GNU bug report logs - #19180
vacuum_weak_hash_table error

Previous Next

Full log

Message #86 received at 19180 <at> debbugs.gnu.org (full text, mbox):

From: ludo <at> gnu.org (Ludovic Courtès)
To: Ricardo Wurmus <ricardo.wurmus <at> mdc-berlin.de>
Cc: Andy Wingo <wingo <at> igalia.com>, 19180 <at> debbugs.gnu.org,
 Christopher Allan Webber <cwebber <at> dustycloud.org>, guile-devel <at> gnu.org
Subject: Re: Weak tables harmful to GC?
Date: Mon, 30 Oct 2017 13:35:56 +0100

[Message part 1 (text/plain, inline)]

Hi Ricardo,

Ricardo Wurmus <ricardo.wurmus <at> mdc-berlin.de> skribis:

> In language/tree-il/analyze.scm:
>   1053:33  3 Exception thrown while printing backtrace:
> ERROR: In procedure assq: Wrong type argument in position 2 (expecting association list): ((system base pmatch) car . #f)
>
> ice-9/boot-9.scm:760:25: In procedure dispatch-exception:
> ice-9/boot-9.scm:760:25: In procedure assq: Wrong type argument in position 2 (expecting association list): ((system base pmatch) car . #f)

It’s a sign that the weak tables were too weak, this time.  :-)

The problem stems from the fact that weak pairs were initialized too
late.  Thus, the first calls to ‘scm_weak_car_pair’ were happening
before the weak-car pair GC descriptor had been initialized; they were
therefore using 0 as their descriptor, and ended up not being traced at
all by the GC.

The fix is to initialize weak pairs before symbols, as in 2.0:

[Message part 2 (text/x-patch, inline)]

modified   libguile/hashtab.c
@@ -1608,10 +1608,11 @@ scm_c_weak_table_fold (scm_t_hash_fold_fn fn, void *closure,
 
 
 
+/* Initialize weak pairs, used by weak hash tables.  This needs to be
+   done early on because it's used by interned symbols etc.  */
 void
-scm_init_hashtab ()
+scm_init_weak_pairs ()
 {
-  /* Initialize weak pairs.  */
   GC_word wcar_pair_bitmap[GC_BITMAP_SIZE (scm_t_cell)] = { 0 };
   GC_word wcdr_pair_bitmap[GC_BITMAP_SIZE (scm_t_cell)] = { 0 };
 
@@ -1627,6 +1628,11 @@ scm_init_hashtab ()
   wcdr_pair_descr = GC_make_descriptor (wcdr_pair_bitmap,
 					GC_WORD_LEN (scm_t_cell));
 
+}
+
+void
+scm_init_hashtab ()
+{
 #include "libguile/hashtab.x"
 }
 
modified   libguile/hashtab.h
@@ -174,6 +174,7 @@ SCM_API SCM scm_hash_map_to_list (SCM proc, SCM hash);
 SCM_API SCM scm_hash_count (SCM hash, SCM pred);
 SCM_INTERNAL void scm_i_hashtable_print (SCM exp, SCM port, scm_print_state *pstate);
 SCM_INTERNAL void scm_init_hashtab (void);
+SCM_INTERNAL void scm_init_weak_pairs (void);
 
 
 /* Guile 2.2.x (x <= 2) weak-table API.  */
modified   libguile/init.c
@@ -390,7 +390,8 @@ scm_i_init_guile (void *base)
 #ifdef GUILE_DEBUG_MALLOC
   scm_debug_malloc_prehistory ();
 #endif
-  scm_symbols_prehistory ();      /* requires weak_table_prehistory */
+  scm_init_weak_pairs ();
+  scm_symbols_prehistory ();      /* requires weak_pairs */
   scm_modules_prehistory ();
   scm_init_array_handle ();

[Message part 3 (text/plain, inline)]

I’m attaching updated patches.  I’ve let the Guix build run to
completion this time.  Let me know if it works for you!

Ludo’.

[0001-Remove-weak-tables-and-revert-to-weak-hash-tables.patch (text/x-patch, attachment)]

[0002-Keep-weak-hash-table-item-count-consistent.patch (text/x-patch, attachment)]

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.