Closed lamyergeier closed 4 years ago
What operating system (+ version) and what Node.js version?
$ node --version v10.15.0
Ubuntu 18.4
$ trash --version 1.4.0
Ok, so there are multiple reasons why it's slower:
rm
. rm
just deletes files, trash
has to actually move the files to a different location on your hard drive. That's much slower.rm
is created in C and trash
is created in Node.js.I'm gonna keep this issue open as "help wanted" to improve performance on Linux.
May be we could send the process to the background.
You can easily do that yourself: alias trash="trash &"
.
@issuehunt has funded $80.00 to this issue.
From my tests:
meow
brings it down ~35msupdate-notifier
brings it down ~25msSo, all in all, should be easier to achieve 2 times faster time. Also, none of optimizations are related to linux directly, they affect all platforms.
Another thing to consider is that most of the time is spent in long chains of requires, because there is a lot of really small modules involved instead of bigger ones.
removing meow brings it down ~35ms
I don't want to do this, but I'm already planning to optimize meow
: https://github.com/sindresorhus/meow/issues/67 https://github.com/sindresorhus/meow/issues/104
removing update-notifier brings it down ~25ms
👍
However, this issue was meant to investigate whether there's a way to optimize the actual Linux deletion code.
@sindresorhus I don't remember the numbers(plus there is some trickery in profiling short-lived processes), but most of the time is spent initializing node and requiring for simple case (trash one-file
), to the point where if you use chrome profiler you won't even see where actual work starts unless you zoom quite a lot.
As for more complex scenarios with multiple files and directories, I don't expect there is much to do, except some place at the beginning where I've seen sync code instead of async, but with much work to do further down the line that shouldn't have much impact.
Did you try profiling on a directory with many subdirectories and thousands of files? That's a better real-world benchmark and what we should optimize for.
My debug using console.time
in 4.14.52-1-MANJARO, file with ~801Mb.
UPDATE: Full log
I actually have some results, and am preparing to make PR.
@sindresorhus Long story short, the fix includes using procfs. After writing 10th implementation of that stuff and working on optimizing it yet again for the 10th time, I looked around for libraries to access procfs
but none of them seem to do such things, only parse some selected areas. So, naturally, I decided to stop writing same code again and again and at the moment I'm close to releasing well optimized library which is able to work with a huge part of procfs
including the stuff needed for trash
. The library is also optimized for require times, so shouldn't be too heavy to include. Would you mind me using it here(after I release it and you review it if you want)?
The library is also optimized for require times, so shouldn't be too heavy to include. Would you mind me using it here(after I release it and you review it if you want)?
Great! More than happy to use it here.
@sindresorhus has rewarded $72.00 to @stroncium. See it on IssueHunt
IssueHunt Summary
#### [ stroncium](https://issuehunt.io/u/stroncium) has been rewarded. ### Backers (Total: $80.00) - [ issuehunt](https://issuehunt.io/u/issuehunt) ($80.00) ### Submitted pull Requests - [#90 Improve performance on Linux, solves #79](https://issuehunt.io/r/sindresorhus/trash/pull/90) --- ### Tips - Checkout the [Issuehunt explorer](https://issuehunt.io/r/sindresorhus/trash/) to discover more funded issues. - Need some help from other developers? [Add your repositories](https://issuehunt.io/r/new) on IssueHunt to raise funds. --- IssueHunt has been backed by the following sponsors. [Become a sponsor](https://issuehunt.io/membership/members)