Difference between
modified post c1545de by Anonymous on Sat 17th Jan 2015 23:19
original post ee9c3 by Anonymous on Wed 14th Jan 2015 00:31
Download diff
Show old version | new version | both versions

20150113.21:13:21        wim | hi guys, after using tmux for a while it seems like it's using quite some memory (using 1.9a).

1	1		20150113.21:13:21 wim \| hi guys, after using tmux for a while it seems like it's using quite some memory (using 1.9a).
2	2		20150113.21:13:44 wim \| 128Mb at the moment for 8 windows
3	3		20150113.21:14:31 wim \| memory leak?
4	4
5	5		20150114.00:47:03 nicm \| tmux returns memory to libc and is reliant on it to return it to the kernel
6	6		20150114.00:47:14 nicm \| glibc does not do so very aggressively
7	7		20150114.00:47:19 nicm \| this is a common issue on linux
8	8		20150114.00:47:48 nicm \| there is a library call you can put into tmux if you want it to be returned more quickly
9	9		20150114.00:48:32 nicm \| malloc_trim(3)
10	10		20150114.00:50:08 nicm \| add malloc_trim(0) somewhere, server_second_callback in server.c would be a good place, or grid_destroy in grid.c
11	11		20150114.00:50:24 nicm \| or just live with it and assume glibc knows best
12	12		20150114.00:51:23 nicm \| i believe OS X is similar, if not worse, although it has different ways to reclaim the memory, i think it is kernel-side there
13	13
14	14		20150114.01:05:47 wim \| nicm: nice! malloc_trim(0) works
15	15
16	16		20150114.01:11:55 freeroute \| wim: are you running Linux? How did you get it working? Just add to source code?
17	17		20150114.01:12:12 wim \| freeroute: yes and yes
18	18		20150114.01:12:35 wim \| in grid_destroy()
19	19		20150114.01:12:35 wim \| free(gd);
20	20		20150114.01:12:35 wim \| malloc_trim(0);
21	21		20150114.01:12:45 wim \| add it under the last free
22	22		20150114.01:14:41 wim \| would be nice if it got into the code though :/
23	23
24	24
25	25		================================================================================================================================
26	26
27	27		So basically, in tmux/grid.c - you want it to look like this:
28	28
29	29		/* Destroy grid. */
30	30		void
31	31		grid_destroy(struct grid *gd)
32	32		{
33	33		struct grid_line *gl;
34	34		u_int yy;
35	35
36	36		for (yy = 0; yy < gd->hsize + gd->sy; yy++) {
37	37		gl = &gd->linedata[yy];
38	38		free(gl->celldata);
39	39		}
40	40
41	41		free(gd->linedata);
42	42
43	43		free(gd);
44	44		malloc_trim(0);
45	45		}
46	46
47		-
	47	+	================================================================================================================================
48	48
49		-
	49	+	:(
50	50
51		-
	51	+	20150117.22:21:01 wim \| nicm: is this something that could be merged? makes it portable? the malloc_trim thing -
52		-
	52	+	\| https://github.com/42wim/tmux/commit/429fc7a7498ca39c5210b26d4e541e8d6969db3a
53		-
	53	+	20150117.22:24:42 freeroute \| wim: does the code say like "If this is going to get compiled by glibc, then also include this code" and by 'this code' I mean that malloc_trim thing.
54		-
	54	+	20150117.22:26:10 freeroute \| if so then that means I can read C :D
55		-
	55	+	20150117.22:26:59 wim \| that's the gist of it yes
56		-
	56	+	20150117.22:45:40 nicm \| not really
57		-
	57	+	20150117.22:46:45 nicm \| it's up to glibc how malloc works really
58		-
	58	+	20150117.22:52:26 wim \| nicm: no power to the people instead of letting the system decide ? :-)
59		-
	59	+	20150117.22:53:46 wim \| nicm: i tried fiddling with the environment variables, but didn't get it to work as well as malloc_trim()
60		-
	60	+	20150117.22:55:48 nicm \| you need to talk to the glibc folks if their environment vars don't work
61		-
	61	+	20150117.22:57:26 nicm \| i don't think we should do this in the code
62		-
	62	+	20150117.22:58:47 wim \| ok thanks, was worth a try
63	63