/dev/chat

1:13 AM

Is Packer still used to maintain immutable infrastructure?

Both on-premise & on lcoud

9:27 AM

@Gilles'SO-stopbeingevil' Weird. I just checked again and it's consistent here: find took an average of 13.88 seconds and ** took 6. How many files do you have? Maybe that's the difference?

$ find . -print0 | grep -c '\0'
find: ‘./.protonvpn-cli’: Permission denied
425576

So, one dir that needs root access, and 425,576 files.

Ah! But the globstar returns far fewer, that's why it's faster!

$ printf '%s\0' ** | grep -c '\0'
48069

I have an autofs-mounted dir in there, but that goes to a pi which is connected to an external drive bay, which I can hear spin up when accessed, so I don't think either approach enters it. Must be something else.

But at least that makes sense. If ** goes through an order of magnitude fewer files, of course it's faster.

Kusalananda

10:15 AM

The scariest thing I've read this week: Right now I'm dealing with it by piping ls into vim and doing a bulk-rename via global commands and writing to !sh.

Stephen Kitt

@Kusalananda a bit like dired in Emacs

Kusalananda

@StephenKitt I'm honestly relieved to not know what that is :-)

And, good morning!

Stephen Kitt

@Kusalananda imagine basically ls’s output in an Emacs buffer, and you can edit it, and have the changes applied to your file system!

@Kusalananda good morning to you too!

Kusalananda

@terdon shopt -s dotglob?

terdon

@Kusalananda EEEeeek!

@Kusalananda Duh! That's almost certainly it, yes. Testing now.

Kusalananda

10:21 AM

@StephenKitt Does it keep track of filenames correctly?

Stephen Kitt

@Kusalananda in my experience, yes

gnu.org/software/emacs/manual/html_node/emacs/…

but I prefer “orthodox” file managers, so I use mc

Kusalananda

I used a file manager on my Amiga 500 back in the day, but not since.

I believe I've accidentally started mc a couple of times, but I don't have it installed on my private systems.

terdon

10:58 AM

Yeah, I don't remember the last time I used a file manager.

By the way, @Kusalananda, as expected you were quite right about dotglob. I can't tell you how long it takes but that's because it's still running, almost an hour later.

Kusalananda

@terdon :-) That's amazing!

I'm still going to start pushing for ** in my answers though, where it makes sense, in addition to showing find alternatives.

terdon

Well, it's running it 10 times, but yeah. Wow.

Kusalananda

Is that from the root down?

terdon

no, $HOME

Ah, just finished.

$ timethis 10 "find . -name findme"
COMMAND: find . -name findme
13.88
$ timethis 10 "printf ./**/findme"
COMMAND: printf ./**/findme
248.6

An average of 248.6 seconds per run. Ouch.

Kusalananda

You're running over a structure with half a million entries, right?

Finding a file in my $HOME (84000+ entries) with (OpenBSD or GNU) find takes 0.7-0.8s. Using ** in bash takes 2.718s, while zsh takes 1.777s to do the same.

Those are approximate medians over about 5 runs.

The ** glob builds a list. find doesn't need to do that. This might have something to do with it being slower. The shell may be allocating and reallocating space for the list of pathnames.

A list of half a million pathnames of an average length of, say, 60 bytes is just under 30 MB.

@StéphaneChazelas Thanks for that edit just now! I'm learning so much I don't know where to put everything :-)

Tim

12:54 PM

@StephenKitt Good morning to you two!

Stephen Kitt

@Tim good morning!

Good morning Jeff too!

Transcript for