> On Jan 8, 2019, at 4:06 PM, Pushkar Pradhan <pprad...@oath.com.INVALID> wrote:
>
> Even RHEL7 sucks, they are also shipping with an old curl version that
> doesn't support http1.1.
> The test was not being skipped, it tries to run the test.
> The quickest thing to do would be add a condition checking for http1.1?
Also, as was mentioned, I’d be fine installing/requiring e.g. httpd24-curl on
the CI boxes etc. for a more modern version of Curl. It’s easy to active this.
As long as we document such package dependencies, it’s all good.
— leif
>
> On Tue, Jan 8, 2019 at 2:32 PM Jason Kenny <jke...@oath.com.invalid> wrote:
>
>> I do plan to add a switch to allow one to error on condition testing. This
>> should be easy to do. However I do report which test are skipped and why (
>> and if that reason is not good enough one can help fix the test as most if
>> not all these should be reason that can be provided with the condition
>> call.)
>>
>> As far as Curl goes. building a curl could be done. It would help with
>> having a consistent curl for testing. However I don't think that this is
>> needed to get http 1.1 support. The issue I think is if you are using rhel6
>> and the supper old curl that come with it by default. you need to install
>> the httpd24-curl package and then enable it. Once this is done the support
>> you need should work as needed. There are a few test that still will not
>> run ( they should be skipped) with this version as it is built with openssl
>> 1.0.1 and this lack certain TLS features that we have tests for. You may
>> want to test if the option to use http 1.1 needs to be added to curl when
>> you run it. I have found that Curl on Fedora and Ubuntu tend to work
>> different from curl on RHEL systems.
>>
>> If you are using RHEL system you may have to configure it correctly.
>>
>> Jason
>>
>> On Mon, Jan 7, 2019 at 9:39 PM Leif Hedstrom <zw...@apache.org> wrote:
>>
>>>
>>>
>>>> On Jan 7, 2019, at 5:28 PM, Pushkar Pradhan <pprad...@oath.com.INVALID
>>>
>>> wrote:
>>>>
>>>> Yes I saw that condition later. I think it could be done for http1.1
>> but
>>> is
>>>> it a good idea to skip tests?
>>>
>>> I’ve had this discussion before, and my personal preference would be that
>>> we never do these types of checks, but rather, have the check fail, with
>> a
>>> good (obvious!) error message explaining what actions needs to be taken.
>>>
>>> The danger with the current approach, which we’ve run into numerous
>> times,
>>> is that it essentially silently skips tests.
>>>
>>> If I have a vote, i’d say get rid of all “feature” tests like this, and
>>> let them fail (again, with good errors). Then we can fix our CI and our
>> dev
>>> tools accordingly.
>>>
>>> My $.01,
>>>
>>> — Leif
>>>
>>>>
>>>> On Mon, Jan 7, 2019 at 4:14 PM Walt Karas <wka...@oath.com.invalid>
>>> wrote:
>>>>
>>>>> Many have this to skip if no http2 support but I don't know if it can
>>>>> be use to test support for http1.1 :
>>>>>
>>>>> # need Curl
>>>>> Test.SkipUnless(
>>>>> Condition.HasProgram("curl", "Curl need to be installed on system
>>>>> for this test to work"),
>>>>> Condition.HasCurlFeature('http2')
>>>>> )
>>>>>
>>>>> On Mon, Jan 7, 2019 at 4:58 PM Pushkar Pradhan
>>>>> <pprad...@oath.com.invalid> wrote:
>>>>>>
>>>>>> I tried running the autests on the master branch and saw many many
>>>>> failures.
>>>>>> Most of them failed because my RHEL box doesn't have a recent curl
>> that
>>>>>> supports http1.1 option.
>>>>>> The fix was obvious but I got no warning about this, I had to figure
>>> this
>>>>>> out.
>>>>>>
>>>>>> Would it be reasonable to include and build the latest curl sources?
>>>>>> Just like how we include yaml-cpp.
>>>>>> --
>>>>>> pushkar
>>>>>
>>>>
>>>>
>>>> --
>>>> pushkar
>>>
>>>
>>
>
>
> --
> pushkar
