Outputting text to the terminal increases run time by up to 30% about ouch HOT 7 CLOSED

How much increased time would be tolerable? I have no idea how much time is normal for a print to take, but it is expected for them to tax the system a little bit.

from ouch.

How much increased time would be tolerable?

I'm not sure, we would need to investigate it further, we might even conclude at the end that our current time is tolerable or can be improved.

Let's see, I ran the same command and redirected the output to the file output_file.

ouch decompress --yes linux-5.14.9.arch2-1-x86_64.pkg.tar.zst &> output_file                                                                                                                                                                      

Now, we can show the contents in the terminal.

time cat output_file

It showed 0.145 seconds in the first run, and 0.045s for some of the consecutive runs.

That means that the terminal can display the same text in amount 0.045 seconds, if we take the minimum (this can be misleading if I'm missing something).

Here is the time it takes to decompressing the linux package with and without displaying the output:

time ouch decompress --yes linux-5.14.9.arch2-1-x86_64.pkg.tar.zst
# 0.95s on average

time ouch decompress --yes linux-5.14.9.arch2-1-x86_64.pkg.tar.zst &> output_file
# 0.82s on average

So it's an addition of 130 milliseconds.

from ouch.

It's common to speed up this by using an additional buffer, here's an example from coreutils:
https://github.com/GrayJack/coreutils/blob/dev/uniq/src/main.rs#L25-L32

This saves on OS calls to write into the stdout file descriptor, but in our case, the problem is probably with how terminals implement their flushing (it depends on the terminal, one of them can try flushing by every frame if too much flush is demanded).

from ouch.

tested on 2c5a57c, got a slightly different result

[...]
  Time (mean ± σ):      6.137 s ±  0.745 s    [User: 3.145 s, System: 2.873 s]
  Range (min … max):    5.589 s …  7.551 s    10 runs

Benchmark #2: target/release/ouch d linux.tar.gz -yo '/tmp/tmp.nhJZkRNvMD' > '/tmp/tmp.nhJZkRNvMD/log'
  Time (mean ± σ):      5.366 s ±  0.087 s    [User: 2.813 s, System: 2.406 s]
  Range (min … max):    5.242 s …  5.490 s    10 runs

Summary
  'target/release/ouch d linux.tar.gz -yo '/tmp/tmp.nhJZkRNvMD' > '/tmp/tmp.nhJZkRNvMD/log'' ran
    1.14 ± 0.14 times faster than 'target/release/ouch d linux.tar.gz -yo '/tmp/tmp.nhJZkRNvMD''

This is the script I tested with, requires hyperfine.

#!/bin/sh

set -e

cargo build --release
[ ! -f linux.tar.gz ] && curl https://github.com/torvalds/linux/archive/v5.14.tar.gz -Lo linux.tar.gz
out=$(mktemp -d)
target/release/ouch d linux.tar.gz -o "$out"
hyperfine --show-output \
    "target/release/ouch d linux.tar.gz -yo '$out'" \
    "target/release/ouch d linux.tar.gz -yo '$out' > '$out/log'"
rm -r "$out"

from ouch.

Interesting, it depends on the archive structure and level of compression, a 14%-ish speed improvement is not a huge priority right now, but I'll leave this open in case we want to optimize in the future.

from ouch.

should be fixed once #643 is solved.

from ouch.

Fixed! #643 was solved by #642 🎉

from ouch.