From: Paul Eggert <eggert@cs.ucla.edu>
To: Eli Zaretskii <eliz@gnu.org>
Cc: jrm@ftfl.ca, mattiase@acm.org, 37006@debbugs.gnu.org
Subject: bug#37006: 27.0.50; garbage collection not happening after 26de2d42
Date: Sat, 14 Sep 2019 00:51:18 -0700 [thread overview]
Message-ID: <b66de82b-d22d-e8fa-22dd-0022db5a5b78@cs.ucla.edu> (raw)
In-Reply-To: <83o90qqzqw.fsf@gnu.org>
[-- Attachment #1: Type: text/plain, Size: 334 bytes --]
On 8/15/19 7:17 AM, Eli Zaretskii wrote:
> OK, but I still hope we will switch to EMACS_INT instead.
I installed the attached patch into master, and it changes the relevant
threshold variables to EMACS_INT, as part of a somewhat-larger cleanup. This
should make a practical difference only on 32-bit Emacs without --with-wide-int.
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Improve-gc-cons-percentage-calculation.patch --]
[-- Type: text/x-patch; name="0001-Improve-gc-cons-percentage-calculation.patch", Size: 12388 bytes --]
From bac66302e92bdd3a353102d2076548e7e83d92e5 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Sat, 14 Sep 2019 00:32:01 -0700
Subject: [PATCH] Improve gc-cons-percentage calculation
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
The old calculation relied on a hodgpodge of partly updated GC
stats to find a number to multiply gc-cons-percentage by.
The new one counts data found by the previous GC, plus half of
the data allocated since then; this is more systematic albeit
still ad hoc.
* src/alloc.c (consing_until_gc, gc_threshold, consing_threshold):
Now EMACS_INT, not intmax_t.
(HI_THRESHOLD): New macro.
(tally_consing): New function.
(make_interval, allocate_string, allocate_string_data)
(make_float, free_cons, allocate_vectorlike, Fmake_symbol): Use it.
(allow_garbage_collection, inhibit_garbage_collection)
(consing_threshold, garbage_collect):
Use HI_THRESHOLD rather than INTMAX_MAX.
(consing_threshold): New arg SINCE_GC. All callers changed.
(bump_consing_until_gc): Return new consing_until_gc, instead of
nil. All callers changed. Don’t worry about overflow since we
now saturate at HI_THRESHOLD. Guess that half of
recently-allocated objects are still alive, instead of relying on
the previous (even less-accurate) hodgepodge.
(maybe_garbage_collect): New function.
(garbage_collect): Work even if a finalizer disables or enables
memory profiling. Do not use malloc_probe if GC reclaimed nothing.
* src/lisp.h (maybe_gc): Call maybe_garbage_collect instead
of garbage_collect.
---
src/alloc.c | 127 ++++++++++++++++++++++++++++++----------------------
src/lisp.h | 5 ++-
2 files changed, 77 insertions(+), 55 deletions(-)
diff --git a/src/alloc.c b/src/alloc.c
index ca8311cc00..497f600551 100644
--- a/src/alloc.c
+++ b/src/alloc.c
@@ -224,7 +224,7 @@ struct emacs_globals globals;
/* maybe_gc collects garbage if this goes negative. */
-intmax_t consing_until_gc;
+EMACS_INT consing_until_gc;
#ifdef HAVE_PDUMPER
/* Number of finalizers run: used to loop over GC until we stop
@@ -238,9 +238,16 @@ bool gc_in_progress;
/* System byte and object counts reported by GC. */
+/* Assume byte counts fit in uintptr_t and object counts fit into
+ intptr_t. */
typedef uintptr_t byte_ct;
typedef intptr_t object_ct;
+/* Large-magnitude value for a threshold count, which fits in EMACS_INT.
+ Using only half the EMACS_INT range avoids overflow hassles.
