80-test_cmp_http.t: Prevent hangs on server not launching/behaving correctly, Improve the way the test server is launched and killed

[DDvO]

This is a follow-up of #15580, picking up loose ends discussed there.

• 80-test_cmp_http.t: Simplify and prevent hangs on server not launching/behaving correctly.
  This should have help diagnosing tricky issues such as Test test_cmp_http failed with icc build #15557 and Test 80-test_cmp_http fails at alpha17 on NonStop x86/ia64 (round 2) #15571.
• 80-test_cmp_http.t: Improve the way the test server is launched and killed
  This should fix test/recipes/80-test_cmp_http.t: Simplify test_cmp_http() #15580 (comment).

• tests are added or updated

[richsalz]

Maybe have the server write it's process ID into a file in the test-runs directory.

I am surprised that this is so complicated, given that the TLSProxy etc do not have these problems.

[slontis]

This appears to now cause a few more infinite loops..

[DDvO]

Maybe have the server write it's process ID into a file in the test-runs directory.

That's a good thought, thank you - I now let the server simply emit its PID along with the ACCEPT string.

[DDvO]

This appears to now cause a few more infinite loops..

These are gone now :)

[levitte]

I am surprised that this is so complicated, given that the TLSProxy etc do not have these problems.

TLSProxy doesn't start the subprocess with an open of a pipe, but with a fork, and control is therefore different. That came with some pains as well, just different pains.

It's also possible that using perl's IPC modules directly would be better in the longer run. I don't know, I've never done that myself, but that might be something worth looking into at some point. (also gotta be careful that the modules used are perl's core modules since version n5.10)

[levitte]

I wonder whether you could add _RLD_LIB_PATH as:

local $ENV{_RLD_LIB_PATH} = $dir;

to account for how NonStop accounts for paths. I currently set this explicitly in the outer shell when testing.

We already have that in util/shlib_wrap.sh, which is being used by util/wrap.pl on Unix family platforms...

[DDvO]

I wonder whether you could add _RLD_LIB_PATH as:

local $ENV{_RLD_LIB_PATH} = $dir;

to account for how NonStop accounts for paths. I currently set this explicitly in the outer shell when testing.

We already have that in util/shlib_wrap.sh, which is being used by util/wrap.pl on Unix family platforms...

NIce that this is now done automatically by calling cmdstr().

[DDvO]

I think this PR is good state now.

[levitte]

Please rebase this on top of a fresh master. I would say that the approvals stand

[DDvO]

Please rebase this on top of a fresh master. I would say that the approvals stand

Done (though there is not even a real merge conflict with commit 8e2b2336fb).

[openssl-machine]

24 hours has passed since 'approval: done' was set, but as this PR has been updated in that time the label 'approval: ready to merge' is not being automatically set. Please review the updates and set the label manually.

[levitte]

Done (though there is not even a real merge conflict with commit 8e2b233).

For some reason, that commit hadn't made its way to github... now it has.

[DDvO]

I think I just referred to an outdated commit ID, which must have changed due to rebasing - sorry for the confusion.

[levitte]

I think I just referred to an outdated commit ID, which must have changed due to rebasing - sorry for the confusion.

No, you didn't. That commit hadn't travelled from OpenSSL's main repository (which isn't on github) to github (openssl/openssl is just a mirror). I had to push that manually.

[DDvO]

Merged - thanks