# Re the blog post: PSA: Thread-local state is no longer recommended
**URL:** https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303
**Category:** Internals & Design
**Created:** [July 7, 2023, 9:25am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303 "2023-07-07T09:25:11Z")
**Posts on this page:** 8
**Page:** 1
### Author:  [@wherrera](https://discourse.julialang.org/u/wherrera)
#### Post date: [July 7, 2023, 9:25am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/1 "2023-07-07T09:25:11Z")
Disfavored in the blog post, due to issues with threadid, was:
```
using Base.Threads: nthreads, @threads, threadid
states = [some_initial_value for _ in 1:nthreads()]
@threads for x in some_data
tid = threadid()
old_val = states[tid]
new_val = some_operator(old_val, f(x))
states[tid] = new_val
end
do_something(states)
```
Proposed solution: **enumerate** the loop with the index as a program-specific task id:
```
using Base.Threads: @threads
states = [some_initial_value for _ in eachindex(x)] # <-- changed
@threads for (id, x) in enumerate(some_data) # <-- the key change is here
old_val = states[id]
new_val = some_operator(old_val, f(x))
states[id] = new_val
end
do_something(states)
```
---
### Author:  [@carstenbauer](https://discourse.julialang.org/u/carstenbauer)
#### Post date: [July 7, 2023, 9:33am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/2 "2023-07-07T09:33:18Z")
Isn’t that pretty much the same pattern as [the one proposed in the other thread](https://discourse.julialang.org/t/psa-thread-local-state-is-no-longer-recommended-common-misconceptions-about-threadid-and-nthreads/101274/8) (the latter just being more explicit/verbose)?
> [@wherrera](#):
>
> `states = [some_initial_value for _ in eachindex(x)] # <-- changed`
`x` isn’t defined here?
---
### Author:  [@wherrera](https://discourse.julialang.org/u/wherrera)
#### Post date: [July 7, 2023, 9:50am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/3 "2023-07-07T09:50:40Z")
Sorry, that could have been some\_data, and I was not posting working code (the example I was quoting did not define some\_data)… I will add below:
```
const some_data = [3, 4, 5]
states = [some_initial_value for _ in eachindex(some_data)] # <-- changed
@Threads.threads for (id, x) in enumerate(some_data) # <-- the key change is here
old_val = states[id]
new_val = some_operator(old_val, f(x))
states[id] = new_val
end
do_something(states)
```
---
### Author:  [@jakobnissen](https://discourse.julialang.org/u/jakobnissen)
#### Post date: [July 7, 2023, 10:10am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/4 "2023-07-07T10:10:33Z")
I use this pattern myself. As written, it doesn’t actually work though.
```julia
julia> result = zeros(10)
Threads.@threads for (i, x) in enumerate(19:28)
result[i] = x ÷ 2
end
ERROR: TaskFailedException
nested task error: MethodError: no method matching firstindex(::Base.Iterators.Enumerate{UnitRange{Int64}})
```
You need to do `collect(enumerate(some_data))`. I wish the implementation didn’t require this, but a single collect over a multithreaded loop is usually insignificant.
---
### Author:  [@Sukera](https://discourse.julialang.org/u/Sukera)
#### Post date: [July 7, 2023, 10:12am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/5 "2023-07-07T10:12:45Z")
Unfortunately, the `collect` version is not always correct, e.g. for channels.
---
### Author:  [@lmiq](https://discourse.julialang.org/u/lmiq)
#### Post date: [July 7, 2023, 11:47am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/6 "2023-07-07T11:47:22Z")
> [@wherrera](#):
>
> `@threads for (id, x) in enumerate(some_data)`
That is ok, but then you need the number of states to be the same as the number of elements in the `some_data` array.
More generally one wants a buffer of length `nthreads`, which is independent of the number of tasks. That is why in general the proposal is to partition the workload in chunks and then use the chunk index (instead of `threadid()`).
---
### Author:  [@tecosaur](https://discourse.julialang.org/u/tecosaur)
#### Post date: [July 8, 2023, 5:31am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/7 "2023-07-08T05:31:06Z")
Sometimes the `collect` for is also very expensive, in terms of allocation/compute. Is there much of a chance of this requirement being removed in the future?
---
### Author:  [@lmiq](https://discourse.julialang.org/u/lmiq)
#### Post date: [July 8, 2023, 8:55am UTC](https://discourse.julialang.org/t/re-the-blog-post-psa-thread-local-state-is-no-longer-recommended/101303/8 "2023-07-08T08:55:27Z")
I think you can use `pairs` there, and you don’t need the collection. But you could also use `eachindex` just indexing the elements inside the loop.