@fbarotov Hey ✋ Sorry for the late feedback. I personally found this approach better than #209 because of 2 reasons:

1. Getting region information with API call is almost costless. It parses region information from the response after sending HEAD request (ref). I wrote simple program to benchmark GetBucketRegion call:

AWS_PROFILE=personal hyperfine --show-output --min-runs 20 './testgetregion --bucket asiaonurtest >/dev/null'
Benchmark #1: ./testgetregion --bucket asiaonurtest >/dev/null
 Time (mean ± σ):     722.0 ms ±  38.0 ms    [User: 57.9 ms, System: 23.6 ms]
 Range (min … max):   690.4 ms … 833.6 ms    20 runs

I think it is quite negligible since we make this call once and use the cached session afterward.

2. IMHO the code will be more readable on this PR. In command/cp: added cross-region transfer support #209, each command needs to parse and handle retryable errors separately (e.g. 1, 2). While adding a new command to s5cmd, we also need to handle retryable errors in order to get region info. If we use this approach and get region info first, it will prevent nested error handling patterns.

What do you think? Dou you want to work on this PR and resolve conflicts before we conclude which implementation is better? 😸

@fbarotov Hey ✋ Sorry for the late feedback. I personally found this approach better than #209 because of 2 reasons:
...

Hi @sonmezonur 🖐

Thanks a lot for the feedback.

Awesome, I also ran your program on my laptop and got the following results:

1. when the default session and bucket have the same regions: 411.65 +- 303.64 ms
2. when the default session and bucket have different regions: 597.10 +- 528.81 ms

indeed, as you mentioned they are negligible.

Yes, the approach in this pr is more readable in my opinion as we would be having more loose coupling between session logic and operations logic.

Let me try to resolve the conflicts :)