Rewrite the assert statement to avoid Socks Proxy triggering the assertions

[oknet]

No description provided.

[oknet]

The assert statement is introduced by PR #4166.

Request @shinrich for review.

[oknet]

Since PR #4166 has been ported back to 7.1.x and 8.0.x, this PR is a candidate for them to fix the SocksProxy.

@zwoop @bryancall

[zwoop]

@ezelkow1 Lets put this on the radar for v7.1.7 as well.

[ezelkow1]

👍 We've seen this crash in 8.1.x at least, so if the issue is on 7 it should definitely go in

[jrushford]

@shinrich and @oknet we've see crashes here on 8.1. this looks like it will fix the crashes we've seen

[ezelkow1]

We pulled this in to a test 8.1.x build yesterday and were still seeing this assert hit

[shinrich]

Seems reasonable. I had not considered the socks path.

[oknet]

@ezelkow1 Have your filed an issue ? Please show me your assert and result of gdb backtrace full command.

[ezelkow1]

@oknet I have not filed an issue since I was just pulling this change in before it was merged. I cant really get a full backtrace, only what bits we have stored from BT since the only way to get this is to run it in production for a while and wait to hit it.

From BT I can see that pending_action is not null, so it must be failing on the second part of the statement, so the pending's cont is different from the vc's. I dont have any insight into the contents of 'vc' during the crashes either so I cant personally do a comparison of the continuation values

[zwoop]

This has merge conflicts in 8.1.x