hyper tests are broken since ca15b7512e

[bagder]

I did this

Build curl with hyper

cd tests
./runtests.pl 1

Problem

I think the new crlf="yes" support from commit ca15b75 should only be applied to the section where it is used, and not on the entire test case like it is now. Test 2500 only uses it for <data> so only that content should get crlf replaced.

[bagder]

The conversion could then be done explicitly without caring if it looks like HTTP or not.

[bagder]

The conversion could then be done explicitly without caring if it looks like HTTP or not.

Eh, no. Because then we also add CRLF to the body in <data> etc and that changes the size of it...

[icing]

The conversion could then be done explicitly without caring if it looks like HTTP or not.

Eh, no. Because then we also add CRLF to the body in <data> etc and that changes the size of it...

I remember myself walking down that line of thinking. That is why I sticked to the HTTP pattern that was in place for hyper. The trouble we have now is that it also replaces \r\n with \r\r\n, it seems.

[bagder]

I'm trying out a few things in #10010. I want to be able to mark individual sections as crlf="yes" so that we long-term can move away from using CRLF lines within the source test files .

[icing]

Another idea would be to have some sort of <reply><http-header>... section where crlf is always normalized.

[bagder]

Or just a <crlf>...</crlf> within <data> etc

[bagder]

A challenge is to introduce such a new/better/different system in a way that can be gradually transitioned into, without us having to update 800+ test cases in a single blow.

[icing]

I was thinking of gradually changing test cases to use the new section. The old way of setting the server response could be left unchanged. But that is just a thought, you know best.

[bagder]

If I can get this working with my current simpler approach, I think I'll stick to this for now.