Fix non blocking eof for TLS stream

[bukka]

Updated version of #3223

[bukka]

The test seems to be failing on Windows but it was working before in #3223 (see https://ci.appveyor.com/project/php/php-src/build/master.build.6841 ).

@weltling what OpenSSL version is now used in AppVeyor (I have tested so far on Linux with 1.0.2 and 1.1.0)? And would you be able to test it locally if you see the same issue? If so it might be interesting to check if rasing proxy usleep changes anything: https://github.com/php/php-src/pull/3752/files#diff-3fbb599e938f2cc62f0620bed8411ee3R79

[bukka]

Ok the test was a bit flaky. I fixed that and all seems good.

Merged as 0c84c2e , d9b2902 and dc2ffde

[nikic]

It looks like the test is still flaky. I've seen a couple of Travis failures that look like this:

========DIFF========
001- string(0) ""
========DONE========
FAIL Bug #76705: feof might hang on TLS streams in case of fragmented TLS records [ext/openssl/tests/bug77390.phpt] 

[valga]

@bukka I think you should increase usleep() value (10x should be enough) and roll back your changes.

The test was made to show that TLS packet that has been split into 3 fragments must cause exactly 3 fread() calls to be received in full. And now it shows something else.

You can also decrease the number of fragments from 3 to 2 to speed up the test.

[valga]

@bukka As an alternative, you could try using phpt_wait() instead of usleep() in the proxy, so it will know when the client is done with previous fragment.

[bukka]

@valga My changes just displayed an empty string once instead of 3 times as it behaves differently with OpenSSL 1.1.1 where you get split to more fragments for some reason.

Looks that travis doesn't even do any split sometimes so I guess we should probably increase usleep or use phpt_wait as you say. Feel free to create a PR for that!

[bukka]

Sorry was hoping to look to this today but was too busy with other stuff and won't probably have much time over the week so if someone wants to take a look and fix it, that would be great!

[billybraga]

Am I right to understand this bug still exists in php 7.1?

[kelunik]

Seems like that, but PHP 7.1 has been end of life for quite some time already now, see https://www.php.net/supported-versions.php.

[bukka]

7.1 doesn't support OpenSSL 3.0 so should be not impacted anyway...

[billybraga]

@bukka I looked around, and I don't see anything about openssl version...

[bukka]

Sorry I confused it with another issue. This is really old though and 7.1 is no longer supported as it was said.