Re: ZSH performance regression in 5.8.1.2-test

[Jordan Patterson <jordanp@xxxxxxxxx>]

I'm not subscribed to the mailing list so I'm missing Jun's original
email, but I'll reply here.

> Jordan, what do you get by the following?
> zsh -xic exit 2>>(wc)

I get 13007 lines, 1534161 characters.

> Can you try the lseek() patch (in my previous post, 50115⁩)?

Sure.  I needed to add a mode argument to the open in your lseek
configure check.  I get this error otherwise on my system:

error: call to '__open_missing_mode' declared with attribute error:
open with o_creat or o_tmpfile in second argument needs 3 arguments

My benchmark results:

Benchmark 1: prefix/5.8/bin/zsh -i -c exit
  Time (mean ± σ):     254.6 ms ±  10.0 ms    [User: 174.6 ms, System: 68.0 ms]
  Range (min … max):   240.1 ms … 271.7 ms    10 runs

Benchmark 2: prefix/5.8.1/bin/zsh -i -c exit
  Time (mean ± σ):     251.7 ms ±   8.8 ms    [User: 179.2 ms, System: 60.0 ms]
  Range (min … max):   233.3 ms … 262.5 ms    11 runs

Benchmark 3: prefix/5.8.1.2-test/bin/zsh -i -c exit
  Time (mean ± σ):      2.583 s ±  0.053 s    [User: 0.516 s, System: 2.033 s]
  Range (min … max):    2.540 s …  2.687 s    10 runs

Benchmark 4: prefix/5.8.1.2-test-lseek/bin/zsh -i -c exit
  Time (mean ± σ):     283.8 ms ±   8.0 ms    [User: 202.7 ms, System: 70.4 ms]
  Range (min … max):   273.0 ms … 300.3 ms    10 runs

Summary
  'prefix/5.8.1/bin/zsh -i -c exit' ran
    1.01 ± 0.05 times faster than 'prefix/5.8/bin/zsh -i -c exit'
    1.13 ± 0.05 times faster than 'prefix/5.8.1.2-test-lseek/bin/zsh -i -c exit'
   10.27 ± 0.41 times faster than 'prefix/5.8.1.2-test/bin/zsh -i -c exit'

On Tue, Apr 26, 2022 at 1:01 AM Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Apr 25, 2022 at 2:27 PM Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
> >
> > Theoretically we can block-read with impunity in cases 1 and 2 (anyone
> > disagree?).  Testing for seek-ability would allow doing the "read too
> > much and back up" trick in case 3.  I don't immediately see any way to
> > avoid reading one byte at a time in case 4, does anyone have a
> > suggestion?
>
> Try this?