You may have noticed that there are quite a few builds being reported as failing in koschei with SEGV errors during testing.
The bad news is that it seems that the builds of node done with gcc 6 seem to have some sort of problem that results in random crashes during garbage collection runs.
I can reproduce it in a rawhide VM that predates the rebuild just by updating the nodejs package to the nodejs-4.2.4-5.fc24 build from the mass rebuild. Going back to -4 fixes it.
I have tried updating node to 4.2.6 but that hasn't helped.
The crash, according to gdb, normally seems to be somewhere in v8 during a garbage collection run.
Tom
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 02/09/2016 03:53 AM, Tom Hughes wrote:
You may have noticed that there are quite a few builds being reported as failing in koschei with SEGV errors during testing.
The bad news is that it seems that the builds of node done with gcc 6 seem to have some sort of problem that results in random crashes during garbage collection runs.
I can reproduce it in a rawhide VM that predates the rebuild just by updating the nodejs package to the nodejs-4.2.4-5.fc24 build from the mass rebuild. Going back to -4 fixes it.
I have tried updating node to 4.2.6 but that hasn't helped.
The crash, according to gdb, normally seems to be somewhere in v8 during a garbage collection run.
Tom
I'm talking with the GCC people about it; can you point to an easy reproducer?
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 02/09/2016 10:45 AM, Stephen Gallagher wrote:
On 02/09/2016 03:53 AM, Tom Hughes wrote:
You may have noticed that there are quite a few builds being reported as failing in koschei with SEGV errors during testing.
The bad news is that it seems that the builds of node done with gcc 6 seem to have some sort of problem that results in random crashes during garbage collection runs.
I can reproduce it in a rawhide VM that predates the rebuild just by updating the nodejs package to the nodejs-4.2.4-5.fc24 build from the mass rebuild. Going back to -4 fixes it.
I have tried updating node to 4.2.6 but that hasn't helped.
The crash, according to gdb, normally seems to be somewhere in v8 during a garbage collection run.
Tom
I'm talking with the GCC people about it; can you point to an easy reproducer?
They also want us to try:
(10:44:14 AM) jakub_: sgallagh: that said, as usually I'd suggest them to try - -fsanitize=undefined, various packages e.g. call methods on NULL pointers which is invalid C++ (10:45:07 AM) jakub_: sgallagh: and/or try -fno-delete-null-pointer-checks , - -fno-strict-aliasing, -fno-aggressive-loop-optimizations , -fwrapv etc. if it makes a difference
Once I have a reproducer, I can try doing some builds with those flags and see what happens.
Please file a BZ to track this and add the information there.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 02/09/2016 10:46 AM, Stephen Gallagher wrote:
On 02/09/2016 10:45 AM, Stephen Gallagher wrote:
On 02/09/2016 03:53 AM, Tom Hughes wrote:
You may have noticed that there are quite a few builds being reported as failing in koschei with SEGV errors during testing.
The bad news is that it seems that the builds of node done with gcc 6 seem to have some sort of problem that results in random crashes during garbage collection runs.
I can reproduce it in a rawhide VM that predates the rebuild just by updating the nodejs package to the nodejs-4.2.4-5.fc24 build from the mass rebuild. Going back to -4 fixes it.
I have tried updating node to 4.2.6 but that hasn't helped.
The crash, according to gdb, normally seems to be somewhere in v8 during a garbage collection run.
Tom
I'm talking with the GCC people about it; can you point to an easy reproducer?
They also want us to try:
(10:44:14 AM) jakub_: sgallagh: that said, as usually I'd suggest them to try -fsanitize=undefined, various packages e.g. call methods on NULL pointers which is invalid C++ (10:45:07 AM) jakub_: sgallagh: and/or try -fno-delete-null-pointer-checks , -fno-strict-aliasing, -fno-aggressive-loop-optimizations , -fwrapv etc. if it makes a difference
Once I have a reproducer, I can try doing some builds with those flags and see what happens.
Please file a BZ to track this and add the information there.
Looks like this is probably due to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68853
The workaround is to add -fno-delete-null-pointer-checks to our compiler flags which will have an unspecified performance hit.
I'm going to build 4.2.7 this afternoon (There's a security release expected today) and I'll add that flag on it.
nodejs@lists.fedoraproject.org
-
Stephen Gallagher -
Tom Hughes