendir(unmeta(*dirp))))
+    if (isrelative(*dirp))
 	return;
+    unmetadir = unmeta(*dirp);
+    if (!(dir = opendir(unmetadir)))
+	return;
+
+    dirlen = strlen(unmetadir);
+    pathbuf = (char *)zalloc(dirlen + PATH_MAX + 2);
+    sprintf(pathbuf, "%s/", unmetadir);
+    pathptr = pathbuf + dirlen + 1;
 
     while ((fn = zreaddir(dir, 1))) {
 	if (!cmdnamtab->getnode(cmdnamtab, fn)) {
-	    cn = (Cmdnam) zshcalloc(sizeof *cn);
-	    cn->node.flags = 0;
-	    cn->u.name = dirp;
-	    cmdnamtab->addnode(cmdnamtab, ztrdup(fn), cn);
+	    char *fname = ztrdup(fn);
+	    struct stat statbuf;
+	    int add = 0, dummylen;
+
+	    unmetafy(fn, &dummylen);
+	    if (strlen(fn) > PATH_MAX) {
+		/* Too heavy to do all the allocation */
+		add = 1;
+	    } else {
+		strcpy(pathptr, fn);
+		/*
+		 * This is the same test as for the glob qualifier for
+		 * executable plain files.
+		 */
+		if (stat(pathbuf, &statbuf) == 0 &&
+		    S_ISREG(statbuf.st_mode) && (statbuf.st_mode & S_IXUGO))
+		    add = 1;
+	    }
+	    if (add) {
+		cn = (Cmdnam) zshcalloc(sizeof *cn);
+		cn->node.flags = 0;
+		cn->u.name = dirp;
+		cmdnamtab->addnode(cmdnamtab, fname, cn);
+	    } else
+		zsfree(fname);
 	}
 #if defined(_WIN32) || defined(__CYGWIN__)
 	/* Hash foo.exe as foo, since when no real foo exists, foo.exe
@@ -664,6 +694,7 @@
 #endif /* _WIN32 || __CYGWIN__ */
     }
     closedir(dir);
+    zfree(pathbuf, dirlen + PATH_MAX + 2);
 }
 
 /* Go through user's PATH and add everything to *


-- 
Peter Stephenson <pws@csr.com>            Software Engineer
Tel: +44 (0)1223 692070                   Cambridge Silicon Radio Limited
Churchill House, Cambridge Business Park, Cowley Road, Cambridge, CB4 0WZ, UK


Member of the CSR plc group of companies. CSR plc registered in England and Wales, registered number 4187346, registered office Churchill House, Cambridge Business Park, Cowley Road, Cambridge, CB4 0WZ, United Kingdom

Re: Exclude non-executables from command completion

[Eike von Seggern]

On Fri, Sep 18, 2009 at 11:27 +0100, Peter Stephenson wrote:
> On Thu, 17 Sep 2009 18:08:26 +0200
> Eike von Seggern <eikevons@yahoo.de> wrote:
> > Is it possible, that command completion only completes on commands that
> > are actually executable? As it happens I have to work on machines that
> > have non-executables in directories in $PATH.
> 
> I'm not sure why we don't check that commands are executables, was there
> ever a reason?
> 
> P.S. we are woefully dependent on PATH_MAX.  That's one of those things it
> would be really good to fix that's not going to change without a ten-fold
> increase in people working on the shell.
 
Hi Peter,

I've had a look in the current HEAD version of hashtable.c, though I
have not compiled and tried it hands-on. But if I'm right, only
regular&&executable files are added to the hashtable. At least on my
Debian system, there are a lot of symlinks in /usr/bin to keep track of
different versions. So I think you should keep symlinks in mind, aswell.
If that's already the case, sorry for this, but my libc knowledge is
really small.

Cheers,
    eike

Re: Exclude non-executables from command completion

[Peter Stephenson]

Eike von Seggern wrote:
> I've had a look in the current HEAD version of hashtable.c, though I
> have not compiled and tried it hands-on. But if I'm right, only
> regular&&executable files are added to the hashtable.

I think we're OK: we test using the information returned by stat whether
the file is regular and executable, but the information was retrieved
using the stat() call (not lstat()), so any symbolic links will have
been dereferenced (unless they're broken, of course).  Tests seem to
confirm this is working.

-- 
Peter Stephenson <pws@csr.com>            Software Engineer
Tel: +44 (0)1223 692070                   Cambridge Silicon Radio Limited
Churchill House, Cambridge Business Park, Cowley Road, Cambridge, CB4 0WZ, UK


Member of the CSR plc group of companies. CSR plc registered in England and Wales, registered number 4187346, registered office Churchill House, Cambridge Business Park, Cowley Road, Cambridge, CB4 0WZ, United Kingdom

Re: Exclude non-executables from command completion

[Eike von Seggern]

On Fri, Oct 02, 2009 at 11:04 +0100, Peter Stephenson wrote:
> Eike von Seggern wrote:
> > I've had a look in the current HEAD version of hashtable.c, though I
> > have not compiled and tried it hands-on. But if I'm right, only
> > regular&&executable files are added to the hashtable.
> 
> I think we're OK: we test using the information returned by stat whether
> the file is regular and executable, but the information was retrieved
> using the stat() call (not lstat()), so any symbolic links will have
> been dereferenced (unless they're broken, of course).  Tests seem to
> confirm this is working.

Cool! And thanks a lot for all your effort!

Best
    eike