extendedglob recursion (/)#

extendedglob recursion (/)#

[Eric Cook]

Playing around in IRC with the command:
setopt extendedglob; print (/)# # or /usr/(/)#, etc.

zsh starts to recurse / repeatedly and eventually(errors and) prints /
// ///...

Mikachu | % pl (/)#|wc -l
Mikachu |  glob.c:276: BUG: statfullpath(): pathname too long

What should the 'proper' behavior be? Recursing into directories below /
or just printing `/'
I am in favor of the latter personally.

Re: extendedglob recursion (/)#

[Eric Cook]

On 02/25/2015 07:51 PM, Eric Cook wrote:
> Playing around in IRC with the command:
> setopt extendedglob; print (/)# # or /usr/(/)#, etc.
>
> zsh starts to recurse / repeatedly and eventually(errors and) prints /
> // ///...
>
> Mikachu | % pl (/)#|wc -l
> Mikachu |  glob.c:276: BUG: statfullpath(): pathname too long
>
> What should the 'proper' behavior be? Recursing into directories below /
> or just printing `/'
> I am in favor of the latter personally.
>
>
On second thought, recursing into the directories below /wouldn't make
much sense with that pattern.
So nevermind about the question, just a bug report.

Re: extendedglob recursion (/)#

[Bart Schaefer]

(Apologies if this ends up appearing multiple times.  Outgoing mail
from my home machine has mysteriously stopped working [I blame
Verizon] so I'm pasting the message into gmail while I figure out what
the heck is going on.)

On Feb 25,  7:51pm, Eric Cook wrote:
}
} Mikachu | % pl (/)#|wc -l
} Mikachu |  glob.c:276: BUG: statfullpath(): pathname too long
}
} What should the 'proper' behavior be? Recursing into directories below /
} or just printing `/'

Recursing into directories below / would definitely be wrong (and would
in any case need to fail with "no match" because there can only be the
one directory named "/").

There's a pretty straightforward fix:

diff --git a/Src/glob.c b/Src/glob.c
index 82f8d62..f48cb8e 100644
--- a/Src/glob.c
+++ b/Src/glob.c
@@ -708,7 +708,8 @@ parsecomplist(char *instr)
         }
         l1 = (Complist) zhalloc(sizeof *l1);
         l1->pat = p1;
-        l1->closure = 1 + pdflag;
+        /* special case (/)# to avoid infinite recursion */
+        l1->closure = (p1->patmlen > 0) ? 1 + pdflag : 0;
         l1->follow = 0;
         l1->next = parsecomplist(instr);
         return (l1->pat) ? l1 : NULL;

Re: extendedglob recursion (/)#

[Bart Schaefer]

On Feb 25,  8:56pm, Bart Schaefer wrote:
}
} +        /* special case (/)# to avoid infinite recursion */
} +        l1->closure = (p1->patmlen > 0) ? 1 + pdflag : 0;

Sorry, that's not correct.  patmlen can be zero in cases other than (/).

diff --git a/Src/glob.c b/Src/glob.c
index 82f8d62..33facde 100644
--- a/Src/glob.c
+++ b/Src/glob.c
@@ -708,7 +708,8 @@ parsecomplist(char *instr)
 	    }
 	    l1 = (Complist) zhalloc(sizeof *l1);
 	    l1->pat = p1;
-	    l1->closure = 1 + pdflag;
+	    /* special case (/)# to avoid infinite recursion */
+	    l1->closure = (*((char *)p1 + p1->startoff)) ? 1 + pdflag : 0;
 	    l1->follow = 0;
 	    l1->next = parsecomplist(instr);
 	    return (l1->pat) ? l1 : NULL;