Why is Fcitx5 taking 3G of RAM?

[codayon]

Chat, is this real?

I'm using OpenBangla Keyboard with it and using Arch Linux. I have not noticed sometime similar to this before.

If I remember correctly, I did not write anything in Bangla.

[wengxt]

1. is it immediately become 3g after you start?
2. can you post fcitx5-diagnose ouput?

[codayon]

1. is it immediately become 3g after you start?

Nope, ideally it uses around 100M. I saw this, when my system uptime was seven and half hours.

2. can you post fcitx5-diagnose ouput?

Not sure how to do that, can you help a bit?

[wengxt]

if you are no using anything like Chinese that loads language model, it should be fairly small on memory (20mb ish).
I don't know what might be leaking, can this issue can be reliably reproduced?

fcitx5-diagnose is a command. you just run it in terminal and copy paste all the output.

[codayon]

it should be fairly small on memory (20mb ish)

Woah!

I don't know what might be leaking, can this issue can be reliably reproduced?

Not sure about that, probably no.

Here is the fcitx5-diagnose output

[wengxt]

I did some preliminary analysis. I think it is indeed related to openbangla.

So, whenever there's an new app connected to fcitx, fcitx will pre-allocate some struct for openbangla.

I tried to create new client and can obviously observe that the usage is increasing with new client. However, even if client goes away, the memory is not going down.

The code there is using RAII so doesn't seem to be an issue in fcitx engine glue code.

I tried a small demo program with only https://github.com/OpenBangla/riti/ and try valgrind and shows no real leak.

This leads me to think it might be due to mem fragmentation.

So one solution seems to be using one context instead of multiple context. after all, openbangla's context data does not share and the json parsing is also an issue to contribute to memory usage.

[wengxt]

OpenBangla/OpenBangla-Keyboard#424