Snapshots for CURRENT

[Julien ÉLIE]

> En réponse à Russ Allbery :
>>> By the way, I have for innd/artparse.t:
>>
>>> ==9259== Invalid write of size 1
>>> ==9259==    at 0x805257F: ARTsetup (art.c:287)
>>> ==9259==    by 0x8051628: initialize (artparse-t.c:98)
>>> ==9259==    by 0x8051A19: main (artparse-t.c:175)
>>> ==9259==  Address 0x42DF927 is 1 bytes before a block of size 1 alloc'd
>>> ==9259==    at 0x401D38B: malloc (vg_replace_malloc.c:149)
>>> ==9259==    by 0x80A296A: x_strdup (xmalloc.c:128)
>>> ==9259==    by 0x8052569: ARTsetup (art.c:286)
>>> ==9259==    by 0x8051628: initialize (artparse-t.c:98)
>>> ==9259==    by 0x8051A19: main (artparse-t.c:175)
>>
>> I haven't looked at that in detail yet, but that definitely looks like a
>> bug.


I tried to do that:

Index: artparse-t.c
===================================================================
--- artparse-t.c        (révision 7651)
+++ artparse-t.c        (copie de travail)
@@ -94,7 +94,8 @@
     if (Log == NULL)
         sysdie("Cannot open /dev/null");
     fdreserve(4);
-    buffer_set(&Path, "", 1);
+    buffer_set(&Path, "news.example.com!", 17);
+    Path.used += 17;
     ARTsetup();
 }


because it would prevent Path.used (= 0) to break ARTsetup in art.c:

  /* Get our Path name, kill trailing !. */
  ARTpathme = xstrdup(Path.data);
  ARTpathme[Path.used - 1] = '\0';


Perhaps we ought to make sure that Path.used is not 0
with an ASSERT here in art.c?



Now, what strikes me most is that there is a new error in valgrind
when I run the test:

==29891== Conditional jump or move depends on uninitialised value(s)
==29891==    at 0x401E215: strlen (mc_replace_strmem.c:246)
==29891==    by 0x8094506: x_strdup (xmalloc.c:127)
==29891==    by 0x8056EB9: ARTsetup (art.c:286)
==29891==    by 0x8051741: main (artparse-t.c:99)

char *
x_strdup(const char *s, const char *file, int line)
{
    char *p;
    size_t len;

    len = strlen(s) + 1;      <----------------------------------------



And it looks the same as what there is in chan.t (the test which currently
makes snapshots fail, with "strcmp (mc_replace_strmem.c:341)").

-- 
Julien ÉLIE

« Vina bibant homines, animantia cetera fontes. »