+ There is no need to fit these counts into fixnums. */
+#define HI_THRESHOLD (EMACS_INT_MAX / 2)
+
/* Number of live and free conses etc. counted by the most-recent GC. */
static struct gcstat
@@ -299,7 +306,7 @@ static intptr_t garbage_collection_inhibited;
/* The GC threshold in bytes, the last time it was calculated
from gc-cons-threshold and gc-cons-percentage. */
-static intmax_t gc_threshold;
+static EMACS_INT gc_threshold;
/* If nonzero, this is a warning delivered by malloc and not yet
displayed. */
@@ -536,6 +543,15 @@ XFLOAT_INIT (Lisp_Object f, double n)
XFLOAT (f)->u.data = n;
}
+/* Account for allocation of NBYTES in the heap. This is a separate
+ function to avoid hassles with implementation-defined conversion
+ from unsigned to signed types. */
+static void
+tally_consing (ptrdiff_t nbytes)
+{
+ consing_until_gc -= nbytes;
+}
+
#ifdef DOUG_LEA_MALLOC
static bool
pointers_fit_in_lispobj_p (void)
@@ -1372,7 +1388,7 @@ make_interval (void)
MALLOC_UNBLOCK_INPUT;
- consing_until_gc -= sizeof (struct interval);
+ tally_consing (sizeof (struct interval));
intervals_consed++;
RESET_INTERVAL (val);
val->gcmarkbit = 0;
@@ -1739,7 +1755,7 @@ allocate_string (void)
MALLOC_UNBLOCK_INPUT;
++strings_consed;
- consing_until_gc -= sizeof *s;
+ tally_consing (sizeof *s);
#ifdef GC_CHECK_STRING_BYTES
if (!noninteractive)
@@ -1859,7 +1875,7 @@ allocate_string_data (struct Lisp_String *s,
old_data->string = NULL;
}
- consing_until_gc -= needed;
+ tally_consing (needed);
}
@@ -2464,7 +2480,7 @@ make_float (double float_value)
XFLOAT_INIT (val, float_value);
eassert (!XFLOAT_MARKED_P (XFLOAT (val)));
- consing_until_gc -= sizeof (struct Lisp_Float);
+ tally_consing (sizeof (struct Lisp_Float));
floats_consed++;
return val;
}
@@ -2535,8 +2551,8 @@ free_cons (struct Lisp_Cons *ptr)
ptr->u.s.u.chain = cons_free_list;
ptr->u.s.car = dead_object ();
cons_free_list = ptr;
- if (INT_ADD_WRAPV (consing_until_gc, sizeof *ptr, &consing_until_gc))
- consing_until_gc = INTMAX_MAX;
+ ptrdiff_t nbytes = sizeof *ptr;
+ tally_consing (-nbytes);
}
DEFUN ("cons", Fcons, Scons, 2, 2, 0,
@@ -3153,7 +3169,7 @@ allocate_vectorlike (ptrdiff_t len)
if (find_suspicious_object_in_range (p, (char *) p + nbytes))
emacs_abort ();
- consing_until_gc -= nbytes;
+ tally_consing (nbytes);
vector_cells_consed += len;
MALLOC_UNBLOCK_INPUT;
@@ -3438,7 +3454,7 @@ Its value is void, and its function definition and property list are nil. */)
MALLOC_UNBLOCK_INPUT;
init_symbol (val, name);
- consing_until_gc -= sizeof (struct Lisp_Symbol);
+ tally_consing (sizeof (struct Lisp_Symbol));
symbols_consed++;
return val;
}
@@ -5477,7 +5493,7 @@ staticpro (Lisp_Object const *varaddress)
static void
allow_garbage_collection (intmax_t consing)
{
- consing_until_gc = consing - (INTMAX_MAX - consing_until_gc);
+ consing_until_gc = consing - (HI_THRESHOLD - consing_until_gc);
garbage_collection_inhibited--;
}
@@ -5487,7 +5503,7 @@ inhibit_garbage_collection (void)
ptrdiff_t count = SPECPDL_INDEX ();
record_unwind_protect_intmax (allow_garbage_collection, consing_until_gc);
garbage_collection_inhibited++;
- consing_until_gc = INTMAX_MAX;
+ consing_until_gc = HI_THRESHOLD;
return count;
}
@@ -5761,11 +5777,13 @@ mark_and_sweep_weak_table_contents (void)
}
}
-/* Return the number of bytes to cons between GCs, assuming
- gc-cons-threshold is THRESHOLD and gc-cons-percentage is
- PERCENTAGE. */
-static intmax_t
-consing_threshold (intmax_t threshold, Lisp_Object percentage)
+/* Return the number of bytes to cons between GCs, given THRESHOLD and
+ PERCENTAGE. When calculating a threshold based on PERCENTAGE,
+ assume SINCE_GC bytes have been allocated since the most recent GC.
