Add libcurl build phase

[hmelder]

There are still some patches required to the libs-base autoconf script to get the right version from pkg-config.

[hmelder]

gnustep/libs-base#269

[triplef]

Hi @hmelder, thanks for setting this up. After battling GitHub rate limit errors this is finally building on CI and seems to work ok.

I ran the tests locally and unfortunately there’s errors from these tests that we would probably still need to fix before merging this:

Running base/NSURLSession/test01.m...
Start set:       test01.m:79 ... NSURLSession test01
*   Trying 127.0.0.1:12345...
*   Trying [::1]:12345...
* Closing connection 0
Failed test:     (2022-12-22 14:42:41.072 +0100)   test01.m:125 ... request completed
Failed test:     (2022-12-22 14:42:41.072 +0100)   test01.m:127 ... unable to connect to host
2022-12-22 14:42:41.072 test01[15060:10120] Error (null)
End set:         test01.m:130 ... NSURLSession test01
Completed file:  test01.m

[hmelder]

Hi @hmelder, thanks for setting this up. After battling GitHub rate limit errors this is finally building on CI and seems to work ok.

I ran the tests locally and unfortunately there’s errors from these tests that we would probably still need to fix before merging this:

Running base/NSURLSession/test01.m...
Start set:       test01.m:79 ... NSURLSession test01
*   Trying 127.0.0.1:12345...
*   Trying [::1]:12345...
* Closing connection 0
Failed test:     (2022-12-22 14:42:41.072 +0100)   test01.m:125 ... request completed
Failed test:     (2022-12-22 14:42:41.072 +0100)   test01.m:127 ... unable to connect to host
2022-12-22 14:42:41.072 test01[15060:10120] Error (null)
End set:         test01.m:130 ... NSURLSession test01
Completed file:  test01.m

Nice! I’ve seen the new commits. I’ll take a look at the errors in the unit tests, but it is probably related to the local web server.

[triplef]

That’d be fantastic, thanks. This test doesn’t have a local webserver, but instead tries to check if a connection to an nonexistent host fails correctly. I’m guessing it’s something with the libcurl integration.

[hmelder]

That’d be fantastic, thanks. This test doesn’t have a local webserver, but instead tries to check if a connection to an nonexistent host fails correctly. I’m guessing it’s something with the libcurl integration.

That's correct. Internally GNUstep uses a curl multi handle that manages multiple easy handles. The handles itself are wrapped into objects, and there are a couple of quite complex handlers (Translating C-based callback functions into the proper NSURLSessionProtocol methods).

It would be great to have a CI log of the test. I guess I can just create a new branch in libs-base and modify the CI config?

[triplef]

I guess I can just create a new branch in libs-base and modify the CI config?

Yes, that’d be great! Thank you.

[triplef]

I updated libs-base to fix unreliable CI runs, and in the process added an option to easily specify a branch from this repo to use. Here’s a run I did with this libcurl branch:
https://github.com/gnustep/libs-base/actions/runs/3834875221/jobs/6527653656

[hmelder]

I am just setting up a new x86 VM on a root server, but compilation of libdispatch fails:

I guess this is a problem in libdispatch upstream, as we are pulling the master branch.

[triplef]

Which Visual Studio version are you running? I'm guessing it's most likely related to that, as there haven't been any upstream changes in libdispatch in a while.

You'll ideally want to use VS
2019, as Clang in 2022 has a number of issues with ObjC.

[hmelder]

You'll ideally want to use VS 2019, as Clang in 2022 has a number of issues with ObjC.

I am using VS 2022 with Windows Kit 11 and Clang 15.
I'll try it with VS 2019 and Clang 16 then.

[hmelder]

MSVC supports atomic<> from version 2012 onwards (https://learn.microsoft.com/en-us/cpp/standard-library/atomic-enums?view=msvc-140). Weird.

[hmelder]

I am using VS 2022 with Windows Kit 11 and Clang 15. I'll try it with VS 2019 and Clang 16 then.

Works with Clang 14

[hmelder]

There is a minor difference between curl error handling on Windows and UNIX which results in different results when parsing them internally in urlErrorCodeWithEasyCode (https://github.com/gnustep/libs-base/blob/efccdb1d71166cfe096f309a97747175f46c1435/Source/GSEasyHandle.m#L271). I get -1001 (NSURLErrorTimedOut) on Windows and -1004 (NSURLErrorCannotConnectToHost) on Linux.

urlErrorCodeWithEasyCode uses the curl error and a system-specific error code to determine the corresponding Foundation Error Code CURLINFO_OS_ERRNO.

    else if (easyCode == CURLE_COULDNT_CONNECT && failureErrno == ETIMEDOUT) 
      {
        return NSURLErrorTimedOut;
      }

I'll look into this the next days. As I need to setup the dev vm and generate debug information.

[triplef]

For some reason I actually don’t get any callback from curl when running the test locally, so the test’s run loop just times out...

I’ll try to look into it further since I’m working on various NSURLSession improvements.

[triplef]

Unfortunately I didn’t manage to track down the NSURLSession/test01 failure, so for now I marked it as hopeful as part of my NSURLSession improvements PR gnustep/libs-base#286.

With that the libs-base CI passes:
https://github.com/gnustep/libs-base/actions/runs/4225184793