Updated the HG prompt

[b0bh00d]

Per our email discussion, I have updated the HG prompt code to use the '-ib' option to 'hg id' so the branch name is always available, regardless of the state of the working copy.

[Stanzilla]

Thanks!

[Tomalak]

Just to throw this in here (cc @Stanzilla)

hg id -ib takes significantly longer than hg branch and hg status combined.

D:\some\repo default
λ hg branch --time
default
time: real 0.016 secs (user 0.016+0.000 sys 0.000+0.000)

D:\some\repo default
λ hg status --time
M test.txt
time: real 0.037 secs (user 0.016+0.000 sys 0.016+0.000)

That's a total of 0.053 seconds, as opposed to

D:\some\repo default
λ hg id -ib --time
6242b58ed6c7+ default
time: real 0.251 secs (user 0.156+0.000 sys 0.094+0.000)

hg id -ib works, i.e. it gets the needed information, but it's awfully annoying to wait a 250ms-ish for every prompt. Maybe this isn't such a good way forward after all.

[Stanzilla]

=/ @b0bh00d

[b0bh00d]

On 1/18/2019 4:50 AM, Martin Böhm wrote: Just to throw this in here (cc @Stanzilla <https://github.com/Stanzilla>) |hg id -ib|takes /significantly/longer than |hg branch|and |hg status|combined. |D:\some\repo default λ hg branch --time default time: real 0.016 secs (user 0.016+0.000 sys 0.000+0.000) D:\some\repo default λ hg status --time M test.txt time: real 0.037 secs (user 0.016+0.000 sys 0.016+0.000) | That's a total of 0.053 seconds, as opposed to |D:\some\repo default λ hg id -ib --time 6242b58ed6c7+ default time: real 0.251 secs (user 0.156+0.000 sys 0.094+0.000)|

|I ran the above commands using Mercurial version 4.7+11 on |||a "pristine" working copy with a 7.1GB disc footprint and more than 27,000 managed files. |Here's the result with some randomly modified files (and this is on an M.2 drive): | |λ hg branch --time| |default.bob.lwk| |time: real 0.005 secs (user 0.000+0.000 sys 0.000+0.000)| || |λ hg status -q --time| |M Bootstrap.bat| |M LightWave\layout\layo.cpp| |M LightWave\layout\main.cpp| |M Purge.bat| || |time: real 0.415 secs (user 0.344+0.000 sys 0.062+0.000)| || |λ hg id -ib --time| |d61a2d50bfd4+ default.bob.lwk| |time: real 0.553 secs (user 0.438+0.000 sys 0.125+0.000)| || |I'd guess my repo is considerably larger than yours, which accounts for the much larger overhead of time on "hg status." Without any modified or unmanaged files in the repo, the "quiet" version of "hg status" takes much less time on my system: | |λ hg status -q --time| || |time: real 0.242 secs (user 0.125+0.000 sys 0.109+0.000)| || |However, if I remove the "quite" option, the time surprisingly skyrockets on a "pristine" working copy: | |λ hg status --time| || |time: real 0.586 secs (user 0.531+0.000 sys 0.047+0.000)| || |"hg id" does indeed take more time, but I only have a difference of 0.133 between them using your commands. Additionally, the Lua code is only running one command instead of two, which would I suppose double the execution overhead, if we're counting milliseconds.  It probably wouldn't add much, but it would certainly reduce the difference by some amount. |||||

|hg id -ib| works, i.e. it gets the needed information, but it's awfully annoying to wait a 250ms-ish for every prompt. Maybe this isn't such a good way forward after all.

|I'd love to have the speed of git, but Mercurial simply doesn't have it.  If 100-200 milliseconds really means that much, then it can certainly be switched to use the branch/status command pair instead.  Just tell me if that's what you want. |

[Tomalak]

Okay, let's do some science. On my machine there currently are a total of 68 Mercurial repos of varying size and complexity, some small, some large, with subrepos and without, some clean, some with a couple of dirty files, some with a number of "unknown" files, too. (None of them has the order of 27,000 files, though I also believe that's not a typical use case.)

Since hg id -ib does not show the state of "unknown" files, I'm using hg status -amrd (added, modified, removed, deleted) in order to produce comparable output and keep the file listing short, because this takes time, too.

So I ran hg branch --time, hg status -amrd --time and hg id --time against all of these repositories and graphed the reported "real" time values in Excel as a stacked bar chart:

Green on top is hg id -ib, red below that is hg branch, blue at the bottom is hg status -amrd.

Its obvious that hg id -ib produces some huge outliers, whereas the performance of hg branch is pretty much constant through the board, and hg status is pretty low as well, but with some variance in the repos where there are modified files to report.

Overall, the timings of hg branch and hg status combined generally lie beneath the 50 millisecond threshold, which I would consider perfectly tolerable. My example above was taken from one of the repos that are an outlier for hg id -ib, that wasn't intentional though.

hg id -ib generally lies below or at the 50 ms threshold as well, but its performance is not as consistent. Also, committing any changes will bring hg status back to ~16ms, so the user has some control over it. On the other hand, hg id -ib seems to take its sweet time for complex repos no matter what state they are in.

(FWIW, for this test I've used Mercurial version 4.3.1, 64bit, on an SSD.)