Ignore locale when creating the HTTP version string from a float

[Zegnat]

This should fix #335.

Rather than using %f, this should use the non-locale aware %F. That way PHP will always use a period (.) for the decimal separator in the HTTP version, and never fallback to a comma (,).

I wasn’t sure how to add a test for this, as I’d want to intercept the socket message somehow. But to illustrate the issue:

php > echo setlocale(LC_NUMERIC, 0);
C
php > echo sprintf("%s %s HTTP/%.1f\r\n", 'GET', '/', 1.1);
GET / HTTP/1.1
php > echo setlocale(LC_NUMERIC, 'de_DE');
de_DE
php > echo sprintf("%s %s HTTP/%.1f\r\n", 'GET', '/', 1.1);
GET / HTTP/1,1
php > echo sprintf("%s %s HTTP/%.1F\r\n", 'GET', '/', 1.1);
GET / HTTP/1.1

[codecov-io]

Codecov Report

Merging #339 into master will not change coverage.
The diff coverage is 100%.

@@            Coverage Diff            @@
##             master     #339   +/-   ##
=========================================
  Coverage     92.11%   92.11%           
  Complexity      760      760           
=========================================
  Files            21       21           
  Lines          1762     1762           
=========================================
  Hits           1623     1623           
  Misses          139      139

Impacted Files	Coverage Δ	Complexity Δ
library/Requests/Transport/fsockopen.php	94.18% <100%> (ø)	68 <0> (ø)	⬇️

---

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 bf35eb2...e52c155. Read the comment docs.

[Zegnat]

Not sure why #335 was closed as a support issue while the bug identified in there, and patched here, is still there.

Remerged with master as per #346 (comment) in the hope to get all tests to pass and get this PR accepted. Something still seems wrong with Travis? Seeing errors connecting to localhost, even from cURL when this patch only touches the fsocket code.

Let me know what I can do to get this resolved!

[rmccue]

Reopened #335, apologies.

The tests are likely failing due to transient issues; the nature of working with servers is that occasionally things fail for no reason. I've restarted those tests.

[rmccue]

Tests are now passing after restarting, thanks!

[Zegnat]

Many thanks for merging! 🎊

[jrfnl]

I've opened PR #521 to add a regression test for this fix.