[yosefe]

No description provided.

[mellanox-github]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
http://bgate.mellanox.com:8000/jenkins-secure/job/gh-ucx-pr/82/
Test PASSed.

[mike-dubman]

line 52 is done not under lock
maybe it is better to split this function into actual init_thread() and get_thread_id() parts?
if get_thread_id function the var "threads_count" can be accessed w/o lock indeed.

[yosefe]

this is a double-checked lock paradigm so no need to take lock every time.
if current thread is not found, it is searched for again with lock held.
it assigns the threads according to the order they are calling this function
calling it on the first time from log_init forces thread 0 to be the main.

[mike-dubman]

ok

[mike-dubman]

👍

[shamisp]

There is a gettid() call, why you don't use it

[yosefe]

i wanted that the thread numbers will be 0..N-1, and not system-wide identifier

[shamisp]

1. Does it mean that that corresponds to threads id's in gdb or something ? How does it help identify actual thread

2. It seems that the function finds thread id and if it can't set threads' id. Maybe it is better to separate to two separate calls ? (not critical)

I think it is useful info to have but it should correspond to some "real" value that is helpful for debug (gdb).

[yosefe]

Unfortunately it does not correspond to gdb.
I think we should fix it to make it same as gdb (probably sort by tid?)

[shamisp]

See below http://stackoverflow.com/questions/2188089/pthread-t-to-gdb-thread-id

[shamisp]

I may ask our debugger guy how the gdb magic works in older gdb versions.