Loading /etc/bash_completion is slow

[Singpolyma]

Loading /etc/bash_completion takes over 0.5s (usually over 0.6s). While this doesn't seem like much, maybe, it's very annoying for a file that must be sourced in .bashrc. If it could go in .bash_profile, that would be fine, but it seems bash is not designed to work that way for completion rules.

I am on a 1.1Ghz Pentium M, so not ancient h/w. The 0.6s delay on opening a new terminal is very annoying.

[Kim Alvefur (zash)]

A way to cache what bash_completion does would be awesome.

[Kim Alvefur (zash)]

Or, improving and including the on-demand version of bash-completion (forgot what it's called)

[Kim Alvefur (zash)]

This is what I was thinking of: http://code.google.com/p/bash-completion-lib/
It's not as good as bash-completion, but it's much faster, since it doesn't load all rules at once.

[AreTee (ryantimmins)]

I also get >0.5s load time on bashrc due to the loading of bash_completion.
I am using Intel(R) Pentium(R) Dual CPU T2330 @ 1.60GHz, 2GB Ram.

[Radosław Bułat (radek-bulat+launchpad)]

What is the progress of this issue? On my computer (core 2 duo, 2.20 GHz) it takes about 1.5-2s to load /etc/bash_completion.

$ time . /etc/bash_completion

real 0m1.656s
user 0m1.290s
sys 0m0.330s

With non warmed up computer (adter rebooting) it could take 2-3x times longer (arrgh).

[Philip Muškovac (yofel)]

This would most likely be improved once someone takes care of http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=467231

[Jason Gunthorpe (jgunthorpe)]

There is more going on here than just loading a lot of completion. Something is weird inside bash, look at the strace of completion loading and you see it is doing a wack of busy work crap like this. The fiddling with FDs looks like it is probably some kind of errant debug feature? This is from maverick with bash 4.1-2ubuntu4. It is annoying that

16:58:30 access("/etc/bash_completion.d/mutt", R_OK) = 0
16:58:30 open("/etc/bash_completion.d/mutt", O_RDONLY) = 3
16:58:30 fstat(3, {st_mode=S_IFREG|0644, st_size=4527, ...}) = 0
16:58:30 read(3, "# mutt completion\n#\n# Mutt doesn"..., 4527) = 4527
16:58:30 close(3) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
16:58:30 open("/dev/null", O_WRONLY|O_CREAT|O_TRUNC, 0666) = 3
16:58:30 fcntl(1, F_GETFD) = 0
16:58:30 fcntl(1, F_DUPFD, 10) = 10
16:58:30 fcntl(1, F_GETFD) = 0
16:58:30 fcntl(10, F_SETFD, FD_CLOEXEC) = 0
16:58:30 dup2(3, 1) = 1
16:58:30 close(3) = 0
16:58:30 fcntl(2, F_DUPFD, 10) = 11
16:58:30 fcntl(2, F_GETFD) = 0
16:58:30 fcntl(11, F_SETFD, FD_CLOEXEC) = 0
16:58:30 dup2(1, 2) = 2
16:58:30 stat(".", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
16:58:30 stat("/usr/sbin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or dir
16:58:30 stat("/usr/bin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or dire
16:58:30 stat("/sbin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or directo
16:58:30 stat("/bin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or director
16:58:30 stat("/usr/bin/X11/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or
16:58:30 stat("/usr/bin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or dire
16:58:30 stat("/usr/local/bin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file o
16:58:30 stat("/usr/games/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or di
16:58:30 stat("/home/jgg/bin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or
16:58:30 stat("/sbin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or directo
16:58:30 stat("/usr/sbin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file or dir
16:58:30 stat("/usr/local/sbin/mutt", 0x7fff6d9e1230) = -1 ENOENT (No such file
16:58:30 write(2, "bash: type: mutt: not found\n", 28) = 28
16:58:30 dup2(11, 2) = 2
16:58:30 fcntl(11, F_GETFD) = 0x1 (flags FD_CLOEXEC)
16:58:30 close(11) = 0
16:58:30 dup2(10, 1) = 1
16:58:30 fcntl(10, F_GETFD) = 0x1 (flags FD_CLOEXEC)
16:58:30 close(10) = 0
16:58:30 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0

[Dawid Ciężarkiewicz (dpc-ucore)]

If anybody is interested -- I think I've hacked some nice solution for this. Seems to be working for me (I havn't tested it extensivly).

http://ucore.info/en/2011/03/faster-bash-completion-loading/
https://github.com/dpc/bash_completion

dpc@mutex:~$ time . /etc/bash_completion.orig

real 0m2.360s
user 0m1.510s
sys 0m0.380s
dpc@mutex:~$ time . /etc/bash_completion.new

real 0m0.149s
user 0m0.100s
sys 0m0.010s

[Dawid Ciężarkiewicz (dpc-ucore)]

I am sorry. I've performed the test for original code with additional links. The real results should be:

dpc@mutex:~$ time . /etc/bash_completion.orig
real 0m0.614s
user 0m0.360s
sys 0m0.120s
dpc@mutex:~$ time . /etc/bash_completion.new
real 0m0.168s
user 0m0.120s
sys 0m0.010s

[Neal H. Walfield (neal-walfield)]

Loading /etc/bash_completion takes 3 seconds on my Alix-based system (which has an AMD Geode 500 MHz CPU).

Bash completions are loaded in a number of places including via /etc/profile. A non-invasive way to disable this is to add

  BASH_COMPLETION=1

to the start of /etc/profile.

/etc/profile.d/bash_completion.sh will only load /etc/bash_completion if BASH_COMPLETION is the empty string.

If you get an error:

  sed: can't read 1: No such file or directory

Then something else is still trying to load /etc/bash_completion. In my base, this was ~/.bashrc.

[Gigabit Software (hlavaty-ondrej-gmail)]

Sorry, wrong button... Loading takes long time also in 11.10... And fix from #8 is not working anymore - didn't find out the reason.