use /usr/bin/vi instead of /usr/share/dict/words as test input

[dgrove-oss]

…ilable to unblock CI work.

[shahmishal]

@dgrove-oss Can you please create a bug for this test, so we can enable the test in the future?

Thanks!

FAIL: dispatch_io
PASS: dispatch_io_net
PASS: dispatch_select
make[4]: Entering directory `/home/buildnode/disk2/workspace/oss-swift-incremental-RA-libdispatch-linux-ubuntu-14_04/Ninja-ReleaseAssert/libdispatch-linux-x86_64/tests'
make[4]: Nothing to be done for `all'.
make[4]: Leaving directory `/home/buildnode/disk2/workspace/oss-swift-incremental-RA-libdispatch-linux-ubuntu-14_04/Ninja-ReleaseAssert/libdispatch-linux-x86_64/tests'
============================================================================
Testsuite summary for libdispatch 1.3
============================================================================
# TOTAL: 26
# PASS:  25
# SKIP:  0
# XFAIL: 0
# FAIL:  1
# XPASS: 0
# ERROR: 0

[shahmishal]

@swift-ci Please test Linux

[MadCoder]

we just need "a large file" could you instead use a file that is here on any distribution that is large enough? /bin/systemctl is 500K /usr/bin/vi is probably also always here and in the MB range

disabling test is wrong

[dgrove-oss]

Sure. I can write the tests to look for different files instead. I've got the configure check working, so I will probably keep it and check for different input files. While we're playing with this, dispatch_io also wants another file (but skips the test without failing if it isn't there):

#if TARGET_OS_EMBEDDED
#define LARGE_FILE "/System/Library/Fonts/Cache/STHeiti-Light.ttc" // 29MB file
#define maxopenfiles 768
#else
#define LARGE_FILE "/System/Library/Speech/Voices/Alex.SpeechVoice/Contents/Resources/PCMWave" // 417MB file
#define maxopenfiles 4096
#endif

Suggestions for a replacement file here that will likely be on a distro?

[parkera]

Can't we just create a file in a temporary directory?

[dgrove-oss]

I'm taking Pierre's suggestion and using /usr/bin/vi as the input file instead. The configure check is probably overkill. Will push an update in a couple of minutes (testing it locally now).

[parkera]

@swift-ci please test linux

[shahmishal]

@swift-ci please test linux

[shahmishal]

Build timed out (after 166 minutes). Marking the build as failed.
Build was aborted

[seabaylea]

The fail was in the dispatch_pingpong test, so not related to this change.

Below is the output of the test log file:

==================================================
[TEST] Dispatch Ping Pong
[PID] 32399
==================================================


[BEGIN] dispatch_queue_create(ping)
    Actual: 0x1b50c40
    Expected: 0x1b50c40
[PASS] dispatch_queue_create(ping)

[BEGIN] dispatch_queue_create(pong)
    Actual: 0x1b50cf0
    Expected: 0x1b50cf0
[PASS] dispatch_queue_create(pong)

[BEGIN] dispatch_group_create
    Actual: 0x1b50d80
    Expected: 0x1b50d80
[PASS] dispatch_group_create

[BEGIN] Process exited
    Actual: 1
    Expected: 0
[FAIL] Process exited (bsdtestharness.c:132)
    bsdtestharness.c:132
[PERF]  wall time: 0.000100
[PERF]  user time: 0.068000
[PERF]  system time: 0.004000
[PERF]  max resident set size: 17580
[PERF]  page faults: 0
[PERF]  swaps: 0
[PERF]  voluntary context switches: 44
[PERF]  involuntary context switches: 10
FAIL dispatch_pingpong (exit status: 1)

[dgrove-oss]

is there something different in the build machines, the steps taken to build/test or the environment for testing pull requests against dispatch vs. the other CI testing that includes building & testing dispatch on Linux? A couple of the recent dispatch pull requests have failed by hanging while running dispatch_pingpong, but as far as I know we've never seen that hang happen in other CI configurations with dispatch and I can't reproduce it locally.

[seabaylea]

I've seen the dispatch_pingpong failure maybe twice in the wider libdispatch CI. The CI env is running with 48 CPUs, which is a little larger than I have locally - that might be part of the issue.

[parkera]

Any update here @seabaylea or @dgrove-oss? We're still blocked on merging URLSession because dispatch tests aren't passing.

[parkera]

I want to see if this fails again in the same way.

[parkera]

@swift-ci test linux

[seabaylea]

This one should be good - we've got a couple of other failures though. I'm currently looking at those.

[parkera]

I'm going to merge this. We will put up a different PR to disable the pingpong test, so as to make some forward progress here.