Performance degradation from 2.14.1 to 2.15.2

[pgorod]

⚠️ This issue respects the following points: ⚠️

• This is a bug. Not a question or feature request.
• The topic is not already reported at Issues. (I've searched it).
• Markor is up to date. See Releases for the latest version. Updates are available from F-Droid and GitHub.
• The bug is still present in the latest development version (git master). (Please download and try the test version of Markor, named Marder. Don't worry; Markor and Marder appear as completely separate applications. You can install both side-by-side, and Markor settings are not touched. In case the issue is resolved there, you don't need to create a bug report. The change will be part of the next Markor update.)

Description

Using a big markdown file, I notice a significant performance degradation when I upgraded Markor from 2.14.1 (normal release) to 2.15.2 (using Marder, latest debug version).

I use a big file everyday, but I can't post it here, so I created a new sample one from a free eBook, which I tried to give the same characteristics as my file (size, number of lines, of headings, kind of markdown used).

My file shows an even greater performance penalty, but I don't know exactly why and the sample seems enough to reproduce the problem:

Chesterton.md

What sort of delay am I seeing?

I am using a OnePlus Nord CE 5G with 8GB of RAM which generally feels fast and snappy.

Syntax highlighting is turned on in all tests.
Turning it off doesn't seem to make a very noticeable difference.

• opening the file takes around 1.5s before upgrade, 3s or 4s after
• opening/closing the keyboard goes from almost immediate to noticeable (1 or 1.5s)
• navigating to a chapter by tapping on the top bar and selecting: no apparent performance hit

Steps to reproduce

1. Open the file
2. Open keyboard or close it
3. Contrast delays in both versions

Information

Android version: 11
Device: OnePlus Nord CE 5G / 8G RAM / SnapDragon 750G
App Version: 2.14.1 / 2.15.2

Source

Github apk

Format / File type

Markdown

[harshad1]

Three things:

1. Marder is a debug build and will be slower
2. Nothing should have changed causing this
3. Google keeps making 'security' improvements which kill filesystem performance.

[pgorod]

Thanks for your comments Harshad!

Just some thoughts 🤔 on your points

1. I thought of this... however my initial experience of slowness was by upgrading Markor to Markor (latest), not Marder. It was immediately obvious to me due to my habit of working on that large file I have daily.
   And I guess the same comparison could be tried with the different Marder versions and the slowness would still be apparent.

2. This is the big riddle here... I admit I don't have an explanation

3. This should not be a factor because I didn't upgrade Google stuff (except Google apps); and even if there were system upgrades without my awareness of them, then reverting to the older Markor wouldn't have solved the problem; but it did. Going back to older Markor makes things fast again.

I was hoping that if you guys can run this on an IDE with debugging (or even with performance profiling) then something would pop up as the cause of the delay...

Or another diagnostic approach would be to list the "diff" of the two versions:
v2.14.1...v2.15.2

... and then use the bisection method to determine which commit introduced the delays:

• try adding half the commits and test
• if it's slow, you know the slowness is one of the added commits. So go back and this time add just half of those; test again.
• if it's fast, leave those commits in and add half of the rest; test again.
• you get the point - go on using the same method until you narrow it down to a single commit which is the culprit. Then debug or profile.

I'm sorry I can't be of much help, I am a developer but not on Android, and not Java... it would be a very steep learning curve for me 😦

Thanks, I really appreciate the work you guys do on this.

[gsantner]

Does toggle of line numbers make a difference?

[pgorod]

Hi. Thanks for looking at this. 👍🏼

I never use line numbers; all my tests were without line numbers.

Anyway, I tried it now and I would say that apparently it doesn't make a difference. Maybe with line numbers there's a slight additional delay, like 0.5s (opening the file), but nothing very noticeable.

Were you able to reproduce the problem with the sample file I sent?

[harshad1]

I was able to reproduce the problem. Not sure where it comes from though

[pgorod]

Well, I am already a happy man to learn that you were able to see it and that I am not imagining things :-)

If you can narrow it down to a single commit that introduces the problem, things should become clearer... hopefully.

Thank you for your help on this.

[Odnes]

Reproduced it myself, the slowness appears on release 2.15.0 already.
Samsung SM-A125F/DSN

EDIT: Actually, didn't try the OP's Chesterton.md file, I just faced the issue with my own overly long QuickNote.txt 😉 (zim syntax).

[pgorod]

the slowness appears on release 2.15.0 already.

Ok, that's good to narrow things down. So I guess the next step is to go through this list of PR's in that release to find out the culprit.

https://github.com/gsantner/markor/releases/tag/v2.15.0

[lrq3000]

That's also what I reported a long time ago, the regression appeared betwee v2.14.1 and v2.15.0. #2652 (comment)

And it is still present in v2.16.0 despite the speed improvements PR in #2700 .

[gsantner]

Improvements are welcome.

I close this, as just saying it is STILL slow in various wordings doesn't help anybody.
Please help with devlopment of Markor if you can.

Thanks.

[lrq3000]

@gsantner Well it was worth retrying after #2700 but unfortunately this did not improve this issue.

#2652 is also a duplicate, I suggest to close it too in favor of the more technically oriented #2739 where I will post my findings.