+ The returned value is positive and no greater than HI_THRESHOLD. */
+static EMACS_INT
+consing_threshold (intmax_t threshold, Lisp_Object percentage,
+ intmax_t since_gc)
{
if (!NILP (Vmemory_full))
return memory_full_cons_threshold;
@@ -5775,42 +5793,33 @@ consing_threshold (intmax_t threshold, Lisp_Object percentage)
if (FLOATP (percentage))
{
double tot = (XFLOAT_DATA (percentage)
- * total_bytes_of_live_objects ());
+ * (total_bytes_of_live_objects () + since_gc));
if (threshold < tot)
{
- if (tot < INTMAX_MAX)
- threshold = tot;
+ if (tot < HI_THRESHOLD)
+ return tot;
else
- threshold = INTMAX_MAX;
+ return HI_THRESHOLD;
}
}
- return threshold;
+ return min (threshold, HI_THRESHOLD);
}
}
-/* Adjust consing_until_gc, assuming gc-cons-threshold is THRESHOLD and
- gc-cons-percentage is PERCENTAGE. */
-static Lisp_Object
+/* Adjust consing_until_gc and gc_threshold, given THRESHOLD and PERCENTAGE.
+ Return the updated consing_until_gc. */
+
+static EMACS_INT
bump_consing_until_gc (intmax_t threshold, Lisp_Object percentage)
{
- /* If consing_until_gc is negative leave it alone, since this prevents
- negative integer overflow and a GC would have been done soon anyway. */
- if (0 <= consing_until_gc)
- {
- threshold = consing_threshold (threshold, percentage);
- intmax_t sum;
- if (INT_ADD_WRAPV (consing_until_gc, threshold - gc_threshold, &sum))
- {
- /* Scale the threshold down so that consing_until_gc does
- not overflow. */
- sum = INTMAX_MAX;
- threshold = INTMAX_MAX - consing_until_gc + gc_threshold;
- }
- consing_until_gc = sum;
- gc_threshold = threshold;
- }
-
- return Qnil;
+ /* Guesstimate that half the bytes allocated since the most
+ recent GC are still in use. */
+ EMACS_INT since_gc = (gc_threshold - consing_until_gc) >> 1;
+ EMACS_INT new_gc_threshold = consing_threshold (threshold, percentage,
+ since_gc);
+ consing_until_gc += new_gc_threshold - gc_threshold;
+ gc_threshold = new_gc_threshold;
+ return consing_until_gc;
}
/* Watch changes to gc-cons-threshold. */
@@ -5821,7 +5830,8 @@ watch_gc_cons_threshold (Lisp_Object symbol, Lisp_Object newval,
intmax_t threshold;
if (! (INTEGERP (newval) && integer_to_intmax (newval, &threshold)))
return Qnil;
- return bump_consing_until_gc (threshold, Vgc_cons_percentage);
+ bump_consing_until_gc (threshold, Vgc_cons_percentage);
+ return Qnil;
}
/* Watch changes to gc-cons-percentage. */
@@ -5829,7 +5839,18 @@ static Lisp_Object
watch_gc_cons_percentage (Lisp_Object symbol, Lisp_Object newval,
Lisp_Object operation, Lisp_Object where)
{
- return bump_consing_until_gc (gc_cons_threshold, newval);
+ bump_consing_until_gc (gc_cons_threshold, newval);
+ return Qnil;
+}
+
+/* It may be time to collect garbage. Recalculate consing_until_gc,
+ since it might depend on current usage, and do the garbage
+ collection if the recalculation says so. */
+void
+maybe_garbage_collect (void)
+{
+ if (bump_consing_until_gc (gc_cons_threshold, Vgc_cons_percentage) < 0)
+ garbage_collect ();
}
/* Subroutine of Fgarbage_collect that does most of the work. */
@@ -5841,7 +5862,6 @@ garbage_collect (void)
bool message_p;
ptrdiff_t count = SPECPDL_INDEX ();
struct timespec start;
- byte_ct tot_before = 0;
eassert (weak_hash_tables == NULL);
@@ -5856,14 +5876,15 @@ garbage_collect (void)
FOR_EACH_BUFFER (nextb)
compact_buffer (nextb);
- if (profiler_memory_running)
