ngtcp2 and altsvc fixes

[piru]

This PR fixes the build for non-IPv6 targets.

In addition Alt-Svc header parsing was pretty much broken:

• before the flags parsing assumed there are single flags section that applied to all alternative protocols, resulting in the code parsing the flags from wrong location on multi-entry header.
• the code would get stuck in a busyloop if protocol wasn't followed by an equal sign.
• the parser would terminate on unknown protocols rather than ignoring them.

[bagder]

ngtcp2.c includes curl_setup.h and there on line 337:

curl/lib/curl_setup.h

Line 337 in f41dedd

#include <assert.h>

Can you figure out why that isn't enough?

(Also, we should probably convert the assert() calls to DEBUGASSERT() to unify with other code.)

[piru]

Oh. that header looks like a mistake indeed.

[piru]

I git incompetence aside, now it should just do the one thing that is really needed, add the #ifdef ENABLE_IPV6 that is.

[piru]

Got a bit carried away with the alt-svc stuff I spotted. Comments welcome.

[piru]

fixed regression on unit test 1654

[bagder]

Did you detect those problems with the existing test case(s) or should we extend them to make sure?

[piru]

Did you detect those problems with the existing test case(s) or should we extend them to make sure?

Existing test cases didn't show these issues. This is a real world header response I got from LiteSpeed server that choked the parser:

quic=":443"; ma=2592000; v="39,43,46", h3-Q039=":443"; ma=2592000, h3-Q043=":443"; ma=2592000, h3-Q046=":443"; ma=2592000, h3-23=":443"; ma=2592000, h3-24=":443"; ma=2592000

The old parser completely gives up on this one at the "quic" entry. It could be extended to include h2 entry at the end:

quic=":443"; ma=2592000; v="39,43,46", h3-Q039=":443"; ma=2592000, h3-Q043=":443"; ma=2592000, h3-Q046=":443"; ma=2592000, h3-23=":443"; ma=2592000, h3-24=":443"; ma=2592000, h2="foo.example.com:443"

If unit test 1654 was adjusted to check for success of parsing this header it'd detect at least this case. To add check for failing to parse flags correctly would require checking the actual parsed items. For example:
h2="foo.example.com:443"; persist=1, h2="bar.example.com:443; persist=0

With old parser both entries will have persist of 1, whereas correct result in that first to have 1, 2nd one have 0. Similar check could be done for ma, but checking for correct value is time dependant.

[bagder]

Thanks, landing this and then we can work on polishing this further.