[ci] an attempt to run tests on Windows#6340
Conversation
|
currently at: 98.8722% passed Most of the (relevant) failures are Windows compatibility bugs, but I am fine with pending those for now: let me know what you think, should we instead run bigger suites (e.g. the MRI ones) and exclude Windows failures? |
|
@kares I am working on Windows a bit lately and would also like to see us running the specs and MRI tests. Specs first, since we know they have been "green enough" in the past. I will play with it a bit today to see how much we need to tag to get specs running to completion. |
|
I have opened #6357 for Windows RubySpec fixes. |
|
have put the Windows failing stuff (as pending) in commit-by-commit, still I got: |
|
was able to locate the tests failures/errors by changing test-unit to always print summary at the end, |
|
I am on Windows and could give those last few items a try here. For now I'd say just mark them pending and we can get to them as soon as we have time. |
|
tests should be green (all failures are pended) but we still get a 1 exit status. |
|
I'll give your branch a try locally and try to figure out the bad exit. Possibly related: what's up with all this: |
|
I notice that -Ptest is still green on Windows, so we could add that too. |
|
Here's the output I get at the end of The one failure can be ignored; my environment has both Java 8 and 11 installed and they are stepping on each other. I'll look into the rbconfig error. |
|
The rbconfig failure appears to be coming from this test: I'm not sure this should actually be expected to pass, since rbconfig.rb lives in the stdlib. We could consider moving it into the main jar, but as it is written I would not expect this test to pass normally. I'm looking into why it triggers this late error, and will also consider how to make it pass. |
|
Reproduces directly when running the contents of this test outside of the test harness, so it is at least a bad test: |
|
I am unsure why the error is delayed. Usually this means we're leaving In any case, I believe the following patch will resolve this, avoiding the attempt to load rbconfig without a JRUBY_HOME set: diff --git a/test/jruby/test_missing_jruby_home.rb b/test/jruby/test_missing_jruby_home.rb
index c820a895d1..f99aa9499c 100644
--- a/test/jruby/test_missing_jruby_home.rb
+++ b/test/jruby/test_missing_jruby_home.rb
@@ -9,7 +9,7 @@ class TestMissingJRubyHome < Test::Unit::TestCase
begin
runtime = org.jruby.Ruby.new_instance
assert_nothing_raised do
- runtime.eval_scriptlet('require "rbconfig"')
+ assert_equal([0, 1, 2], runtime.eval_scriptlet('3.times.to_a'))
end
ensure
java.lang.System.set_property('jruby.home', old) if oldI'll file a bug about rbconfig and some other libraries being stdlib-dependent when they may be safer to keep in the jar. |
kares
force-pushed
the
win-ci
branch
2 times, most recently
from
feb6c1f to
acd6515
Compare
|
fixed the remaining Windows compatibility issues and added |
|
I believe this can be merged any time and we can investigate the pending items and TODOs from there. I think some of your fixes here may help specs so I want to land this before I tag off specs. |
|
I'm going to go ahead ad land #6357 as is since it makes enough fixes to get specs running to completion. I did not add a Windows target since it makes more sense for me to wait until your PR lands. Then we can collab on getting a spec run and making it greener. |
2 participants
a bit of Windows CI de-ja-vu, this time trying to run
test:jrubychosen JRuby suite as it is very stable, and in out control, compared to say ruby-specs
motivated by: #6322 (can re-target to
jruby-9.2if needed)