bug#21666: 25.0.50; Random segfaults

[Paul Eggert <eggert@cs.ucla.edu>]

[-- Attachment #1: Type: text/plain, Size: 694 bytes --]

Those two backtraces both have a call to mapcar1 with a long list (leni=2544 in 
one, leni=2502 in the other).  When Fmapcar uses SAFE_ALLOCA_LISP, it'll ask for 
(say) 2502*8 bytes, or 20016 bytes, and this is more than the 16 KiB MAX_ALLOCA 
limit, which means SAFE_ALLOCA_LISP will invoke xmalloc and make_save_memory 
rather than invoking AVAIL_ALLOCA.  One possibility is that the xmalloced memory 
isn't being properly analyzed by the garbage collector.  To test that theory, 
can you please try something like the attached patch?  This wouldn't fix the 
bug, but it would mean mapcar should work for larger lists and you won't get a 
backtrace until the list gets about twice as large.


[-- Attachment #2: t.diff --]
[-- Type: text/plain, Size: 345 bytes --]

diff --git a/src/fns.c b/src/fns.c
index dfd48a2..bb0af4a 100644
--- a/src/fns.c
+++ b/src/fns.c
@@ -2584,6 +2584,7 @@ SEQUENCE may be a list, a vector, a bool-vector, or a string.  */)
   register Lisp_Object *args;
   Lisp_Object ret;
   USE_SAFE_ALLOCA;
+  sa_avail = 32 * 1024;
 
   len = Flength (sequence);
   if (CHAR_TABLE_P (sequence))