tab completion is really slow

[dwijnand]

in a clone of https://github.com/scala/scala run tig blame src/library/scala/math/Order<TAB> and (on my machine) it'll take 14 seconds to give you:

$ tig blame src/library/scala/math/Order
src/library/scala/math/Ordered.scala    src/library/scala/math/Ordering.scala

tig version 2.3.3
git version 2.15.1

[pzrq]

@dwijnand Great question. Could you run $ time tig blame src/library/scala/math/Order<TAB> and copy/paste the output here?

If you can think of any other way to debug, profile or otherwise get visibility into what is causing the slowness that would be great too.

[jonas]

The bash tab completion is still based on a very old copy of the git completion. Could be part of the problem.

…

[dwijnand]

Could you run $ time tig blame src/library/scala/math/Order<TAB> and copy/paste the output here?

you can't time the bash completion - it just outputs the same thing:

$ time tig blame src/library/scala/math/Order
src/library/scala/math/Ordered.scala    src/library/scala/math/Ordering.scala

[pzrq]

Then is this a tig issue or a bash issue or some interaction of the two?

[dwijnand]

it's an issue with tig's bash completion.

for comparison completing blame src/library/scala/math/Order<TAB> is less than a second for me.

[askedrelic]

I've noticed this as well on a git repo with 14K tags. I'm assuming tags are part of the autocomplete algorithm and that is what I assume really slows it down. I think it's more due to bash being slow at large amounts of data.

tig version 2.3.3
git version 2.16.3

[orehmane]

Just ran into this myself, with the following versions:

git version 2.20.1
tig version 2.4.1
ncurses version 5.7.20081102
readline version 7.0

The repo in question has 42468 tags, which I assume is the cause (since it's only happening in that one directory). It hangs seemingly indefinitely.

[ingydotnet]

On a huge repo I use daily, tig Bash completion hangs for 2 minutes. If I type tig and press tab, I'm screwed. Git completions on same repo are fast.

Here's a fun trick:

set -x; tig <tab><tab>

Try that on a big repo and sit back with some coffee and watch the show.

[rohieb]

The culprit is probably __tig_complete_file:

linux.git $ time __tig_complete_file 

real    0m29.363s
user    0m27.805s
sys     0m1.600s

linux.git $ time __git_complete_file

real    0m0.014s
user    0m0.003s
sys     0m0.008s

In the long run I would probably rewrite the tig completion so it uses the git completion for all git-related options. Most users should already have this installed, and it is faster in most cases, as well as better maintained :)

[rohieb]

linux.git $ time __tig_complete_file 

real    0m29.363s
user    0m27.805s
sys     0m1.600s

linux.git $ time __git_complete_file

real    0m0.014s
user    0m0.003s
sys     0m0.008s

These times are reproducible btw, so no caching issues.

[benpaxton-hf]

the loop in _tigcomp is the problem; I replaced it locally with an equivalent but significantly faster awk script before finding this issue and the PR - my tests showed that for the specific repo I was having issues with, it was taking 15+ seconds, iterating approximately 9000 times and building a 327kB string; the awk script took ~0.01s and built an identical string.

#960 is probably the "right" solution here, though.

[dwijnand]

Wonderful to have this fixed! Thank you, everyone involved.