Repository navigation
test/parallel/test-npm-install.js fails (Node 4.4.4) #6648
Description
- Version: 4.4.4
- Platform: Linux
- Subsystem: unknown
Running make testthe test test/parallel/test-npm-install.js fails.
=== release test-npm-install ===
Path: parallel/test-npm-install
assert.js:90
throw new assert.AssertionError({
^
AssertionError: npm install should run without an error
at ChildProcess.handleExit (/home/phw/devel/nodejs4/src/node-v4.4.4/test/parallel/test-npm-install.js:40:10)
at ChildProcess.<anonymous> (/home/phw/devel/nodejs4/src/node-v4.4.4/test/common.js:385:15)
at emitTwo (events.js:87:13)
at ChildProcess.emit (events.js:172:7)
at Process.ChildProcess._handle.onexit (internal/child_process.js:200:12)
Command: out/Release/node /home/phw/devel/nodejs4/src/node-v4.4.4/test/parallel/test-npm-install.js
Here is the npm-debug.log generated.
I see multiple issues with this test:
- Obviously npm is expecting a package.json in the build directory, which isn't there.
- This npm call uses the previously installed node 4.4.3 for testing and not the newly compiled version in my source tree.
- This test even runs if I configured the build with the
--without-npmflag (not sure whether this is indeed an issue or not).
Activity
scorpp commentedon May 9, 2016
Similar issue here with fresh ArchLinux. More over I tried to downgrade to 4.4.3 (which i had installed and working fine beforehand) and faced similar issue.
Even more to add - node binary segfaults on almost any operation. E.g. I cannot install npm.
I have another ArchLinux PC which doesn't have update installed for few days and 4.4.4 has been built on that PC without this issue. And it doesn't segfault. So I built it on 'older' PC and transferred and installed package on my laptop (which is suffering the topic issue). And 4.4.4 build on 'older' linux box works just fine on a 'newer' one.
Here is list of packages which are newer on suffering PC
MylesBorins commentedon May 9, 2016
This is an interesting edge case... it looks like two things are happening that should
I'm going to dig into this and see what is going on.
MylesBorins commentedon May 9, 2016
OK, so here is where I am at, and perhaps you can tell me a bit more about your systems because this is a bit odd.
This is being created and written into the
common.tmpdirhere. Thecwdof the spawned process is set tocommon.tmpdir, so it is very odd that is not being foundI have a PR submitted to make sure PATH is explicitly including the build directory which should fix this
Is that flag related to make install? AFAIK we don't touch or change anything in the npm directory, this test should be completely self contained to the repo, and that flag should have no affect on it's ability to pass
phw commentedon May 10, 2016
Thanks for looking into this. I have as a quick test now tried building both 4.4.3 and 4.4.4 again, both fail with the same error while previously
make testdid run without any error when I installed 4.4.3 some time ago. Also it is not only the npm install test but also a few other tests failing. So I think there is something wrong with my system (Arch Linux also, just like @scorpp ), I will investigate what changed since then and report back.phw commentedon May 10, 2016
I am clueless. Both version 4.4.3 and 4.4.4 fail on a couple of tests. The resulting binary is very unstable and core dumps in many cases. The last time I compiled and installed 4.4.3 was May 3, and that build went fine. But I don't see any critical update since then, and I even reverted some updates to try whether they had any effect. Just for having it here below is the output of
make test. But ultimately I think this is probably not really a node issue but something broken on the Arch side.addaleax commentedon May 10, 2016
test-tick-processor is a known flaky test (#4427), so you can probably ignore that.
bnoordhuis commentedon May 10, 2016
@pwh Are you compiling with g++ v6.x by any chance? If yes, see #6272 - it's going to be fixed in the next release but in the mean time try building with
make -j8 CXX="g++ -fno-delete-null-pointer-checks".scorpp commentedon May 10, 2016
I'm compiling with g++ v6.1.1, can confirm that adding
-fno-delete-null-pointer-checksfixes the issue. Both tests and segfaults. I'm on ArchLinux.Many thanks @bnoordhuis for pointing the right thread
MylesBorins commentedon May 10, 2016
@bnoordhuis I dug through #6272
Looks like the v8 changes are being backported in #6669
@scorpp if all is well this should be fixed in v4.4.5
pwh commentedon May 10, 2016
Hey, I'm pwh!
I'm not phw, and I know not node
so I'm unsubscribing the thread.
(Programmers sweat the details!)
MylesBorins commentedon May 10, 2016
off by one
phw commentedon May 10, 2016
The real @phw here ;) I still don't quite understand what happened that test-npm-install failed the way it did, but compiling with
-fno-delete-null-pointer-checksfixes the issue (and all other build errors). Thanks to all for helping out. Hope #6669 will go through for v4.4.5. At least this issue report led to the small patch by @thealphanerd to fix the node version used by the test :)Otherwise I think this issue can be closed if nobody objects.
test: ensure test-npm-install uses correct node
test: ensure test-npm-install uses correct node
test: ensure test-npm-install uses correct node
test: ensure test-npm-install uses correct node