Don't issue redundant stdin detection warning when is in place.

[isidentical]

This patch fixes a race happening between 2 IO bound operations:

• read(2) with sys.stdin
• select(2) with sys.stdin

select(2) is a function that takes files and tells when they are ready to be processed (read/written etc.). It is used for concurrently dealing with multiple sockets (e.g you have multiple open connections, and with select(2) you can check which one of those connections have some data to be processed). In HTTPie however, we used it to detect whether there is any incoming data to the sys.stdin, so that when there is no data we could issue a warning.

The problem is that, the initial version of this code unknowlingly created a race between the select.select call:
https://github.com/httpie/httpie/blob/30cd862fc0e173698fc17487c4b96d8f64b701ea/httpie/uploads.py#L98-L110

and the file.read() call:
https://github.com/httpie/httpie/blob/30cd862fc0e173698fc17487c4b96d8f64b701ea/httpie/uploads.py#L124-L132

Both of these functions are blocking, and we ran them concurrently to get away from that issue. But since Python has a GIL, when you run multiple threads together, depending on the CPU-related activity and some other details it switches between them on certain intervals. (green arrows indicate the interpreter given the execution to the Main Thread, red indicate it is given to Observer Thread)

So to put it simply, we used to run the read() and the select() at the same time and whichever gets choosen to be executed at that time would win. If it was select(), we'd get the correct result (about whether there is any data or not) and if it was read() we'd always get the wrong result (no incomiong data). This caused to intermittent / random failure. From what I understood this problem happens randomly (and in a OS-dependant way, because of the underlying thread scheduler), and there is no fair way of testing it (except manually).

The race has been solved by moving the select(2) out of the thread (as a guard) and using threading.Events as indicators about whether we've seen any data or not.

[codecov-commenter]

Codecov Report

Merging #1303 (75ae325) into master (4d7d6b6) will decrease coverage by 0.75%.
The diff coverage is 95.12%.

@@            Coverage Diff             @@
##           master    #1303      +/-   ##
==========================================
- Coverage   97.28%   96.53%   -0.76%     
==========================================
  Files          67       85      +18     
  Lines        4235     6027    +1792     
==========================================
+ Hits         4120     5818    +1698     
- Misses        115      209      +94     

Impacted Files	Coverage Δ
tests/test_binary.py	100.00% <ø> (ø)
httpie/compat.py	31.11% <27.90%> (-68.89%)	⬇️
tests/conftest.py	77.14% <58.33%> (-9.82%)	⬇️
httpie/output/lexers/http.py	59.09% <59.09%> (ø)
tests/test_ssl.py	91.01% <66.66%> (-3.93%)	⬇️
httpie/manager/__main__.py	82.35% <82.35%> (ø)
httpie/output/lexers/metadata.py	82.35% <82.35%> (ø)
httpie/models.py	94.23% <90.00%> (-3.14%)	⬇️
tests/test_uploads.py	97.70% <92.59%> (-2.30%)	⬇️
httpie/manager/core.py	92.85% <92.85%> (ø)
... and 71 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ad613f2...75ae325. Read the comment docs.

[jkbrzt]

Good stuff. Let’s just add a mention to CHANGELOG