- tot_before = total_bytes_of_live_objects ();
+ byte_ct tot_before = (profiler_memory_running
+ ? total_bytes_of_live_objects ()
+ : (byte_ct) -1);
start = current_timespec ();
/* In case user calls debug_print during GC,
don't let that cause a recursive GC. */
- consing_until_gc = INTMAX_MAX;
+ consing_until_gc = HI_THRESHOLD;
/* Save what's currently displayed in the echo area. Don't do that
if we are GC'ing because we've run out of memory, since
@@ -5975,7 +5996,7 @@ garbage_collect (void)
unblock_input ();
consing_until_gc = gc_threshold
- = consing_threshold (gc_cons_threshold, Vgc_cons_percentage);
+ = consing_threshold (gc_cons_threshold, Vgc_cons_percentage, 0);
if (garbage_collection_messages && NILP (Vmemory_full))
{
@@ -6008,11 +6029,11 @@ garbage_collect (void)
gcs_done++;
/* Collect profiling data. */
- if (profiler_memory_running)
+ if (tot_before != (byte_ct) -1)
{
byte_ct tot_after = total_bytes_of_live_objects ();
- byte_ct swept = tot_before <= tot_after ? 0 : tot_before - tot_after;
- malloc_probe (min (swept, SIZE_MAX));
+ if (tot_after < tot_before)
+ malloc_probe (min (tot_before - tot_after, SIZE_MAX));
}
}
diff --git a/src/lisp.h b/src/lisp.h
index 024e5edb26..02f8a7b668 100644
--- a/src/lisp.h
+++ b/src/lisp.h
@@ -3824,9 +3824,10 @@ extern void mark_maybe_objects (Lisp_Object const *, ptrdiff_t);
extern void mark_stack (char const *, char const *);
extern void flush_stack_call_func (void (*func) (void *arg), void *arg);
extern void garbage_collect (void);
+extern void maybe_garbage_collect (void);
extern const char *pending_malloc_warning;
extern Lisp_Object zero_vector;
-extern intmax_t consing_until_gc;
+extern EMACS_INT consing_until_gc;
#ifdef HAVE_PDUMPER
extern int number_finalizers_run;
#endif
@@ -5056,7 +5057,7 @@ INLINE void
maybe_gc (void)
{
if (consing_until_gc < 0)
- garbage_collect ();
+ maybe_garbage_collect ();
}
INLINE_HEADER_END
--
2.17.1
next prev parent reply other threads:[~2019-09-14 7:51 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <5075406D-6DB8-4560-BB64-7198526FCF9F@acm.org>
2019-08-11 16:23 ` bug#37006: 27.0.50; garbage collection not happening after 26de2d42 Mattias Engdegård
2019-08-11 17:07 ` Eli Zaretskii
[not found] ` <83h86nu0pq.fsf@gnu.org>
[not found] ` <86pnlbphus.fsf@phe.ftfl.ca>
2019-08-12 2:31 ` Eli Zaretskii
2019-08-12 14:34 ` Joseph Mingrone
2019-08-12 16:49 ` Eli Zaretskii
2019-08-12 17:00 ` Mattias Engdegård
2019-08-13 15:37 ` Eli Zaretskii
2019-08-13 16:48 ` Mattias Engdegård
2019-08-13 17:04 ` Eli Zaretskii
2019-08-13 17:29 ` Mattias Engdegård
2019-08-13 17:21 ` Paul Eggert
2019-08-13 17:53 ` Eli Zaretskii
2019-08-13 19:32 ` Paul Eggert
2019-08-14 16:06 ` Eli Zaretskii
2019-08-15 1:37 ` Paul Eggert
2019-08-15 14:17 ` Eli Zaretskii
2019-08-15 18:51 ` Paul Eggert
2019-08-15 19:34 ` Eli Zaretskii
2019-09-14 7:51 ` Paul Eggert [this message]
2019-09-14 8:30 ` Eli Zaretskii
2019-08-11 12:39 Joseph Mingrone
2019-08-11 15:13 ` Eli Zaretskii
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=b66de82b-d22d-e8fa-22dd-0022db5a5b78@cs.ucla.edu \
--to=eggert@cs.ucla.edu \
--cc=37006@debbugs.gnu.org \
--cc=eliz@gnu.org \
--cc=jrm@ftfl.ca \
--cc=mattiase@acm.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this external index
https://git.savannah.gnu.org/cgit/emacs.git
https://git.savannah.gnu.org/cgit/emacs/org-mode.git
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.